{ "schema_version": "newruntime-weeks-index-v0.1", "type": "week_in_review_index", "generated_at": "2026-10-28T22:00:01.749Z", "count": 18, "index_contract": "Field Notes grouped by the ISO week (Monday–Sunday) of their published date, then by topic hub. Follow the note routes under /posts/{slug}.json for content.", "routes": { "html": "https://newruntime.com/weeks/", "json": "https://newruntime.com/weeks.json" }, "weeks": [ { "key": "2026-w44", "label": "Oct 26 – Nov 1, 2026", "start": "2026-10-26", "end": "2026-11-01", "count": 299, "html": "https://newruntime.com/weeks/2026-w44/", "summary": null, "topics": [ { "slug": "claude", "label": "Claude", "count": 43, "notes": [ "artificial-analysis-launches-intelligence-index-v5-with-60-59ecd7", "blind-german-writing-test-kolibri-loses-all-136-e80a2a", "claude-code-2-1-312-adds-an-oidc-8c6cf2", "claude-sonnet-5-5-outage-hits-claude-cowork-8dd894", "deepseek-harness-ships-dsh-v0-2-2-alpha-ca4e22", "orcarouter-launches-orca-connect-letting-local-gpu-models-fa1c6b", "the-claude-code-plugin-is-open-source-as-f04b4e", "tradingagents-v0-7-1-fixes-look-ahead-leak-2a99b3", "virginia-vassilevska-williams-responds-to-backlash-over-the-a1d725", "ai2-has-released-old-ideas-novel-problems-42d2cf", "anthropic-has-expanded-its-agreement-with-the-us-1883af", "anthropic-reports-early-progress-using-claude-in-drug-e3a07b", "antibody-scientists-say-claude-s-scorpion-binders-are-dc73dd", "artificial-analysis-launched-code2video-aa-benchmarking-27-models-0bbef6", "claude-code-2-1-311-speeds-pro-max-9f4329", "claude-fable-5-1-saw-another-round-of-e9ebd5", "claude-for-financial-advisors-a6ef91", "cognition-groups-models-by-cost-and-capability-on-6ee91e", "cognition-launched-frontiercode-2-0-a-coding-benchmark-c6b49a", "delta-from-zed-makes-claude-code-sessions-multiplayer-39fadc", "delta-cli-sync-and-share-agent-sessions-from-38699e", "elevated-errors-hit-claude-mythos-5-1-fable-3b66a3", "glean-is-now-available-on-the-claude-marketplace-6930ed", "hallmark-is-a-rebuilt-design-skill-for-agents-a59090", "lukewarm-recommendations-for-improving-ai-agents-interactions-with-52fce9", "openbot-is-a-source-available-desktop-app-from-d564db", "openclaw-ships-v2026-10-7-patch-for-updates-2f5151", "the-spending-has-outrun-the-math-if-your-062a00", "tom-s-hardware-v4-1-flash-1017-is-b5e63d", "webframe-evaluated-as-a-scraper-and-a-sandbox-97924b", "a-10-year-old-cnn-outperforms-frontier-llms-7e50e0", "claude-cowork-and-chat-come-together-for-enterprise-b47caf", "cognition-launched-devin-security-2-0-combining-frontier-d42d7f", "expanding-claude-for-nonprofits-4d3f19", "last-february-epoch-senior-researcher-anson-ho-attempted-7d9df8", "lyra-c-released-neuron-explorer-a-steering-first-3f6fc9", "moonshot-s-kimi-k3-1-pro-flirts-with-73e739", "openclaw-released-v2026-10-7-146cda", "r-opensource-banned-anything-made-with-ai-projects-5bc090", "stale-instruction-files-override-the-code-in-coding-438cf9", "starting-today-staff-at-sweden-s-government-offices-de5e4c", "theo-browne-open-sourced-tstack-nine-agent-skills-04a9f8", "zed-1-25-2-stable-release-b4f5f7" ] }, { "slug": "google", "label": "Google", "count": 19, "notes": [ "building-conversational-commerce-with-multi-agent-swarms-on-ea0f9d", "cisa-adds-brotli-flaw-cve-2026-111497-node2shell-4aada4", "fast-reliable-a2ui-with-swappable-inference-formats-f5e911", "mercury-agent-1-4-2-rebuilds-keyless-web-9f6f64", "opera-ships-webai-early-preview-local-in-browser-a13927", "akamai-launches-akamai-inference-with-12-open-weight-be7d9b", "business-insider-reports-companies-selling-slack-crm-and-a8f5fa", "gemini-robotics-embodied-reasoning-2-enters-public-preview-e0fdff", "google-and-oracle-to-offer-google-tpus-on-23f0b0", "google-and-voltus-announce-one-gigawatt-byoc-agreement-e45876", "google-deepmind-unveiled-genie-4-a-real-time-24f142", "google-research-introduced-a-sleep-framework-letting-llms-7a9ea6", "google-research-reports-blind-sweep-ultrasound-matching-expert-7c756d", "google-s-litert-community-publishes-an-on-device-5dfaca", "google-s-self-improvements-in-llms-18-minute-26a79c", "preview-edit-and-collaborate-on-markdown-md-files-498bb5", "tenable-public-pocs-and-a-first-in-the-d9ad85", "google-developer-program-benefits-faq-ea21fb", "how-google-beam-connects-military-families-globally-4d6e80" ] }, { "slug": "agent-security", "label": "Agent security", "count": 18, "notes": [ "meta-launches-its-muse-personal-ai-agent-in-106056", "nvidia-openshell-v0-1-6-makes-gateway-authentication-243f82", "1password-for-agents-goes-live-for-meta-muse-5f8af2", "coworkers-build-your-own-ai-teammates-each-with-e58268", "emudhra-security-advisory-eight-intermediate-ca-keys-were-bb9034", "litellm-enterprise-proxy-flaw-let-unauthenticated-requests-leak-971578", "mattermost-introduces-sovereign-agentic-solutions-agents-on-customer-b44083", "nvidia-and-meta-expand-collaboration-on-meta-s-59a54e", "nvidia-says-european-and-nordic-cloud-partners-are-7b04ac", "openclaw-has-released-v2026-10-8-a-patch-4309eb", "vercel-observability-adds-custom-monitors-with-threshold-and-bb5dca", "a-configuration-mismatch-in-glm-5-3-on-8c8082", "cohere-releases-traces-and-a-reproduction-script-for-11db72", "cohere-s-sandbox-caught-qwen3-8-flash-next-e7776b", "gpt-6-1-sol-helps-coderabbit-strengthen-security-9718fe", "openai-trains-its-ai-on-a-fake-copy-57c8f2", "oracle-introduces-fusion-claw-a-governed-agentic-execution-925c3c", "texas-sued-openai-and-sam-altman-over-chatgpt-d96996" ] }, { "slug": "github", "label": "GitHub", "count": 17, "notes": [ "deepseek-harness-dsh-v0-2-1-rc-4-1b2ef5", "github-brings-its-fine-tuned-secret-detection-model-03c54b", "github-copilot-cli-1-0-105-lets-users-77ef51", "kapitsa-0-2-22-runs-flash-next-from-9160f3", "kapitsa-0-2-23-lands-an-ordering-fix-75ad97", "t3-code-nightly-lets-t3-chat-pro-and-40cf7e", "thinkrail-vibe-code-with-pi-in-a-lightweight-2e7924", "cloudflare-computer-now-helps-agents-recover-from-interruptions-5e7ee2", "github-blendi-remade-dioramas-a-free-open-source-cdd3b4", "introducing-pg-pinecone-open-source-search-for-your-0f87cb", "microsoft-maia-300-s-40-tokens-per-megawatt-5a1a47", "slotstream-runs-a-125b-moe-model-on-16-1dfd4c", "ub-mode-skill-md-at-master-unclebob-ub-6220de", "updates-to-github-secret-scanning-8b5197", "entire-led-by-ex-github-ceo-thomas-dohmke-e13ea8", "gpt-6-1-terra-is-generally-available-in-c81298", "reasoning-from-scratch-implement-a-reasoning-llm-in-2d2073" ] }, { "slug": "models", "label": "Models", "count": 14, "notes": [ "new-progress-index-v1-uses-item-response-theory-dcd58b", "per-iteration-cost-drops-to-o-nm-versus-a7c828", "built-on-rumodernbert-base-with-three-stage-training-9166f4", "deepvk-released-user2-base-a-149m-russian-sentence-70f5d3", "development-starts-in-november-with-a-small-model-a4f874", "fern-3-minute-read-64d096", "fleet-wide-computer-vision-lifecycle-at-the-edge-0e9dce", "full-method-detailed-in-the-arxiv-paper-from-7df950", "tencent-open-sourced-hy-simulmt-a-real-time-df6271", "tested-on-five-7b-mlrms-with-consistent-gains-fd8b39", "farbod-tavakkoli-released-otel-2-0-decision-26b-b85555", "gpt-6-1-terra-model-openai-api-865b61", "jack-dorsey-s-block-made-a-voice-ai-65ec90", "qwen-released-a-4-bit-awq-quantization-of-290649" ] }, { "slug": "harness-engineering", "label": "Harness engineering", "count": 13, "notes": [ "mai-cyber-1-1-flash-sharper-findings-and-01681c", "amp-lets-every-workspace-self-host-orbs-free-5ee832", "frontier-models-ability-to-control-robot-arms-is-193384", "groundhog-day-for-ai-agents-prompt-caching-patterns-3e941d", "how-we-built-our-design-agent-48f162", "inclusionai-releases-ling-3-1-flash-1019-an-9765f4", "independent-lab-run-of-flint-skipped-coding-scores-963341", "langchain-open-sources-design-agent-content-studio-its-4aab2f", "liquid-ai-launches-lfm3-an-open-weight-2b-d00d7d", "meta-s-ai-chief-apologizes-as-muse-wonder-6d4b7f", "bitter-lessoned-in-automated-theorem-proving-13-minute-2e2a5c", "cursor-released-the-agent-sdk-for-defining-bots-47b7e0", "netlify-adds-gpt-6-1-terra-to-its-019086" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 12, "notes": [ "anthropic-study-analysts-followed-a-hidden-ai-agent-e66db0", "copilotkit-released-openduet-an-mit-licensed-template-for-63ae8b", "a-dive-into-why-openai-anthropic-cut-llm-8966c7", "georgia-officials-say-anthropic-s-agent-used-a-0731f6", "harmonic-extends-aristotle-to-formal-verification-of-c-db045f", "ollama-0-40-9-expands-flint-from-macos-67a821", "oss-scanner-s-first-public-scorecard-114-projects-332e2c", "researchers-dispute-anthropic-s-claim-that-its-ai-d01fee", "security-researchers-say-anthropic-s-iot-worm-tests-16ffd3", "an-anthropic-researcher-had-claude-on-a-statistics-c87ffb", "anthropic-publicly-filed-its-s-1-to-list-14d9cd", "anthropic-publicly-filed-its-s-1-with-the-acb9e6" ] }, { "slug": "mcp", "label": "MCP", "count": 11, "notes": [ "letta-code-v0-34-26-adds-cursor-coding-a35cf4", "nvidia-released-auto-ontology-an-apache-2-0-90349b", "pydantic-ai-v2-62-0-adds-mid-conversation-92bdb6", "staygreen-uptime-monitor-for-agent-built-apps-goes-e3964b", "upstash-ships-mcp-tasks-support-in-its-mcp-6565fb", "canva-mcp-is-now-generally-available-to-developers-b92553", "parallel-decisions-api-web-grounded-classification-dc4d7c", "ponte-open-source-end-to-end-encrypted-channel-eb44a0", "served-over-mcp-parses-documents-with-sarvam-vision-3537ad", "vercel-sandbox-commands-now-accept-stdin-enabling-stdio-d16cc1", "void-calendar-0-1-0-beta-7-for-e0748a" ] }, { "slug": "api-design", "label": "API design", "count": 10, "notes": [ "introducing-authenticated-fetch-browserbase-cb64c4", "bell-the-personal-assistant-stack-91721b", "elevenlabs-on-ai-gateway-b33c16", "much-requested-and-now-yours-to-connect-icloud-8fa82d", "oracle-fusion-companion-allows-enterprise-workers-utilize-different-13d75e", "ships-a-kicad-fork-with-an-ipc-schematic-ceec78", "siemens-partners-with-nvidia-to-advance-ai-driven-18e469", "perplexity-has-released-decider-for-mac-19b242", "pitchbook-in-north-cohere-826460", "vercel-cli-adds-vercel-connect-discover-to-search-c421c4" ] }, { "slug": "codex", "label": "Codex", "count": 10, "notes": [ "openai-reports-another-codex-chatgpt-and-api-degradation-6124a2", "metr-and-redwood-research-say-openai-agent-monitoring-a8b5ad", "openai-expands-partnership-with-oracle-to-bring-its-272a5a", "vercel-ai-sdk-harnessagent-now-accepts-image-inputs-2ddc63", "chatgpt-s-expanding-world-of-work-16036a", "codex-helped-build-this-token-usage-dashboard-for-8dc689", "gpt-6-1-terra-is-live-on-ai-b29cde", "infini-ai-lab-open-sources-hermit-a-framework-8dfe5a", "octopus-si-os-is-an-open-source-agentic-1c0cd4", "openai-releases-gpt-6-1-terra-a-cheaper-d843fa" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 9, "notes": [ "amazon-formally-withdraws-its-data-center-application-for-af11b5", "announcing-soci-indexed-images-for-faster-sagemaker-ai-f9f202", "gartner-predicts-over-80-of-neocloud-providers-will-a74fdb", "bdn-sealed-volumes-publish-once-mount-anywhere-a3d065", "elisa-completes-europe-s-first-live-network-ai-ac73b0", "in-china-the-ai-bubble-has-already-burst-2bf514", "meta-launches-35-billion-underwritten-stock-offerings-as-b5ce36", "mws-ai-has-launched-cowork-8d8134", "together-ai-details-auto-repair-for-together-gpu-a330ef" ] }, { "slug": "generative-media", "label": "Generative media", "count": 9, "notes": [ "sglang-merges-hy-nomas-support-for-offline-and-372c96", "spaitial-launches-echo-3-world-model-and-iphone-dc6701", "artlist-lists-ideogram-vector-1-0-as-a-3981df", "court-allows-umg-and-sony-to-depose-ex-6bff33", "getty-images-files-8-k-on-143-million-91355a", "meta-is-testing-the-live-avatar-feature-for-48a6de", "microsoft-s-mai-image-2-7-preview-ranks-ccadfd", "quantstack-released-gguf-quantizations-of-wan2-2-i2v-9462a3", "runway-launched-app-studio-a-no-code-builder-0ef8e8" ] }, { "slug": "evals", "label": "Agent evals", "count": 8, "notes": [ "independent-self-hosted-test-of-astabrief-8b-finds-92ecbc", "mercury-agent-1-4-1-adds-learned-work-39078b", "alongside-the-fork-the-project-ships-deterministic-viewers-50cfd0", "continual-learning-exam-benchmark-finds-best-llm-agent-44b039", "how-ai-is-changing-the-pace-of-discovery-8fdbba", "on-sierra-s-tau-knowledge-matched-baseline-accuracy-e2114f", "scale-ai-s-agentenv-agent-evals-can-now-f2ea3e", "working-with-academic-communities-to-build-scientific-understanding-bdc335" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 8, "notes": [ "openbmb-open-sources-minicpm-ocr-1-0-and-457493", "reaches-42-8-top-1-and-68-top-09386e", "superplane-v1-0-0-first-stable-release-of-75d162", "western-australia-wants-a-refund-from-deloitte-over-5a4db4", "with-your-approval-1password-for-agents-can-sign-60ab5e", "humanlayer-the-best-way-to-get-ai-coding-5e49d4", "maintaining-software-with-the-gpt-6-family-2a875d", "oracle-announces-deep-trust-ai-architecture-to-enforce-92cc0e" ] }, { "slug": "security", "label": "Security", "count": 8, "notes": [ "autonomous-ai-agent-cipher-finds-four-epics-base-15c4dc", "git-for-windows-2-56-0-4-ships-b48890", "openai-s-daybreak-lead-says-ai-generated-bug-83da82", "cisa-adds-php2shell-flaws-to-the-known-exploited-9f2c05", "connectwise-misses-october-26-cloud-backup-restore-onboarding-5ba284", "introducing-the-cloudflare-php-capture-the-flag-challenge-529e4d", "ox-security-93-6-of-measurable-php-hosts-e44b08", "php2shell-confirmed-as-working-in-the-wild-rce-1c9b9a" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 4, "notes": [ "11-840-tps-on-a-single-gpu-d7e41f", "cursor-says-x-subscription-switches-stay-optional-and-16c11c", "vercel-agent-sessions-now-cost-67-less-on-3cfbee", "typesafe-s-jev-has-skyrocketed-in-recent-months-c410b5" ] }, { "slug": "compute", "label": "Compute", "count": 4, "notes": [ "meta-networking-at-the-2026-ocp-global-summit-2c00fc", "nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f", "elon-musk-said-on-x-over-the-weekend-e6435c", "spacex-closes-40-billion-debt-deal-for-ai-52d0f4" ] }, { "slug": "generative-ui", "label": "Generative UI", "count": 4, "notes": [ "langfuse-has-launched-halo-an-assistant-inside-its-ee4d1e", "on-tuesday-brilliant-labs-officially-launched-its-much-d3cab3", "vllm-semantic-router-releases-d3-ultra-a-26b-37f577", "bluejay-adds-typesafe-s-jev-decision-model-as-ad40c8" ] }, { "slug": "open-models", "label": "Open models", "count": 4, "notes": [ "apache-2-0-licensed-onnx-available-with-a-404ae2", "baidu-open-sources-unlimited-ocr-2-a-3b-b8360f", "new-hugging-face-merge-minimax-h3-singularity-x-17c3f3", "tencent-hunyuan-open-sources-hy-simulmt-a-30b-fd28ee" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 3, "notes": [ "box-s-deep-agent-architecture-for-box-agent-657eb5", "memory-tax-paper-finds-many-dedicated-agent-memory-f40f56", "shopify-moves-its-ai-agent-knowledge-base-into-031d7f" ] }, { "slug": "skills", "label": "Agent skills", "count": 3, "notes": [ "amazon-mwaa-adds-four-ai-agent-skills-to-93d318", "how-some-malicious-ai-skills-hide-instructions-from-7a9f9b", "moonshot-stripped-hacking-skills-out-of-its-new-851b3b" ] }, { "slug": "ai", "label": "AI", "count": 3, "notes": [ "biotech-startup-pathos-ai-raises-237-million-3-48c324", "colossal-biosciences-unveils-genomorph-an-ai-tool-that-9b7b50", "warp-sciences-emerges-from-stealth-with-215-million-503cae" ] }, { "slug": "ai-security", "label": "AI security", "count": 3, "notes": [ "oracle-launches-ai-security-mesh-a-managed-vulnerability-1232c1", "openai-faces-us-wiretapping-suit-over-chatgpt-ads-40af50", "nask-says-its-openai-ai-safety-agreement-is-98989f" ] }, { "slug": "computer-use", "label": "Computer use", "count": 3, "notes": [ "inclusionai-releases-lbs-27b-a-qwen3-8-27b-214d7f", "openai-has-introduced-gpt-6-1-terra-a-b073a1", "openai-has-introduced-gpt-6-1-terra-a-43aa54" ] }, { "slug": "engineering-management", "label": "Engineering management", "count": 3, "notes": [ "your-next-sales-teammate-should-be-a-grok-69efa3", "how-check-point-accelerates-vulnerability-research-with-daybreak-64c29a", "you-know-the-drill-someone-asks-for-a-eaadbc" ] }, { "slug": "inference", "label": "Inference", "count": 3, "notes": [ "humming-block-fp8-path-fails-deepseek-v4-1-ccbde0", "infinigence-ai-and-three-body-universe-launch-sophon-cc8c5e", "vllm-blog-serving-deepseek-v4-1-flash-at-073f14" ] }, { "slug": "local-models", "label": "Local models", "count": 3, "notes": [ "apple-introduces-the-new-14-inch-macbook-pro-fff5b1", "community-mlx-quant-fits-nanbeige4-3-33b-into-c49936", "release-desktop-v0-0-49-cline-cline-b90716" ] }, { "slug": "agent-loops", "label": "Agent loops", "count": 2, "notes": [ "devin-desktop-3-11-228-fixes-remote-ssh-12526a", "knowing-how-is-not-knowing-what-14-minute-edc126" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 2, "notes": [ "parallel-has-spent-about-a-year-building-web-5860d5", "rea-agents-6-8-0-adds-local-agent-f153cc" ] }, { "slug": "ai-products", "label": "ai-products", "count": 2, "notes": [ "cursor-launches-agent-sdk-for-building-and-deploying-custom-89cfee", "moonshot-unveils-kimi-k3-1-pro-open-weight-150efb" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 2, "notes": [ "ver-di-ties-amazon-s-16-billion-germany-ade59a", "microsoft-offers-voluntary-buyouts-as-ai-leads-to-1d29e7" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 2, "notes": [ "unlimited-ocr-2-builds-on-deepseek-ocr-deepseek-8f6931", "unsloth-ships-dynamic-3-0-gguf-quants-of-1e5a1f" ] }, { "slug": "observability", "label": "Observability", "count": 2, "notes": [ "vercel-agent-can-now-validate-production-alerts-before-51e95a", "vercel-observability-adds-ai-traffic-insights-showing-how-62373c" ] }, { "slug": "open-source", "label": "Open source", "count": 2, "notes": [ "assisted-by-ai-tags-reached-14-5-of-a4792b", "murces-minecraft-server-tui-for-linux-servers-meant-3369b3" ] }, { "slug": "tools", "label": "Tools", "count": 2, "notes": [ "our-records-know-when-they-re-played-135a98", "simon-willison-has-released-seam-cutter-a-small-14180f" ] }, { "slug": "agent-identity", "label": "Agent identity", "count": 1, "notes": [ "at-the-gartner-it-symposium-xpo-this-week-e90f5f" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 1, "notes": [ "privacylib-privacy-controls-at-the-data-access-boundary-2d11c3" ] }, { "slug": "agent-protocols", "label": "Agent protocols", "count": 1, "notes": [ "deprecating-the-graphql-api-timeline-and-migration-paths-f8bc7c" ] }, { "slug": "agents", "label": "Agents", "count": 1, "notes": [ "daytona-raised-60m-at-a-1b-valuation-to-00dcde" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 1, "notes": [ "meta-launches-35-billion-underwritten-offerings-and-a-5363fd" ] }, { "slug": "automation", "label": "Automation", "count": 1, "notes": [ "bedrock-robotics-raises-600-million-4-1-billion-00cc91" ] }, { "slug": "code-review", "label": "Code review", "count": 1, "notes": [ "greptile-takes-bug-detection-beyond-code-review-with-7582d7" ] }, { "slug": "design-systems", "label": "Design systems", "count": 1, "notes": [ "cursor-design-mode-adds-annotate-for-visual-prompts-f89d89" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 1, "notes": [ "your-plain-dockerfile-now-deploys-on-vercel-fdccd6" ] }, { "slug": "enterprise-agents", "label": "Enterprise agents", "count": 1, "notes": [ "sarvam-s-knowledge-fabric-compiles-enterprise-docs-into-4ffcbd" ] }, { "slug": "gemini", "label": "Gemini", "count": 1, "notes": [ "google-releases-the-gemini-api-cli-a-zero-4a8b64" ] }, { "slug": "governance", "label": "Governance", "count": 1, "notes": [ "us-treasury-proposes-crack-down-on-tech-investments-494c6b" ] }, { "slug": "knowledge-work", "label": "Knowledge work", "count": 1, "notes": [ "new-benchmarks-cognition-s-frontiercode-2-0-merge-a78e19" ] }, { "slug": "model-routing", "label": "Model routing", "count": 1, "notes": [ "cloudflare-agents-sdk-0-29-0-updates-the-5e6b75" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 1, "notes": [ "freight-search-composing-a-shared-search-index-from-17718c" ] }, { "slug": "tool-use", "label": "Tool use", "count": 1, "notes": [ "diffuspace-releases-clover-35b-a3b-preview-an-open-bd5943" ] }, { "slug": "other", "label": "Other notes", "count": 18, "notes": [ "31-states-and-dc-object-to-character-ai-1f5875", "broadcom-vmsa-2026-0008-vmware-sddc-manager-hit-534cf6", "bun-1-4-4-patches-bundled-brotli-decoder-d5a54b", "character-ai-files-revised-bidding-procedures-order-after-d144a2", "delft-honeypot-data-shows-72-000-php2shell-exploitation-daa61f", "diffuspace-releases-clover-35b-a3b-preview-an-open-463b05", "federal-grand-jury-indicts-ex-openai-engineer-abdulrahman-42c1e6", "grok-imagine-releases-a-7-minute-odyssey-episode-85aab1", "hy-nomas-technical-report-published-on-arxiv-2610-9d073d", "load-full-manhattan-building-footprints-from-compressed-data-53e76f", "loudoun-supervisors-voice-doubts-about-amazon-deal-fast-8318b8", "lovett-and-karingula-post-a-simplified-proof-of-a20347", "musk-puts-first-external-optimus-deliveries-around-end-f82a5d", "opencode-data-shows-kimi-k3-1-pro-at-911d98", "opencode-makes-deepseek-v4-1-flash-free-on-635004", "opencode-raises-deepseek-v4-1-flash-go-allowance-f0044e", "reylo-0-6-0-adds-slonik-support-try-ce5302", "ruby-releases-4-0-9-3-4-12-543af9" ] } ] }, { "key": "2026-w43", "label": "Oct 19–25, 2026", "start": "2026-10-19", "end": "2026-10-25", "count": 690, "html": "https://newruntime.com/weeks/2026-w43/", "summary": { "label": "New Runtime synthesis", "headline": "Agents are moving from demos into durable work, but control and verification are becoming the price of autonomy", "paragraphs": [ "This week’s shift is from agents that answer or code to agents that persist, use tools and operate in managed environments. Scheduled Codex automations, Notion’s 24-hour computer, durable MCP jobs and sandboxed runtimes make delegated work easier to keep running; open models and inference work widen the set of teams that can afford it. This continues the prior weeks’ move toward persistent agents, but shifts the hard question from whether a model can act to whether its work can be bounded and resumed safely.", "The gains in access come with sharper contradictions. Open-weight models are cheaper to run and increasingly available in production workflows, yet notes on unreliable small models and open models trailing in cybersecurity caution against treating low cost as equivalent capability. Meanwhile, reports of agents disabling monitoring, spending unintended funds or probing customer accounts sit beside new sandboxes, identity controls and security scans. Builders gain flexibility; buyers inherit more responsibility for permissions, audit trails and recovery.", "The practical change is to evaluate the whole operating path, not just model scores: test bounded tasks, tool permissions, sandbox escape and what happens when an agent claims completion. As decision APIs and durable runtimes spread, buyers should require evidence of control and verification before expanding agent access, and builders should make those controls part of the harness rather than bolt-ons." ], "key_stories": [ { "slug": "openai-agents-api-adds-scheduled-automations-with-cron-cf145e", "why": "Scheduled runs make agent work persistent, increasing the importance of explicit scope and recovery controls." }, { "slug": "aws-and-uc-berkeley-open-source-infinity-a-5436d7", "why": "A runtime designed to suspend agents at tool calls shows infrastructure adapting to large-scale, long-lived execution." }, { "slug": "perplexity-open-sourced-the-sandbox-that-keeps-its-10afd0", "why": "Sandbox testing found network-rule gaps despite failed escape attempts, illustrating that containment needs ongoing validation." }, { "slug": "openai-s-own-agents-turned-off-security-monitoring-4d1aee", "why": "Agents weakening their own monitoring makes safety controls a runtime concern, not just a model behavior issue." }, { "slug": "us-banks-move-to-restrict-third-party-ai-0bad17", "why": "Banks restricting third-party agents show that delegated access can collide with account security and user attribution." }, { "slug": "minimax-releases-open-weights-for-minimax-m3-1-4125a5", "why": "An open-weight model aimed at agents and coding makes capable execution more accessible, while leaving deployment risk with operators." }, { "slug": "anthropic-launched-oss-scanner-free-claude-powered-vulnerability-de807d", "why": "Continuous vulnerability scanning and fix tracking show security tools shifting from one-off checks to ongoing agent-era maintenance." } ], "source_note_count": 690, "generated_at": "2026-10-26T12:50:23.482Z" }, "topics": [ { "slug": "claude", "label": "Claude", "count": 71, "notes": [ "ex-ussr-kinship-research-evidence-first-genealogy-skill-29b9cf", "stepfun-has-published-api-documentation-for-step-5-e1edb8", "a-habr-author-has-described-claude-limits-7785aa", "claude-api-claude-ai-and-claude-code-saw-00966e", "claude-code-v2-1-310-brings-the-monitor-24f7bd", "claude-cowork-and-chat-come-together-for-enterprise-df71e0", "deepseek-releases-updated-v4-1-flash-model-it-4cb23a", "feat-mods-ui-stack-viewer-map-every-component-545e53", "frontiermemory-frontiermemory-benchmarking-continual-learning-through-memory-ca7ffa", "harvey-is-an-open-source-autonomous-sales-agent-1d49a4", "humanlayer-open-sources-its-coding-agent-control-plane-124fca", "open-weight-models-trail-on-cybersecurity-eeeb2c", "simon-willison-published-quick-study-b7aa90", "syntropylabs-launches-publicly-as-an-ai-observability-evaluation-ec1bfd", "zed-1-25-2-stable-cuts-binary-size-6413ee", "a-new-form-of-eyewear-ee9c95", "allenai-astabrief-8b-hugging-face-af4afe", "anthropic-and-theorem-used-claude-to-formally-verify-255192", "anthropic-and-theorem-claude-wrote-machine-checked-no-9a7d2c", "anthropic-s-claude-code-changelog-records-a-set-1c9505", "anthropic-s-claude-platform-documentation-now-lists-claude-291167", "artificial-analysis-has-released-version-2-0-of-8a8c4d", "claude-code-2-1-283-wraps-up-cleanly-163a7c", "claude-developer-platform-release-notes-d9ca32", "claude-for-government-changelog-bda6dd", "claude-haiku-5-5-docs-1m-context-128k-1c1b3a", "claude-launches-claude-startup-stack-with-up-to-66a59c", "critpt-v2-launches-with-70-revised-research-level-d7f03a", "dynamic-workflows-for-stateful-agents-and-decision-models-acdbcd", "firecrawl-launches-free-government-legal-index-for-chatgpt-41d295", "harshpreet-hand-v1-0-0-a-handwriting-font-2d7124", "humanlayer-open-sources-its-core-coding-agent-platform-54b0aa", "introducing-1password-for-agents-7-minute-read-a76f79", "kwant-launches-as-an-open-source-self-hosted-8521c0", "memory-as-middleware-for-self-improving-ai-agents-abd8b2", "obra-superpowers-v7-0-1-retunes-brainstorming-skill-b75957", "obra-superpowers-v7-0-1-retunes-the-rebuilt-227ea3", "release-notes-delta-docs-6253bf", "the-agent-said-it-was-done-the-database-3a02f5", "the-full-sky-in-the-ultraviolet-bc1a68", "the-team-released-a-proof-archive-the-machine-58d819", "use-claude-code-with-your-pro-or-max-c83685", "use-claude-cowork-on-web-desktop-and-mobile-c6be57", "world-of-oldcraft-the-windows-build-the-full-ac9a28", "5-production-challenges-in-multi-agent-systems-and-48676c", "announcing-new-ways-to-customize-models-with-sagemaker-030e9c", "claude-mods-flipping-the-harness-inside-out-11-33c280", "claude-platform-is-now-available-on-google-cloud-73c65b", "claude-security-can-now-watch-your-code-continuously-d038f0", "claude-security-gets-continuous-scans-and-fix-tracking-188085", "crowdstrike-uncovers-new-agentic-ai-campaigns-621213", "get-started-with-claude-duet-b54952", "how-does-fable-5-1-hide-messages-in-5226a9", "introducing-the-free-government-legal-index-for-chatgpt-b6ff71", "meta-s-muse-hit-10-million-downloads-in-38d0de", "release-v1-25-0-pre-zed-industries-zed-9e6201", "sparqlite-sparql-1-1-inside-sqlite-written-in-cb516c", "why-now-microsoft-ai-has-its-own-stance-633383", "ai-agents-codex-and-claude-performed-a-full-0349db", "building-effective-agent-automations-claude-dev-blog-55ed08", "claude-duet-is-an-early-research-preview-available-ac51e7", "claude-security-now-scans-continuously-and-tracks-fixes-3fd743", "cua-lets-your-agents-run-jobs-across-every-5b86dc", "introducing-claude-duet-claude-by-anthropic-05310c", "a-smarter-model-won-t-save-your-bad-36f226", "claude-code-paste-preview-see-and-mark-up-c2df45", "one-hacker-used-ai-agents-to-break-into-79fa67", "software-engineering-agents-such-as-agentless-swe-agent-e4c26b", "structured-outputs-for-anthropic-claude-models-on-amazon-c9a274", "a-coffee-chain-s-vibe-coded-dashboard-leaked-e75f1e", "mods-claude-code-mods-by-dave-kiss-analysis-e1be5d" ] }, { "slug": "models", "label": "Models", "count": 62, "notes": [ "inclusionai-open-sourced-five-ming-3-0-speech-0bcc34", "nemotron-3-5-lightning-is-a-30b-hybrid-5ce211", "block-unveils-dessa-a-full-duplex-speech-model-62567c", "cohere-has-documented-cohere-transcribe-arabic-668717", "hugging-face-transformers-merges-add-kolibri-1-pr-d3aab5", "kimi-android-app-v3-1-5-ships-with-1bf2cb", "knowledge-distillation-lets-smaller-language-models-inherit-capabilities-d1d19a", "mclr-fine-tunes-diffusion-models-to-maximize-inter-1d808b", "moonshot-ai-has-added-kimi-k3-1-pro-7f366e", "moonshot-released-kimi-k3-1-pro-a-2-c3d4fa", "moonshot-says-kimi-k3-1-pro-was-trained-d50b53", "new-nvidias-cosmos-3-tools-simulation-specific-models-3af45b", "paper-inherited-not-invented-finds-the-wait-marker-bc3007", "paper-4-bit-n-gram-embedding-tables-cut-1822b8", "reflection-publishes-beam-customization-recipes-self-distillation-distillation-5a2a4b", "technology-innovation-institute-tii-introduced-falcon-asr-da7399", "beam-model-by-reflection-is-now-available-on-62d87f", "chatgpt-voice-gpt-live-1-1-brings-new-a7dd16", "designing-kolibri-architecture-trade-offs-from-first-principles-c04638", "does-your-model-run-on-aws-trainium-look-014c50", "mistral-released-timestral-its-first-time-series-foundation-e44ba8", "moonshot-launches-kimi-k3-1-pro-on-the-d876cd", "ocria-1-document-parsing-model-from-llamaindex-is-74e755", "openai-says-80-to-90-percent-of-its-54dd1f", "qd-rec-open-ended-evolution-of-recommendations-via-89ff95", "the-strict-threshold-for-gaussian-ellipsoid-fitting-research-699248", "weighted-conditional-flow-matching-loss-is-proven-to-980e80", "ensemble-forecasting-with-ibm-granite-time-series-models-18fae7", "get-started-with-llm-training-on-modal-c92ce5", "hugging-face-biology-research-has-published-carbon-x-43b3e8", "jolt-extends-opsd-or-on-policy-self-distillation-7a545f", "llamaindex-has-introduced-ocria-1-a-4-billion-652dae", "perplexity-and-nvidia-released-jolt-a-joint-teacher-0cb0aa", "recursive-released-flint-a-40b-total-2-5b-37dd1a", "train-muse-glimmer-for-free-our-recipe-for-4f4f83", "train-qwen3-on-aws-trainium-with-torchtitan-75a96f", "write-your-own-training-loop-without-managing-gpus-12e654", "contrastive-lm-has-a-public-hugging-face-organization-fa03e0", "deepseek-v4-1-flash-1017-is-now-included-323f0e", "granite-speech-5-0-transcribe-translate-and-attribute-fa8fd7", "humans-released-ampersand-v0-1-a-collaborator-model-2e438d", "rep2text-reconstructs-text-from-a-single-last-token-a808b9", "sakana-ai-released-qd-rec-a-quality-diversity-d37692", "train-in-simulation-deploy-on-a-real-robot-9d85ca", "zero-shot-to-new-robots-introducing-dream-1-b1810d", "211-designs-were-lab-tested-at-adaptyv-bio-1b76fb", "26-best-text-to-speech-models-for-voice-2b8c30", "amazon-sagemaker-jumpstart-2026-year-to-date-in-edbb27", "artificial-analysis-has-launched-aa-personalitybench-a-public-01433e", "cheap-but-unreliable-smaller-ai-models-are-tempting-057529", "glas-reports-fine-tuning-gigaam-v3-for-russian-a418b7", "gradium-open-sourced-phonon-a-family-of-31m-b6b3d6", "introducing-sky-001-the-first-world-model-for-13eabc", "prime-intellect-has-released-prime-rl-v0-10-2e34fd", "rethinking-confidence-in-reasoning-models-analysis-f362d0", "together-post-training-frontier-grade-model-shaping-the-bff980", "truth-anchoring-tac-maps-raw-uncertainty-scores-to-ef386b", "voxtral-3-voice-agent-sdk-speech-models-in-44fa6b", "introducing-clairic-a-new-system-one-class-of-a47edc", "mercury-3-from-inception-ai-is-live-on-c7ea9a", "minimax-m3-1-flash-goes-toe-to-toe-b64392", "torchneuron-running-hf-models-on-aws-trainium-is-787908" ] }, { "slug": "google", "label": "Google", "count": 40, "notes": [ "a-habr-author-described-a-research-prototype-for-8261b9", "announcing-llm-d-1-0-open-distributed-inference-7539eb", "four-evaluation-engineering-lessons-learned-from-building-agent-302c05", "google-adk-for-python-v2-13-0-adds-13d2bb", "inside-agent-compute-engine-building-the-compute-foundation-cb5231", "openai-s-patch-the-planet-discloses-brotli-decoder-ad9480", "ai-ready-biological-data-1-8-billion-global-e2cf6d", "cloud-billing-release-notes-b0c915", "gemini-4-argon-lands-for-google-antigravity-for-cf22e8", "google-cloud-launches-agent-compute-engine-for-sandboxed-229085", "google-cloud-offers-day-0-support-for-reflection-82afff", "google-deepmind-paper-four-gemini-4-argon-agents-ddbaec", "google-deepmind-researchers-resigned-over-the-company-s-ce105f", "openai-s-red-team-agents-found-an-unauthenticated-7fb79e", "relevance-isn-t-enough-what-it-takes-to-b2317f", "understanding-auto-in-the-gemini-model-picker-a323f8", "according-to-meta-s-latest-earnings-report-the-3b1652", "death-to-long-lived-static-credentials-ce34e4", "not-all-forests-are-equal-mapping-the-world-a7e56a", "openai-and-google-cloud-said-that-openai-s-efd7e1", "openai-models-now-available-across-all-three-major-cf50f7", "streaming-comes-to-api-gateway-real-time-and-ab194f", "uber-brings-chatgpt-work-to-all-employees-with-85121f", "with-your-permission-gemini-can-now-reference-your-abe8d1", "google-and-meta-bring-native-pytorch-support-to-3b9b77", "google-released-embeddinggemma-2-a-740m-open-weight-8a4f42", "openai-models-are-coming-to-google-cloud-8edcd4", "powering-the-agent-development-lifecycle-with-any-framework-414ca1", "announcing-new-vllm-and-sglang-inference-engines-on-be000f", "free-my-tv-a-fast-ad-free-home-cdd033", "google-launched-agentic-video-understanding-for-gemini-flash-ad81da", "how-ai-can-help-doctors-spend-more-time-41842f", "hugging-face-announced-torchtpu-a-native-pytorch-backend-7d05d9", "integrating-torchtpu-into-torchtitan-and-optimizing-models-for-6f942a", "unifying-edge-ai-how-coral-npu-s-open-d7ec2f", "cleanyourtv-debloat-any-android-or-google-tv-over-85ed5a", "firecrawl-for-google-cloud-analysis-0dc927", "native-model-authoring-for-edge-with-litert-tensor-435140", "step-by-step-guide-build-a-voice-agent-259b68", "tencent-says-its-ai-agents-went-off-script-48251d" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 35, "notes": [ "a-habr-article-by-sopranodev-sets-out-a-74e414", "myna-v0-2-0-open-source-bring-your-77783c", "anthropic-has-published-single-agent-loop-kit-4e80d5", "claude-duet-safety-privacy-and-security-b02eb4", "isec-news-opencode-says-breach-tied-to-distillation-e57018", "kimi-k3-1-pro-debuts-at-8-on-bd08e0", "uk-launches-ai-biosecurity-lab-with-100m-public-cdc041", "agentic-credit-analysis-ramp-builders-651933", "anthropic-commits-5-of-its-shares-to-a-d72e9d", "anthropic-frontier-red-team-shows-claude-can-formally-aa997f", "anthropic-has-published-formal-math-a-public-github-589726", "anthropic-publishes-alignment-assessment-of-claude-mythos-5-987d90", "anthropic-publishes-first-annual-public-benefit-report-assessing-a403c5", "anthropic-sonnet-5-5-launch-price-features-and-f086bd", "claude-models-on-google-cloud-anthropic-managed-is-ecc3eb", "deepseek-raises-100-billion-yuan-to-compete-with-44ae5d", "public-benefit-report-anthropic-f7766e", "review-of-the-misalignment-in-high-stakes-settings-34ff43", "1password-announced-its-agents-platform-in-collaboration-with-c6e982", "anthropic-released-api-version-2026-10-21-its-c6768c", "building-agents-when-to-fold-subagents-into-your-a9d2d9", "donating-agent-skills-to-the-agentic-ai-foundation-4b43e5", "anthropic-s-draft-safety-bill-would-let-a-da657d", "musk-says-spacex-and-tesla-are-exploring-a-04d877", "physicist-christopher-stubbs-joins-anthropic-s-long-term-c3f915", "anthropic-turns-to-lawyers-to-help-ai-agents-9d48fc", "how-to-review-your-agents-for-security-issues-63fda7", "openai-is-reportedly-racing-to-raise-30-billion-dc430d", "redesigning-source-control-at-anthropic-anthropic-analysis-dd7491", "strengthening-our-defenses-against-distillation-attacks-anthropic-analysis-533a67", "anthropic-announces-a-public-fund-as-agents-drive-d495a3", "anthropic-is-reportedly-offering-blackstone-over-10-to-fe8362", "anthropic-launched-oss-scanner-free-claude-powered-vulnerability-de807d", "how-ai-is-reshaping-anthropic-s-workforce-17-e67383", "replit-switches-its-default-ai-provider-to-openai-35a0f0" ] }, { "slug": "generative-media", "label": "Generative media", "count": 34, "notes": [ "comfy-org-repackaged-flux-2-klein-4b-as-497873", "comfy-org-flux-1-dev-repackages-flux-1-9e7612", "comfy-org-flux-2-dev-packages-flux-2-c197e1", "deepbeepmeep-publishes-a-quantized-repack-of-minimax-h3-d7ffe5", "hosted-api-priced-at-0-014-per-megapixel-b7af43", "apache-2-0-weights-drop-in-workflow-templates-2c8979", "augment-your-livekit-agent-with-modal-gpus-3c5d04", "better-speech-synthesis-for-audiobooks-bbec1d", "bytedance-s-volcano-engine-releases-seedance-2-5-a67d17", "comfy-org-published-an-archival-re-upload-of-bd8372", "comfy-org-publishes-repackaged-single-file-safetensors-of-9c4ca5", "openspline-give-a-face-to-your-voice-ai-72a768", "tencent-releases-hy-nomas-an-open-interactive-video-c43eeb", "universal-modder-adds-um-voidless-a-fourth-asset-fedeb6", "ace-step-1-5-the-most-powerful-local-c356d6", "and-dj-and-producer-josh-fawaz-saw-an-37b3c3", "answer-of-defendant-suno-inc-to-complaint-874f46", "ant-group-released-ming-3-0-tts-0-0b239e", "asi-es-la-odisea-creada-con-ia-de-a3f66f", "canva-launches-craft-image-model-with-2k-output-e06976", "canva-releases-canva-craft-its-first-in-house-429648", "flux-3-image-klein-f506b6", "getty-images-files-for-bankruptcy-amid-pressure-from-43fb4e", "ideogram-launches-vector-1-0-its-first-model-784bea", "nvidia-s-video-generation-with-matching-3d-motion-85d452", "project-continuum-runway-labs-a75d6f", "though-listeners-are-actively-searching-for-imperfect-sound-dc70d3", "training-ai-to-paint-with-code-8043dc", "wan2-2-animate-14b-is-alibaba-s-open-35181e", "amazon-polly-delivers-a-major-upgrade-to-its-8bb03e", "apache-2-0-licensed-available-at-huggingface-co-1e39f3", "one-year-in-putting-human-artistry-first-on-8d6f08", "black-forest-labs-and-fal-released-flux-3-c519d2", "nvidia-cosmos3-h-dreams-bringing-clinical-surgery-to-3caa7a" ] }, { "slug": "agent-security", "label": "Agent security", "count": 33, "notes": [ "misalignment-reports-and-notices-openai-alignment-34c551", "cohere-launched-confidential-computing-on-model-vault-in-4abaea", "gitlab-critical-patch-release-19-5-1-19-a1d606", "meta-launches-its-muse-personal-ai-agent-in-cd8116", "meta-s-muse-wonder-model-card-chem-bio-832f11", "over-the-past-several-months-we-have-substantially-7f9243", "reuters-openai-resumed-frontier-ai-training-on-october-5af8a9", "us-banks-move-to-restrict-third-party-ai-0bad17", "browserbase-1password-secure-sign-in-for-browser-agents-b90402", "engadget-openai-research-agent-spent-another-wallet-s-fff526", "fastmcp-4-1-1-bounds-regex-tool-search-c332f6", "microsoft-open-sources-quicksand-a-python-api-for-c3438d", "microsoft-quicksand-core-0-14-0-adds-bidirectional-08d1ef", "nvidia-launches-openshell-reference-architecture-for-safe-ai-bbb154", "one-bank-executive-told-axios-they-struggle-to-fb53c8", "openai-defends-firing-three-safety-researchers-citing-breach-6d1e2f", "openai-says-research-environment-agents-sent-training-and-d8fce9", "openclaw-enterprise-the-open-agent-platform-a47826", "us-prosecutors-memo-seeks-pretrial-detention-of-ex-a5f3a7", "zhipu-s-zcode-deletes-data-and-announces-compensation-f79f2b", "zuckerberg-called-muse-ready-despite-the-risks-new-87d5d4", "cohere-model-vault-verifiable-confidential-ai-inference-815b97", "community-banks-deserve-advanced-tools-to-keep-their-5d5394", "confidential-computing-on-cohere-model-vault-5e675b", "node-js-security-advisory-a-high-severity-rce-4b2e5e", "securing-quick-access-from-corporate-networks-with-aws-e38f6b", "the-emudhra-ca-compromise-and-ngrok-what-s-11293b", "typically-when-talking-about-ai-features-and-privacy-a5db1b", "update-gitlab-by-matching-your-installation-to-the-6043e2", "aimori-the-guardian-for-your-ai-prompts-browser-562acd", "the-department-of-homeland-security-published-a-notice-7fe61c", "the-php-project-released-security-updates-on-21-3da3dc", "yujunzhixue-s-dsh-purge-repository-describes-a-local-ce354b" ] }, { "slug": "harness-engineering", "label": "Harness engineering", "count": 29, "notes": [ "kimi-k3-tech-blog-open-frontier-intelligence-600d83", "the-fox-early-access-release-provides-a-scaffold-8c90b2", "caict-report-china-s-normalized-llm-token-prices-d63fcb", "deepseek-open-sources-deepinfer-the-inference-framework-behind-893234", "full-engineering-writeup-covers-dep2-prefill-wide-ep-b2079d", "kimi-k3-1-pro-is-live-on-ai-e474c2", "nvidia-has-published-the-technical-report-model-checkpoints-5c5ae7", "corridor-introduces-argos-a-fine-tuned-qwen3-5-bf46d6", "deepseek-harness-dsh-v0-2-0-rc-2-380dc2", "deepseek-v4-1-flash-from-deepseek-ai-is-baea9f", "diagrid-releases-diagrid-goose-durable-dapr-workflow-execution-5e5bb5", "harness-acquires-augment-code-s-cosmos-auggie-cli-93e0c6", "harness-ceo-jyoti-bansal-details-augment-code-asset-824e78", "lerobot-adds-dexmal-s-dm0-5-vision-language-c365ab", "managed-deepagents-0-9-0-released-on-pypi-a59d19", "minimax-releases-open-weights-for-minimax-m3-1-4125a5", "nvidia-nemotron-3-5-bolt-e4b-small-open-c12f6c", "stanford-cs146s-lecture-by-lee-robinson-explains-how-dc4775", "the-methodology-of-nerfbench-7d0aa8", "we-gave-ai-agents-a-mobile-dev-environment-b3e860", "zed-1-22-0-stable-release-19bc66", "adaptive-agentic-intelligence-framework-using-strands-agents-78743d", "factory-now-supports-slackhq-delegations-b8719e", "how-cerebras-deploys-trillion-parameter-ai-models-06db8d", "ifm-open-sourced-rl360-the-rl-infrastructure-behind-4e4318", "nvidia-released-flare-a-local-inference-server-for-594e36", "recursive-releases-flint-40b-model-for-local-coding-6ca7fb", "recursive-s-ai-built-coding-model-runs-on-1efcdc", "the-harness-integrates-with-sglang-vllm-and-the-186be6" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 27, "notes": [ "opencode-s-experimental-background-compaction-setting-lands-in-0dafb8", "researchers-introduce-decision-oriented-programming-a-paradigm-making-57c70c", "awesome-gamedev-agent-skills-74-game-dev-skills-edfd71", "a-habr-author-published-a-hands-on-evaluation-404e56", "linus-torvalds-on-lkml-rules-against-using-ai-2a65c2", "linus-torvalds-rejects-ai-coding-bans-as-linux-be23ea", "llm-echo-0-5-released-on-pypi-with-d33b97", "measuring-the-ai-frontier-for-real-world-alignment-e8b10d", "moonshot-ai-quietly-rolls-out-kimi-k3-1-94b950", "opencode-go-adds-moonshot-s-kimi-k3-1-b43adf", "python-sdk-v5-and-js-ts-sdk-v6-b93c5c", "release-desktop-v0-0-43-cline-cline-696a12", "scale-agentic-coding-workflows-with-meta-s-muse-8364b4", "the-verge-openai-says-its-gpt-6-astra-70d07c", "torvalds-merges-docs-7-4-pull-removing-the-b9de7a", "atlassian-s-2026-state-of-the-developer-experience-ea0fd8", "how-to-use-decision-models-for-evals-this-cdc2f3", "openai-s-own-agents-turned-off-security-monitoring-4d1aee", "turning-database-grading-into-rl-rewards-post-training-46f950", "cognition-launched-efficiency-based-auto-approvals-for-devin-657f14", "factory-now-supports-linear-delegations-analysis-aafd6f", "my-obituary-for-programming-analysis-d2d60d", "people-are-using-ai-to-make-their-smart-93d055", "reason-network-is-an-open-source-rust-daemon-b71aac", "advancing-coding-in-low-resource-programming-languages-analysis-920015", "earshot-gives-coding-agents-calibrated-hearing-describe-match-1bc321", "factory-is-partnering-with-databricks-to-bring-autonomous-400e46" ] }, { "slug": "security", "label": "Security", "count": 24, "notes": [ "bringing-post-quantum-cryptography-to-geographically-fenced-key-507697", "jpmorgan-has-invested-5b-in-us-security-and-2348be", "positioned-as-local-counterpart-to-hosted-orcacyber-zero-2f993d", "cisa-urges-immediate-patching-of-php2shell-php-flaws-b9b2c3", "cloudflare-and-deutsche-telekom-join-forces-for-a-d48bfc", "critical-vulnerabilities-in-php-cert-eu-security-advisory-7570e7", "how-weaviate-responded-to-the-emudhra-certificate-compromise-87b538", "openai-and-poland-s-nask-partner-to-advance-18aeae", "php-serves-up-to-70-of-the-web-333d66", "cloudflare-has-achieved-south-korea-s-isms-certification-e4bb8e", "cloudflare-deploys-protections-against-php2shell-vulnerability-9a828a", "emudhra-intermediate-ca-compromise-analysis-a954d2", "nvidia-advances-agentic-security-for-media-from-european-d09a9b", "php2shell-security-advisory-two-critical-rce-flaws-in-1be379", "using-locked-chats-in-chatgpt-analysis-480f44", "ben-werdmuller-s-october-20-2026-article-says-ea6446", "building-overexposed-understanding-application-exposure-analysis-8c2f2d", "homeland-security-ordered-every-federal-civilian-agency-to-28b4de", "keep-sensitive-conversations-out-of-view-with-locked-919d7c", "on-public-cyberbench-deepseek-v4-1-flash-scores-db4c38", "us-government-bans-deepseek-from-federal-civilian-agencies-b72faf", "cve-2026-88771-lets-attackers-run-commands-on-3e350b", "openai-added-four-more-misbehaving-agent-incidents-to-8cc160", "phishing-emails-are-baiting-your-ai-assistant-analysis-71783e" ] }, { "slug": "github", "label": "GitHub", "count": 23, "notes": [ "terrascope-braille-rendered-terminal-world-map-with-live-9ce3e2", "code-at-github-com-agostini335-mvmae-accepted-to-1e2613", "discover-local-models-in-github-copilot-cli-1b8ee0", "factory-says-agent-effectiveness-is-now-generally-available-e5d7b0", "github-open-sources-reviewbench-a-reproducible-benchmark-and-b50aea", "how-vercel-agent-accesses-production-safely-01e121", "hydrafusion-updates-and-new-max-option-ffb3d8", "poppy-on-cloudflare-workers-github-repository-irvinebroque-poppy-c78edd", "release-rea-agents-v6-7-0-morluto-rea-db3ec6", "release-v0-2-1-rc-3-deepseek-ai-4fc601", "reylo-0-3-0-released-import-sql-files-e8f7de", "rfc-7230-6-1-compliance-issue-hop-by-6c9c29", "v1-7-3-least-privilege-postings-checked-before-c5c48b", "a-crop-mapping-benchmark-for-the-tropics-73ab07", "contained-oct-21-2026-litellm-release-v1-107-695e77", "dataset-on-huggingface-code-on-github-accepted-at-5c6cec", "scenario-open-sourced-its-blender-plugin-with-500-38c6f4", "what-the-heck-is-the-public-suffix-list-c28ae3", "as-traffic-grew-exponentially-we-moved-99-99999-9d909d", "asimov-1-v1-of-asimov-an-open-source-9897d4", "compromised-litellm-1-107-0-on-pypi-installs-138876", "code-and-pipeline-available-at-github-com-ponhvoan-62f346", "starsift-check-whether-the-accounts-behind-a-github-4f2a81" ] }, { "slug": "evals", "label": "Agent evals", "count": 22, "notes": [ "agentphishing-cmu-benchmark-measures-how-llm-email-agents-b197a6", "four-new-benchmark-reviews-we-graded-one-as-d411d1", "our-latest-batch-includes-4-new-reviews-1-b778bd", "techcrunch-colossal-biosciences-unveils-genomorph-ai-platform-to-cbae2f", "universal-modder-merges-game-recon-step-1b-with-09b78e", "we-don-t-review-our-own-benchmarks-instead-e4fd67", "we-recently-launched-benchmark-reviews-an-initiative-led-073033", "benchmirt-disentangling-safety-and-general-capabilities-in-llm-7d74ef", "bring-deep-sagemaker-ai-expertise-to-your-coding-53da91", "humanity-s-sixth-sense-hss-18ce4e", "nvidia-cosmos-opens-new-tools-to-train-evaluate-55aee4", "openai-s-math-solutions-aren-t-meeting-the-1fc6f6", "psychopomp-describe-a-scene-in-code-and-get-d0012a", "beyond-g-mean2-complementary-metrics-for-monitorability-27-a0386b", "experiments-now-show-which-metric-moved-and-why-cff081", "kaggle-and-rehearsals-launched-human-behavior-bench-scoring-0fd223", "mistral-advances-ai-for-engineering-with-nvidia-analysis-a7149a", "the-use-of-ai-in-science-raises-questions-3f80f1", "using-thousands-of-ai-agents-to-uncover-new-6b957c", "artificial-analysis-launched-aa-personalitybench-scoring-36-llms-01939f", "strengthening-ai-safety-evaluation-through-independent-third-party-f7e046", "unified-uv-run-eval-cli-replaces-separate-eval-573358" ] }, { "slug": "codex", "label": "Codex", "count": 21, "notes": [ "agmsg-1-6-1-fixes-agmsgd-notifications-to-51ceb7", "openai-ships-codex-cli-0-164-0-with-7e9e4a", "we-serve-data-in-milliseconds-to-more-than-0c5559", "daybreak-openai-for-cybersecurity-9bad37", "distributed-agent-sun-et-al-a04202", "fresh-dots-updates-create-in-the-mobile-app-7fedc1", "grounding-frontier-intelligence-in-enterprise-context-bc04bd", "how-stackai-by-asana-used-astra-codex-and-99cf2e", "openai-devday-2026-recap-for-developers-b08bb2", "openai-devday-2026-the-biggest-news-and-announcements-71fc9d", "openai-gives-codex-reusable-cloud-environments-that-work-3819f9", "openai-halves-pro-200-usage-and-launches-a-f01113", "openai-reports-a-full-codex-outage-on-sept-77f1d3", "provenance-signals-in-openai-generated-content-be979b", "response-speed-and-usage-resets-in-work-and-a51cd3", "then-on-thursday-metr-and-redwood-research-published-b866d7", "to-our-knowledge-frontier-ai-found-these-vulnerabilities-493b48", "cloudflare-links-vulnerability-discovery-to-defense-with-openai-10fe99", "metr-and-redwood-research-spent-three-weeks-embedded-fd7d77", "if-you-are-wondering-whether-your-website-is-0632e0", "shared-context-between-gpt-rosalind-s-scientific-reasoning-c64ee5" ] }, { "slug": "inference", "label": "Inference", "count": 19, "notes": [ "disaggregated-deployments-can-assign-precision-by-serving-phase-01b3fb", "engineers-at-humans-accelerated-mixture-of-experts-decode-840662", "spark-ds41f-1025-release-serves-deepseek-v4-1-86e402", "hi3d-open-sources-hi3d-v1-0-formerly-sparc3d-7328f0", "speculative-decoding-uses-the-smaller-draft-model-to-48c074", "fix-mega-mhc-split-barriers-under-cuda-graph-6ee79b", "hugging-face-has-published-webgpu-kernels-ai-onnx-13e715", "kapitsa-v0-2-15-an-open-source-apple-e8f061", "liquid-ai-has-published-lfm2-5-24b-a2b-d82305", "liquid-ai-launched-ripl-a-sub-50ms-inference-10eb44", "announcing-humming-fast-flexible-kernels-for-quantized-inference-67a0bf", "stop-guessing-auto-scaling-targets-for-tts-on-48e817", "eagle-4-ships-as-eagle-harness-a-source-7bdb0f", "fal-and-fireworks-published-a-technical-analysis-of-fb2e4d", "deepseek-open-sourced-deepinfer-the-apache-2-0-4c5d90", "kindling-ai-released-a-preview-exlr8-v2-quantization-830057", "new-default-rl-loss-ipo-masks-on-probability-e29bff", "how-deepgram-optimized-voice-ai-at-scale-with-3d2163", "improve-ttft-for-multi-turn-agentic-workloads-with-207bc9" ] }, { "slug": "mcp", "label": "MCP", "count": 19, "notes": [ "print3d-v0-1-0-open-source-agent-skill-1dd00d", "claude-code-cli-gains-mcp-tasks-for-durable-1711e7", "devin-desktop-3-11-216-fixes-devin-local-96caa1", "open-responses-api-spec-joins-the-agentic-ai-f44597", "uber-uses-a-similar-setup-to-clean-up-200090", "cursor-web-release-notes-c60558", "fly-io-launches-the-fly-mcp-server-mcp-a5495a", "framer-ships-official-mcp-connector-for-claude-codex-79c0ab", "getting-the-source-right-not-just-the-fact-829a2e", "holo4-powering-generalist-computer-use-agents-7a01cd", "krea-s-changelog-says-that-every-workspace-can-e56efd", "bundles-meshy-s-first-party-mcp-server-pinned-6d952a", "just-ai-published-a-technical-account-of-training-0e5091", "slack-v0-docs-3bf948", "build-a-semantic-layer-for-agentic-ai-on-b577b3", "build-a-semantic-layer-for-agentic-ai-on-ec1cc0", "from-data-to-decisions-conversational-supply-chain-intelligence-b8bdf2", "look-ma-we-re-famous-analysis-6cba07", "now-in-agno-stateless-mcp-serving-analysis-b6980e" ] }, { "slug": "ai", "label": "AI", "count": 16, "notes": [ "cu-anschutz-one-year-kysa-8-data-show-74751d", "fighting-back-an-experimental-treatment-has-helped-people-853c71", "pocket-lab-a-10-chip-could-put-a-a46b63", "devanthro-becomes-plim-robotics-reveals-a-home-robot-e8f4a7", "the-tiny-robot-designed-to-solve-a-massive-a04ad2", "this-pixar-inspired-robot-will-help-you-find-42f412", "colossal-unveils-genomorph-ai-platform-and-avian-advances-8bd429", "deep-robotics-field-tests-dr02-humanoid-robots-for-475b57", "francis-halzen-icecube-principal-investigator-wins-2026-physics-824661", "iros-2026-by-the-numbers-papers-countries-topics-6777be", "meteomatics-launches-strato-a-reusable-multicopter-weather-drone-92fffc", "meteomatics-unveils-strato-meteodrone-a-multicopter-that-reached-cb30c3", "vk-ai-engineers-described-how-the-recommendation-system-7205fe", "ai-lab-offers-free-plumbing-and-hvac-repairs-892565", "idealab-ai-a-free-business-idea-catalog-launched-e18971", "published-in-foundations-of-computational-mathematics-preprint-on-fefe35" ] }, { "slug": "open-models", "label": "Open models", "count": 15, "notes": [ "kimi-k3-1-pro-tech-blog-advancing-open-73b81c", "introducing-our-training-api-6c2d0c", "kolibri-has-landed-a-sovereign-open-weight-model-55b53f", "leading-european-financial-services-organizations-embrace-open-models-3ceeb3", "nvidia-open-models-power-a-new-wave-of-29b9c7", "nvidia-transforms-european-retail-from-digital-discovery-to-5e0584", "pioneers-across-europe-deploy-nvidia-earth-2-open-91795f", "stepfun-publishes-step-5-preview-open-weights-600b-5a7d50", "olmoearth-v1-3-state-of-the-art-embeddings-30cce8", "olmoearth-v1-3-state-of-the-art-embeddings-d07deb", "how-shopify-turns-open-models-into-task-experts-edc9e2", "muse-spark-open-on-fireworks-meta-s-open-316301", "nvidia-open-models-advance-europe-s-sovereign-intelligence-fdf60b", "fine-tuning-gpt-oss-20b-for-multilingual-reasoning-f9bab1", "hemmingway-1-is-a-27b-open-weights-fine-5cab3a" ] }, { "slug": "compute", "label": "Compute", "count": 12, "notes": [ "moonshot-ai-releases-kimi-k3-1-pro-a-50d51b", "taiwan-to-add-advanced-ai-chips-and-servers-8ce017", "magic-state-cultivation-on-a-superconducting-quantum-processor-22a37c", "news-announcements-loudoun-county-va-e9cca8", "v1-9-0-gpu-indexing-and-host-binding-4b353b", "nvidia-cudss-now-generally-available-to-accelerate-engineering-1cdba3", "spain-based-startups-advance-regional-innovation-with-barcelona-511fb0", "datacenters-2027-the-year-of-acceleration-50gw-in-cc5fcf", "meta-detailed-prometheus-its-gigawatt-scale-supercluster-in-a2cf33", "meta-developed-dynamo-dimmer-to-adjust-gpu-power-a819b7", "nvidia-expands-advanced-technology-access-for-europe-s-840f0e", "three-things-are-pressing-at-once-doubts-about-7fdd6b" ] }, { "slug": "gemini", "label": "Gemini", "count": 12, "notes": [ "gemini-is-the-closest-it-s-been-to-43ce2b", "paperclip-v2026-1023-0-adds-private-tasks-bring-6dfa4b", "figuring-out-which-document-parser-to-use-can-80bc24", "four-gemini-4-argon-agents-beat-minecraft-in-165ca3", "gemini-4-argon-on-cursor-solid-security-score-2ae32c", "gemini-4-argon-safety-security-report-92f6da", "gemini-api-release-notes-changelog-d8dec3", "gemini-canvas-turns-dense-pdfs-into-interactive-web-513d58", "gemini-notebook-becomes-a-connected-app-inside-the-4fb418", "release-desktop-v0-0-47-cline-cline-80da65", "simon-willison-s-llm-gemini-package-is-available-5eb0ee", "today-s-php-security-release-and-how-we-cac9ed" ] }, { "slug": "agents", "label": "Agents", "count": 11, "notes": [ "grok-bot-v0-75-5-adds-a-vault-b509eb", "tiktok-launches-an-ai-shopping-assistant-and-one-ba3275", "agents-rebuilt-to-do-more-for-less-analysis-76f6d4", "bad-company-data-sends-your-gtm-agents-after-09ae49", "introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b", "patch-the-planet-global-response-coordination-as-agents-c9b065", "using-grok-bot-to-detect-and-respond-to-d14ef7", "apate-sends-ai-agents-undercover-to-get-scammed-abbe18", "eka-is-worth-1-45b-thanks-to-a-b400d5", "notion-s-agents-now-get-their-own-computer-2b4eb8", "stripe-shared-data-on-70-ai-assistants-with-253e75" ] }, { "slug": "generative-ui", "label": "Generative UI", "count": 11, "notes": [ "openai-developer-community-announces-decisions-api-general-availability-759b4d", "perplexity-releases-decider-for-mac-to-run-pplx-25c390", "pydantic-ai-v2-61-0-adds-structured-citations-2f80e1", "clef-cloudflare-cloudflare-workers-ai-docs-e1bf6c", "decisions-api-is-now-available-in-public-beta-3cfa31", "open-d1-edge-decision-models-for-text-vision-e8468e", "build-more-responsive-apps-with-the-decisions-api-e5e19c", "mammoth-1-and-elephant-1-frontier-decision-models-389fb0", "the-top-ranked-decision-models-in-the-world-291d74", "two-new-decision-models-from-baseten-are-free-942221", "nvidia-kumo-relational-outperforms-supervised-machine-learning-on-6b45ee" ] }, { "slug": "api-design", "label": "API design", "count": 10, "notes": [ "mercury-agent-1-4-0-adds-a-plugin-b95875", "vercel-cli-adds-vercel-budgets-to-view-and-297ab7", "warp-documents-microsoft-teams-and-azure-devops-services-555b23", "cloudflare-forge-open-source-pipeline-for-generating-sdks-0c287e", "gateway-and-workers-integration-is-now-in-public-6bbcfb", "grok-bot-gets-its-own-email-address-at-ab6388", "new-web-search-tool-in-the-agents-sdk-9959ac", "openai-decisions-api-reaches-general-availability-with-gpt-8932f3", "sakana-ai-launched-aquarium-an-ontology-based-agent-87bed2", "factory-now-supports-jira-delegations-485380" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 9, "notes": [ "introducing-the-baseten-environment-network-ed257f", "dsx-pod-reference-design-for-ai-factory-infrastructure-5a11ea", "the-science-behind-torchneuron-amazon-s-new-pytorch-5e7fcd", "amazon-to-open-berlin-lab-invest-16-billion-ea466b", "nvidia-introduces-dsx-pod-ai-factories-to-accelerate-e0ac7f", "introducing-torchneuron-native-pytorch-backend-for-aws-trainium-17b313", "microsoft-reportedly-pulled-out-of-a-planned-ai-f2ccf9", "sonicboom-sonicboom-is-an-ai-framework-written-from-31e9d0", "neocloud-stocks-slide-amid-rising-borrowing-costs-shares-773eee" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 8, "notes": [ "ex-openai-science-head-kevin-weil-launches-warp-95f46e", "techcrunch-openai-acquires-four-year-old-biotech-software-b23d45", "disney-agrees-to-hire-the-majority-of-character-dfc753", "meta-proposes-a-new-class-of-stock-to-5fd9e9", "the-ai-app-that-triggered-kids-safety-fears-e36423", "how-we-built-chatgpt-s-novel-ad-stack-aac391", "disney-bids-to-buy-character-ai-from-bankruptcy-bf658f", "disney-hires-more-than-60-character-ai-employees-287414" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 8, "notes": [ "zoom-acquires-onyx-open-source-enterprise-search-that-c840e0", "integrated-embedding-on-the-documents-api-b97822", "when-vectors-became-an-implementation-detail-1f8f00", "eu-data-residency-for-the-parallel-search-api-d73c08", "cohere-rerank-5-more-relevant-context-for-rag-bba6d4", "exa-crosses-100m-in-annualized-revenue-analysis-7e991a", "nvidia-kumo-forecast-combines-pretrained-time-series-representations-1929c9", "new-scifullbench-benchmark-provides-full-paper-queries-and-9c116b" ] }, { "slug": "agent-protocols", "label": "Agent protocols", "count": 7, "notes": [ "ben-werdmuller-argues-that-news-will-increasingly-be-5502fa", "openai-usage-policies-effective-november-20-require-prior-c0d549", "guidance-for-travel-concierge-using-amazon-bedrock-agentcore-122af0", "walmart-wants-ai-agents-to-do-your-shopping-6c44ac", "almost-100-000-commerce-com-domains-now-give-9ba7e5", "runity-published-a-case-study-on-extending-its-f0364a", "companion-deepseek-recipe-mit-handles-chat-completions-messages-eba1d4" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 6, "notes": [ "the-model-that-didn-t-exist-so-you-f74bab", "vals-ai-launched-the-inference-provider-index-comparing-60fbd1", "introducing-engines-ai-with-motion-8ae976", "opencode-makes-60-deepseek-v4-1-flash-quota-d49960", "exa-ranks-1-on-the-artificial-analysis-search-60afad", "openclaw-creator-says-he-spent-10-million-in-729428" ] }, { "slug": "ai-security", "label": "AI security", "count": 6, "notes": [ "aisi-has-committed-to-security-changes-including-running-caa0c2", "china-linked-hackers-breached-connectwise-using-ai-penetration-a92d8d", "openai-employee-arrested-for-allegedly-hiding-a-sophisticated-0d3e4f", "four-z-ai-founders-co-led-ai-project-6be122", "nvidia-confidential-computing-expands-enabling-frontier-model-builders-d991c7", "gpt-6-sol-was-accessed-via-openai-s-92dbca" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 5, "notes": [ "building-laminar-ramp-s-managed-agents-platform-c45176", "there-s-new-free-models-available-in-cline-326dd7", "making-every-harness-durable-analysis-9f2d73", "open-apache-2-0-harness-and-27b-checkpoint-2846e5", "three-recursive-learning-mechanisms-credit-optimization-harness-replay-aa797b" ] }, { "slug": "agent-loops", "label": "Agent loops", "count": 5, "notes": [ "openai-agents-api-adds-scheduled-automations-with-cron-cf145e", "langchain-s-langsmith-cloud-changelog-says-langsmith-engine-573fd2", "humans-launched-boba-a-text-based-assistant-for-1b4853", "qwopus3-8-27b-flash-v2-is-an-apache-526d19", "welcome-rl-environments-to-the-hub-478980" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 5, "notes": [ "ai-engineer-nyc-2026-day-1-livestream-bridgewater-9a4497", "chatgpt-space-is-a-shared-hub-for-work-173b17", "chroma-joins-replit-chroma-cloud-will-shut-down-b20eaf", "introducing-frontiermemory-a-new-way-to-evaluate-agent-43a03f", "every-memory-has-two-clocks-when-the-event-ecf7f8" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 5, "notes": [ "release-candidate-v1-0-0-rc-3-huggingface-32da6a", "git-for-windows-2-56-0-3-dbc4ba", "simon-willison-released-version-0-2-of-llm-8816ed", "websocket-support-in-vercel-functions-is-now-generally-1650a6", "introducing-tsrs-a-port-of-typescript-to-rust-2e9bee" ] }, { "slug": "governance", "label": "Governance", "count": 5, "notes": [ "ebay-updates-global-privacy-policy-to-disclose-ai-518e9b", "meta-was-ordered-to-pay-25b-in-new-df69ea", "openai-safety-report-lead-david-robinson-resigns-says-2b8f3a", "chatgpt-now-lets-you-lock-sensitive-chats-behind-62b9cb", "super-intelligence-force-to-address-cyber-bio-ai-9bf5f4" ] }, { "slug": "open-source", "label": "Open source", "count": 5, "notes": [ "waytator-screenshots-and-video-recording-on-wayland-d8839b", "mrdoob-releases-slop-new-york-an-open-source-67f6ee", "hi3d-1-0-formerly-sparc3d-image-conditioned-3d-63944a", "obsidian-1-15-0-replaces-mermaid-js-with-2aaba8", "roboparty-released-rp1-a-full-stack-open-source-2ac942" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 4, "notes": [ "america-s-sbdc-announces-partnership-with-openai-to-c0b3a8", "andrej-karpathy-says-only-5-000-people-have-6315fb", "asu-knight-center-report-catalogs-160-newsroom-ai-784b65", "the-wharton-school-of-the-university-of-pennsylvania-cb97aa" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 4, "notes": [ "1password-surpasses-500-million-arr-on-enterprise-ai-0fc53e", "advancing-digital-sovereignty-with-greater-control-and-choice-0b7e76", "micro1-announces-commitment-to-spend-1b-on-enterprise-bea414", "germany-wants-to-lead-the-world-in-industrial-1fdd5d" ] }, { "slug": "local-models", "label": "Local models", "count": 4, "notes": [ "ist-austria-daslab-releases-gsq-rco-non-uniform-f02d77", "nvidia-released-nemotron-3-5-bolt-e4b-4gb-5d9d76", "mps-in-pytorch-2-15-analysis-02ed3b", "perplexity-computer-now-runs-fully-local-on-apple-c23b92" ] }, { "slug": "memory", "label": "Memory", "count": 4, "notes": [ "svs-in-milvus-reshaping-the-recall-memory-speed-0a935b", "modelbest-and-openbmb-launch-minicpm-v-4-7-23d8e1", "the-algorithmic-self-portrait-deconstructing-memory-in-chatgpt-8c013f", "altk-evolve-how-learned-memory-tops-the-2-fae5de" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 4, "notes": [ "beats-supervised-biomedclip-and-chexagent-baselines-on-mimic-1a7014", "datologyai-open-sources-zephon-an-elastically-deterministic-data-1c88d9", "meituan-s-longcat-2-5-preview-on-opencode-a86fea", "validated-on-56-hours-of-wild-lemur-footage-9761ce" ] }, { "slug": "sandbox", "label": "Sandbox", "count": 4, "notes": [ "nvidia-vera-cpus-double-throughput-for-modal-sandboxes-14b1aa", "openai-s-agents-api-documentation-now-describes-pre-c3eb29", "thehumanbox-free-civilization-sandbox-in-your-browser-analysis-117bec", "perplexity-open-sourced-the-sandbox-that-keeps-its-10afd0" ] }, { "slug": "agent-tools", "label": "Agent tools", "count": 3, "notes": [ "vercel-ai-sdk-adds-a-web-fetch-tool-24b855", "vercel-releases-ai-sdk-tool-web-fetch-1-8fc36b", "simon-willison-released-llm-tools-datasette-0-1-3c9d33" ] }, { "slug": "ai-products", "label": "ai-products", "count": 3, "notes": [ "google-cloud-page-returns-a-404-948dc5", "ideogram-offers-editable-svg-generation-and-image-54b01a", "openai-describes-a-conversation-aware-chatgpt-ads-c8c989" ] }, { "slug": "automation", "label": "Automation", "count": 3, "notes": [ "tesla-has-begun-producing-its-first-commercial-grade-9d0d9e", "aligning-dispatch-to-reality-using-reinforcement-learning-to-d0b2f6", "west-japan-railway-company-jr-west-has-decided-8ee0e1" ] }, { "slug": "enterprise-agents", "label": "Enterprise agents", "count": 3, "notes": [ "autosynthdata-generating-training-data-for-enterprise-agents-fc7609", "nvidia-agent-toolkit-expands-grounding-enterprise-agents-in-af91c3", "set-up-and-manage-teams-and-team-tasks-92191b" ] }, { "slug": "agent-identity", "label": "Agent identity", "count": 2, "notes": [ "glean-makes-agent-identity-generally-available-so-agents-808423", "that-s-a-wrap-on-day-2-of-205b9b" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 2, "notes": [ "aws-and-uc-berkeley-open-source-infinity-a-5436d7", "openclaw-has-released-v2026-10-6-a-patch-f66d83" ] }, { "slug": "skills", "label": "Agent skills", "count": 2, "notes": [ "skills-docs-by-langchain-87dc10", "specification-agent-skills-d638fc" ] }, { "slug": "code-review", "label": "Code review", "count": 2, "notes": [ "reviewbench-an-open-benchmark-for-ai-code-review-c9dcf4", "chopin-proof-driven-development-kit-from-githubnext-and-d99fcc" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 2, "notes": [ "origin-range-requests-serving-large-files-without-range-b93c0a", "serverless-routing-made-sticky-analysis-5547c6" ] }, { "slug": "model-routing", "label": "Model routing", "count": 2, "notes": [ "llm-router-v0-1-0-go-service-exposing-43bb2e", "cutting-our-inference-bill-by-68-by-being-8790d7" ] }, { "slug": "observability", "label": "Observability", "count": 2, "notes": [ "pricing-cloudflare-observability-docs-1c8117", "profiling-torchneuron-part-1-system-level-profiling-on-6f6b9b" ] }, { "slug": "orchestration", "label": "Orchestration", "count": 2, "notes": [ "how-doordash-built-a-next-generation-real-time-eb6790", "the-thin-orchestrator-how-we-made-our-ai-be8c58" ] }, { "slug": "agent-context", "label": "Agent context", "count": 1, "notes": [ "replit-is-acquiring-chroma-the-open-source-vector-12b314" ] }, { "slug": "agent-interfaces", "label": "Agent interfaces", "count": 1, "notes": [ "simular-s-agent-s4-evolves-agent-interfaces-so-0ce52b" ] }, { "slug": "computer-use", "label": "Computer use", "count": 1, "notes": [ "cursor-s-in-editor-agent-gains-computer-use-2e9fa3" ] }, { "slug": "context-engineering", "label": "Context engineering", "count": 1, "notes": [ "new-paper-proposes-abstract-compression-for-conversational-context-b183c5" ] }, { "slug": "design-systems", "label": "Design systems", "count": 1, "notes": [ "mistral-makes-its-fourth-acquisition-in-eight-months-47fa7b" ] }, { "slug": "engineering-management", "label": "Engineering management", "count": 1, "notes": [ "ruby-on-rails-creator-declares-pencils-down-on-08a48a" ] }, { "slug": "frontend", "label": "Frontend", "count": 1, "notes": [ "tanstack-start-1-0-is-stable-501a22" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 1, "notes": [ "current-affairs-staff-ratify-union-contract-that-bans-51167b" ] }, { "slug": "hermes", "label": "Hermes", "count": 1, "notes": [ "nous-research-s-hermes-agent-mit-licensed-self-a50abd" ] }, { "slug": "knowledge-work", "label": "Knowledge work", "count": 1, "notes": [ "litpivot-developing-research-ideas-through-dynamic-contextualization-within-5ab8c2" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 1, "notes": [ "office-intelligence-for-long-running-ai-work-3-4eba7c" ] }, { "slug": "org-design", "label": "Org design", "count": 1, "notes": [ "ai-can-weaken-relationships-between-people-and-teams-473058" ] }, { "slug": "pricing", "label": "Pricing", "count": 1, "notes": [ "moonshot-gives-every-kimi-subscriber-a-one-time-23ac78" ] }, { "slug": "prompt-injection", "label": "Prompt injection", "count": 1, "notes": [ "a-habr-author-tested-prompt-injection-against-a-fd59e3" ] }, { "slug": "prompts", "label": "Prompts", "count": 1, "notes": [ "on-tuesday-amazon-unveiled-kiro-trace-an-internal-a63578" ] }, { "slug": "rumors", "label": "rumors", "count": 1, "notes": [ "unconfirmed-posts-say-claude-helped-verify-123-open-source-6dd32a" ] }, { "slug": "tool-use", "label": "Tool use", "count": 1, "notes": [ "meta-released-muse-glimmer-30b-an-open-weight-aa31c3" ] }, { "slug": "tools", "label": "Tools", "count": 1, "notes": [ "add-experimental-reterminal-sticky-muse-and-link-e-675e6c" ] } ] }, { "key": "2026-w42", "label": "Oct 12–18, 2026", "start": "2026-10-12", "end": "2026-10-18", "count": 398, "html": "https://newruntime.com/weeks/2026-w42/", "summary": { "label": "New Runtime synthesis", "headline": "Open models and agent infrastructure are making deployment easier, while oversight and security become the constraint", "paragraphs": [ "The main shift is from model choice toward deployable systems. Muse Spark Open, Beam and other open-weight releases are appearing across inference platforms, while local models and lower-cost routing widen where builders can run them. At the same time, agents are moving into on-call work, research, shopping and enterprise workflows. Together, these notes suggest access to capable models is broadening, but the value increasingly lies in the harness, integrations and operating boundary around them.", "That expansion makes some work cheaper and easier to start, but it raises the cost of control. OpenTelemetry traces and agent-session visibility are becoming product features; yet an audit found weaknesses in many third-party sandboxes, and reports of agents taking consequential actions without adequate grounding underscore that autonomy can fail outside the code editor. Security teams and buyers gain leverage from observability and containment; teams that treat model access as equivalent to safe execution take on more risk.", "This continues the prior weeks’ move toward persistent agents and decision-oriented systems, but broadens it: this week’s releases put open models, robotics, voice and multimodal tools into more channels, while benchmarks increasingly test reliability and real tasks rather than just model output. Builders and buyers should evaluate the full system—cost per completed task, traceability, sandbox boundaries and failure behavior—before increasing autonomy." ], "key_stories": [ { "slug": "meta-released-muse-spark-open-an-816b-total-5c27ac", "why": "A major open-weight release arriving through multiple platforms shows that model availability is spreading beyond a single vendor’s serving stack." }, { "slug": "arena-s-new-agentic-max-router-matches-claude-47c5aa", "why": "Lower cost per task for a routed agent suggests orchestration and model selection can matter as much as choosing a top model." }, { "slug": "cognition-launched-devin-oncall-an-ai-sre-that-44f031", "why": "An AI SRE that investigates alerts and opens fixes shows agents moving from code generation toward operational responsibility." }, { "slug": "audit-of-10-third-party-sandboxes-found-8-54fdd3", "why": "Sandbox weaknesses make the execution boundary a practical security issue as agents gain access to tools and networks." }, { "slug": "claude-managed-agents-the-messages-api-and-the-094b3f", "why": "OpenTelemetry support makes tracing agent activity part of the platform layer needed to manage more autonomous workflows." }, { "slug": "benchmark-grades-mergeability-using-rubrics-from-20-open-7bcbc3", "why": "A mergeability benchmark based on maintainer rubrics emphasizes whether agent-produced code can be accepted, not merely whether tests pass." }, { "slug": "open-weight-takes-71-5-of-tokens-and-1f3202", "why": "Open-weight models taking most token usage but a smaller share of spend points to widening access without equivalent control of economics." } ], "source_note_count": 398, "generated_at": "2026-10-26T12:50:23.482Z" }, "topics": [ { "slug": "models", "label": "Models", "count": 41, "notes": [ "a-habr-analysis-argues-that-prismml-s-bonsai-eb9001", "liquid-ai-released-lfm2-5-24b-a2b-a-c47d5c", "looprpt-is-a-reinforcement-pre-training-framework-built-6a2c60", "deepseek-v4-1-flash-1017-is-on-ai-d34087", "introducing-s1-in-context-learning-for-robotics-analysis-85f282", "meta-shakes-up-the-frontiers-open-and-closed-f37025", "bank-of-america-brought-in-1-000-outside-504a49", "data-mixing-minus-the-guesswork-analysis-9fc7d6", "introducing-voice-standard-bench-50-models-ranked-by-71e9f1", "microsoft-ai-says-it-has-completed-pretraining-mai-9757e7", "openai-s-release-notes-say-gpt-live-1-eaaab0", "reflection-released-beam-a-501b-parameter-moe-with-d3b81d", "scaling-indic-pretraining-2-8-trillion-tokens-and-cc2843", "the-gradient-is-a-covariance-8-minute-read-5099f4", "are-all-system-one-models-created-equal-parallel-bb3793", "cern-and-mistral-join-forces-to-accelerate-ai-3361fa", "granite-switch-4-1-chat-decisions-and-speech-3d85e2", "muse-wonder-from-meta-superintelligence-labs-is-now-84824a", "open-weight-takes-71-5-of-tokens-and-1f3202", "proxyattn-a-training-free-method-from-researchers-at-59312c", "researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1", "thinking-machines-released-dimuon-a-new-in-house-16cede", "vercel-token-pass-bundles-glm-and-kimi-credits-1c8418", "a-separate-public-benchmark-analysis-evaluated-23-models-6f6beb", "about-400-scientists-want-a-global-watchdog-for-ac9849", "amp-puts-ai-agents-to-work-beyond-code-15c9f6", "artificial-analysis-launched-the-decision-index-v1-0-5ce031", "deploy-runway-s-video-models-on-baseten-analysis-3b89fa", "how-to-be-more-ambitious-a-tutorial-for-eb5e7a", "nemotron-3-5-lightning-is-free-on-ai-eb8db3", "physical-intelligence-launched-the-pi-api-beta-for-30a424", "spur-recently-published-the-content-telemetry-standard-which-beaccb", "a-habr-research-write-up-reports-a-controlled-b48dcc", "aa-3d-ranks-21-models-by-human-votes-ae34ef", "amazon-science-reports-a-method-called-priming-that-c86e28", "langchain-and-scaledown-announced-an-integration-for-using-504920", "prime-intellect-acquired-moondream-bringing-open-vision-models-1a82c6", "sarvam-launched-bulbul-v4-flash-a-distilled-tts-dea6f0", "when-do-we-need-on-policy-distillation-distilling-ca2f7b", "nobody-shipped-a-frontier-model-this-weekend-meta-404e76", "pplx-decider-v1-1-27b-hugging-face-repo-44e8c0" ] }, { "slug": "claude", "label": "Claude", "count": 23, "notes": [ "arena-s-new-agentic-max-router-matches-claude-47c5aa", "claude-reveals-ecdsa-curve-origins-analysis-d4ef04", "claude-code-changelog-md-at-main-anthropics-claude-59f743", "how-the-claude-code-team-does-code-review-efacb7", "why-i-m-rebuilding-my-indieweb-projects-with-59a150", "claude-mythos-5-designed-quantum-circuits-beating-best-1a6bcd", "an-author-on-habr-describes-building-aiworks-a-b51ea4", "claude-managed-agents-the-messages-api-and-the-094b3f", "simon-willison-reports-that-he-used-the-default-f8ddfb", "claude-code-can-write-multi-file-changes-faster-ee444e", "claude-opus-5-5-tops-vals-ai-s-3c4a0e", "claude-platform-can-now-emit-opentelemetry-traces-from-5c6a70", "gpt-6-1-sol-max-and-claude-fable-b17123", "simon-willison-released-web-edit-0-1a1-with-6d392c", "trace-agents-with-opentelemetry-on-the-claude-platform-a01c87", "claude-code-templates-cli-tool-components-mods-ui-54bc60", "epic-games-told-the-team-making-its-marvel-1021b8", "marketing-research-in-chatgpt-claude-analysis-bb13ab", "artificial-analysis-launched-a-leaderboard-for-cognition-s-b160a4", "claude-is-reportedly-turning-old-console-hits-into-bf574f", "during-a-test-claude-sent-philadelphia-police-a-2aad16", "kernel-git-torvalds-linux-git-linux-kernel-source-fe3569", "the-case-that-will-get-quoted-claude-haiku-5fcb2a" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 22, "notes": [ "deepseek-released-v4-1-flash-1017-a-snapshot-8347f0", "factory-made-delegations-generally-available-for-handing-droid-30660f", "our-harnesstax-analysis-examines-how-harness-choice-changes-80954d", "cognition-launched-devin-oncall-an-ai-sre-that-44f031", "cognition-wants-ai-to-take-your-on-call-7f4701", "ecc-packages-an-engineering-harness-for-coding-agents-b3b9b6", "helpfreedom-s-nv-repository-presents-a-neovim-0-974f8c", "ibragim-bad-published-a-guide-and-a-companion-e8d8ee", "larger-sibling-muse-wonder-full-is-locked-to-e67a35", "sakana-ai-released-the-agentic-scientist-an-open-918e08", "deepseek-v4-1-flash-usage-cost-rank-analysis-0123cf", "giving-coding-agents-a-docs-link-doubles-task-630606", "how-meta-built-forum-with-ai-native-engineering-832700", "muse-spark-open-an-open-weight-model-for-ddd53f", "a-habr-experiment-compared-a-solo-coding-agent-be4e61", "cognition-launched-a-devin-integration-for-pylon-that-80f825", "releasing-ai-generated-software-8-minute-read-analysis-12b186", "simon-willison-released-web-edit-0-1a0-analysis-8c7282", "this-concern-shows-up-in-developer-surveys-too-019ab4", "claude-motion-a-motion-design-toolkit-for-coding-e1ccf6", "habr-published-a-practical-article-on-reducing-coding-76eaaa", "solo-developers-are-using-ai-to-clone-big-fa0ad4" ] }, { "slug": "generative-media", "label": "Generative media", "count": 21, "notes": [ "beeftv-is-an-open-source-mit-licensed-ai-7e3a2d", "comfy-org-published-repackaged-moss-image-weights-for-2aed52", "now-you-see-me-imagine-making-your-tattoo-76ce8d", "odyssey-says-it-placed-a-fruit-fly-brain-62f71e", "notable-links-october-16-2026-analysis-210c26", "pika-launched-a-chatgpt-plugin-extension-embedding-its-aa1ead", "additionally-bmg-signed-its-own-partnership-with-suno-3c6210", "in-september-sony-became-the-first-music-company-bf4f5b", "jasco-adds-flow-matching-text-to-music-with-052f26", "major-labels-and-industry-firms-including-warner-bmg-6de432", "openai-and-meta-told-a-federal-judicial-panel-8bbdcf", "suno-aside-august-universal-sony-and-warner-backed-49f78f", "udio-suno-s-biggest-rival-has-faced-similar-1414b3", "welcome-flux-3-image-klein-in-diffusers-analysis-5e5c27", "drama-3-from-fishaudio-is-live-on-ai-9ed1b5", "flux-3-video-is-now-up-to-2x-2d1fca", "a-man-got-18-months-for-streaming-ai-a2e132", "day-one-comfyui-support-with-bf16-and-int8-7812b2", "delivering-real-time-text-to-speech-with-gradium-3fc8eb", "fish-audio-released-drama-3-an-open-weight-208ee1", "iris-3b-is-a-3b-pixel-space-text-71c4cd" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 19, "notes": [ "anthropic-s-october-16-research-covered-two-distinct-37747a", "the-business-transformation-has-become-a-software-process-e718c9", "rivian-wants-to-put-third-party-ai-agents-9175e5", "openai-and-anthropic-each-got-more-than-50-28ee98", "anthropic-tells-investors-quarterly-revenue-up-nearly-50-188b40", "working-out-how-best-to-use-ai-in-149c27", "anthropic-s-economic-index-ranks-singapore-second-among-e1c03c", "claude-plugins-official-plugins-code-modernization-readme-md-000530", "preparing-for-the-2026-us-midterms-anthropic-analysis-23217f", "redesigning-source-control-at-anthropic-analysis-bca01f", "singapore-office-and-government-mou-analysis-fffeb8", "the-defensive-argument-is-the-useful-part-and-9e95da", "ai-models-may-be-capable-of-self-replicating-445288", "anthropic-says-claude-produced-most-of-scion-s-97253d", "anthropic-s-system-trust-and-reporting-page-last-c4ac93", "in-the-blog-post-anthropic-says-it-identified-d21030", "over-the-last-few-months-we-ve-made-218d38", "redesigning-source-control-at-anthropic-anthropic-analysis-47501c", "amazon-is-reportedly-in-talks-to-buy-decart-f81dcb" ] }, { "slug": "google", "label": "Google", "count": 19, "notes": [ "announcing-dpop-support-for-google-oauth-securing-refresh-0a4963", "how-we-re-helping-the-world-prepare-for-78e4da", "humanizer-a-local-12b-model-that-rewrites-ai-b81290", "a-separate-public-health-study-cited-by-google-1c668d", "analyzing-technical-documents-with-deep-review-12-minute-37bb1c", "developers-built-google-docs-and-figma-so-others-78a17f", "google-antigravity-accelerating-antimicrobial-peptide-search-analysis-460207", "google-drops-the-developer-account-for-personal-vibe-4fa9ad", "google-harness2-for-recursive-self-improvement-github-repo-89c90b", "google-research-unveiled-a-geospatial-reasoning-agent-combining-8edb23", "research-2026-novel-capabilities-and-intelligence-analysis-ebd1fb", "an-off-the-shelf-openai-model-helped-hack-fa471e", "google-deepmind-and-singapore-government-announce-new-ai-1abbdb", "google-released-gemma-4-e4b-qat-q4-0-17e441", "use-google-ai-plan-credits-analysis-7930ff", "xbow-built-a-chrome-full-chain-exploit-affecting-097c4b", "gemini-4-argon-lets-you-reimagine-your-video-a06d77", "google-deepmind-s-alphaproof-nexus-solved-9-open-07c5ac", "isomorphic-labs-is-eyeing-a-40b-valuation-the-c815df" ] }, { "slug": "inference", "label": "Inference", "count": 19, "notes": [ "qwen-released-official-gguf-weights-for-qwen3-vl-6d4ccb", "vllm-ships-a-specialized-cute-dsl-decode-path-dd574e", "advancing-generative-recommendation-inference-our-collaboration-with-nvidia-67cd59", "janelia-and-collaborators-release-dpms-a-variational-framework-788229", "the-technical-report-describes-a-40-layer-causal-ba1f4b", "underlying-flux-2-klein-is-a-4b-rectified-4a4a74", "meta-s-muse-spark-open-weight-release-is-e68623", "lumi-documents-an-offline-ollama-desktop-integration-analysis-7a0d20", "push-based-kv-transfer-to-improve-ttft-and-0a2b7c", "ps5lm-llama-cpp-on-a-jailbroken-ps5-run-aba8a6", "scaling-agent-teams-reach-the-frontier-at-30-4e95cd", "mixture-of-experts-kernel-engineering-on-amazon-sagemaker-abd782", "toward-rl-on-mixed-precision-mixed-accelerator-samplers-e45555", "ai-scaling-vs-human-scaling-12-minute-read-2d8916", "built-on-gsq-and-rco-methods-from-ista-c27305", "endpoint-candidates-evaluate-and-promote-inference-endpoints-analysis-3de079", "project-maya-is-an-mit-licensed-inference-stack-19a0c1", "run-multi-node-inference-workloads-using-ray-llm-2fa041", "what-is-speculative-decoding-70-minute-read-analysis-3094d3" ] }, { "slug": "evals", "label": "Agent evals", "count": 18, "notes": [ "introduces-a-fast-track-rfc-for-agent-generated-d83647", "new-open-dataset-synth-dopaco-releases-8-800-91cde0", "rawmodels-unslop-harbor-datasets-at-hugging-face-analysis-7de3de", "turns-out-le-chonk-is-into-rocket-science-3f9b33", "a-team-of-35-researchers-put-together-50-1f091c", "ai-safety-nonprofit-metr-has-published-a-proposal-0b8b05", "alignment-metr-would-test-whether-current-or-near-f9db7f", "epoch-will-release-1-000-gameplay-videos-and-a6ce53", "extractbench-pro-the-long-tail-of-document-extraction-9e4ec3", "metr-explains-its-approach-in-a-companion-post-888250", "that-sparse-signal-permits-inefficient-traces-to-earn-3dcb56", "vals-ai-launched-robo-bench-testing-llms-on-226ca5", "arena-ai-replaced-bash-recovery-with-bash-reliability-0fc075", "artificial-analysis-publishes-per-benchmark-results-with-the-03859f", "deepseek-v4-1-flash-joins-the-aa-agentperf-8ea814", "teams-evaluating-the-pattern-should-define-the-target-7bd81f", "the-expensive-loop-problem-22-minute-read-analysis-03f819", "benchmark-grades-mergeability-using-rubrics-from-20-open-7bcbc3" ] }, { "slug": "security", "label": "Security", "count": 13, "notes": [ "oauth-2-0-for-browser-based-apps-analysis-8950b5", "bank-of-america-hires-1-000-contractors-after-b91da7", "two-robot-dogs-now-work-the-night-shift-f99546", "cloudflare-gateway-and-data-loss-prevention-have-inherited-26628b", "weaviate-security-release-medium-severity-fix-for-a-e0cb9c", "how-block-is-building-ai-powered-security-with-2ec292", "how-tamnoon-built-ai-powered-cloud-security-remediation-f9b887", "natural-language-security-queries-at-orca-security-using-e6d046", "northern-district-of-california-saudi-national-arrested-for-326e38", "on-tuesday-a-coalition-of-more-than-400-7351e5", "sentinelone-cuts-malware-analysis-from-weeks-to-hours-e2e382", "uber-s-ciso-on-building-an-insider-threat-6196c2", "hackers-heat-french-supermarket-fridges-sparking-fires-and-016080" ] }, { "slug": "agents", "label": "Agents", "count": 12, "notes": [ "vercel-and-shopify-make-agents-first-class-shoppers-b5d607", "who-let-the-bots-in-rethinking-captchas-in-6b5d53", "browserbase-and-parallel-the-right-search-for-every-c1e068", "trust-in-depth-giving-agents-more-autonomy-without-3ff173", "european-cities-deploy-nvidia-powered-ai-agents-to-5534db", "fanuc-accelerates-physical-ai-in-arc-welding-with-af0359", "moderna-s-co-founder-wants-ai-agents-regulated-a16753", "from-earth-to-orbit-france-s-ecosystem-delivers-bcf917", "how-we-made-our-docs-test-themselves-analysis-13a388", "no-more-data-entry-how-to-set-up-c9af92", "openclaw-published-version-v2026-10-3-a-patch-c55036", "using-specialized-data-to-improve-research-agent-accuracy-eade5e" ] }, { "slug": "github", "label": "GitHub", "count": 10, "notes": [ "aa-agentperf-methodology-replays-trajectories-from-agents-resolving-cf52ce", "sgs-slam-is-the-first-semantic-visual-slam-b618ff", "tbdub-is-a-github-project-for-local-video-138055", "the-si-ocp-paper-was-accepted-to-cdc-ecf333", "github-availability-report-september-2026-analysis-558a44", "technical-report-introduces-two-new-benchmarks-geoexpertqueries-and-e1a6b1", "nvidia-s-gpu-sharing-software-is-now-free-8c9a45", "all-systems-go-global-pc-app-ecosystem-ready-b7d255", "cadcraft-cadcraft-computer-aided-design-and-drafting-analysis-805f96", "code-available-at-github-com-gen-verse-rsi-665baf" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 10, "notes": [ "building-fast-filesystems-for-vms-analysis-b70ce3", "2026-world-cup-breaking-records-spiky-charts-and-83ce38", "backends-on-vercel-python-containers-and-daemons-analysis-d251c0", "bring-your-own-cloud-on-aws-is-now-bf9422", "vercel-daemons-is-now-in-public-beta-analysis-fbd09f", "inside-hyperdrive-how-we-solved-a-rare-hyperdrive-f53a20", "sber-said-its-data-center-in-balakovo-in-81a3ba", "when-strikes-hit-yandex-data-centers-many-assumed-8aa5a4", "under-the-hood-off-the-hot-path-how-2a2716", "ukrainian-drones-shut-down-yandex-s-vladimir-data-7b192a" ] }, { "slug": "open-models", "label": "Open models", "count": 10, "notes": [ "inclusionai-released-ling-3-1-flash-a-560b-af9ab3", "meta-s-muse-spark-open-now-available-in-831a6b", "meta-s-muse-spark-open-page-on-hugging-9dd8ba", "meta-muse-spark-open-is-live-on-together-2ca117", "meta-released-muse-spark-open-an-816b-total-5c27ac", "muse-spark-open-is-now-available-on-modal-c28c51", "the-base-model-card-describes-the-first-open-bbeb0a", "the-companion-jais-asr-notes-describe-an-open-1cbe78", "open-models-have-crossed-a-threshold-analysis-8bf2ef", "alibaba-s-upstream-release-is-a-7-billion-2604f5" ] }, { "slug": "gemini", "label": "Gemini", "count": 9, "notes": [ "optimizing-a-200-000-line-python-codebase-overnight-c6f69a", "instacart-opentable-adobe-shopify-gemini-4-argon-is-c60a77", "developing-gemini-4-argon-responsibly-and-safely-analysis-442f9c", "gemini-4-argon-also-brings-major-improvements-in-2795ad", "gemini-4-argon-is-now-available-for-developers-fbfa82", "gemini-4-argon-is-now-available-in-cursor-31ead0", "gemini-4-argon-now-available-in-opencode-it-a13315", "gemini-4-argon-the-first-gemini-4-model-e3012c", "you-can-connect-more-than-40-apps-to-7b067f" ] }, { "slug": "compute", "label": "Compute", "count": 8, "notes": [ "zephon-integrates-with-torchtitan-to-keep-global-batches-289602", "the-nordics-and-nvidia-show-what-s-possible-3c495c", "asml-posted-record-11-7b-q3-sales-and-e19104", "ensuring-similar-computational-resources-to-stateful-workloads-running-45481c", "from-enterprise-to-earthshot-uk-s-ecosystem-moves-520391", "meta-s-generative-ads-model-gem-scaling-the-dfded4", "microsoft-introduced-the-maia-300-ai-chip-4-bee46a", "microsoft-to-expand-azure-maia-system-to-multi-1b104b" ] }, { "slug": "governance", "label": "Governance", "count": 8, "notes": [ "china-s-ministry-of-commerce-and-ministry-of-f79f2f", "eight-states-sue-openai-over-alleged-failure-to-aaa3e2", "openai-agrees-not-to-use-trade-secrets-claimed-97bcb8", "attorney-general-james-urges-court-to-protect-children-b588a5", "iowa-kansas-and-six-other-states-sued-openai-016548", "as-a-result-the-standards-for-publisher-usage-c1f984", "microsoft-president-brad-smith-says-ai-faces-fewer-a351c6", "who-funds-the-independent-ai-evaluators-12-minute-52cdf6" ] }, { "slug": "local-models", "label": "Local models", "count": 8, "notes": [ "new-optimal-transport-method-aggregates-independently-trained-mixture-fa1819", "only-18-mb-of-onnx-weights-no-pytorch-61535a", "computers-built-for-local-ai-analysis-bb5065", "prismml-introducing-ternary-bonsai-2-2b-4b-and-8cd13c", "set-up-openclaw-out-of-the-box-on-386334", "scragnog-released-a-gguf-conversion-of-openmoss-s-90fc8e", "run-fast-private-agents-on-device-with-lfm2-2d40db", "ukisai-has-released-four-mixed-precision-gguf-versions-5d1285" ] }, { "slug": "agent-interfaces", "label": "Agent interfaces", "count": 7, "notes": [ "a-new-buddy-for-your-ios-home-screen-685849", "added-native-decisions-integration-in-gpt-live-analysis-ff9213", "introducing-grok-voice-3-0-analysis-045705", "muse-spark-open-from-aiatmeta-is-now-available-d826ea", "teaching-students-to-think-and-learn-through-questions-0a7a70", "trace-shadcn-s-component-for-showing-what-your-da66fe", "apple-introduced-homeview-a-home-display-with-a-d843ce" ] }, { "slug": "ai", "label": "AI", "count": 7, "notes": [ "some-scientists-think-a-hidden-force-could-be-a21c34", "mosqitter-is-turning-its-mosquito-traps-into-an-9ad670", "new-robot-design-uses-your-old-dusty-phone-cd826e", "tiny-robots-speed-medicine-through-body-to-the-fb03a7", "volvo-has-expanded-its-partnership-with-nvidia-to-9d6f9c", "paper-published-in-transportation-research-part-a-doi-7ceebb", "clinical-trial-update-first-patient-with-stargardt-disease-2a32a4" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 7, "notes": [ "agile-robots-is-training-robots-using-videos-of-9a916d", "deepseek-v4-flash-vision-exp-release-multimodal-api-0831ab", "liquid-ai-released-d1-ultra-a-multimodal-decision-3c7286", "apertus-1-5-release-notes-analysis-7dac13", "day-0-chip-support-for-rockchip-houmo-intel-1bc826", "openbmb-released-minicpm-v-4-7-1b-an-6db742", "a-community-bun-grounding-demo-already-renders-bounding-d31287" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 7, "notes": [ "search-and-fetch-apis-now-public-at-0-5421c8", "yukun-huang-sanxing-chen-jian-pei-manzil-zaheer-eea460", "cloudflare-and-openai-making-ai-discovery-better-for-b2ae17", "on-demand-mode-for-pinecone-byoc-analysis-e48587", "ruri-v3-is-a-japanese-text-embedding-family-a95e29", "web-search-for-browserbase-is-now-powered-by-52db78", "castle-training-natural-language-search-before-the-first-6a8432" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 6, "notes": [ "bumble-agrees-to-a-1-3b-acquisition-deal-805e5f", "nvidia-is-buying-a-stake-in-doctolib-france-b65f6a", "spacexsi-bundles-x-premium-grok-and-cursor-into-c88115", "an-update-on-character-ai-analysis-c52d53", "the-getty-family-and-koch-equity-development-commit-3a4c16", "altman-backed-meanwhile-raised-37-5-million-at-d71da2" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 6, "notes": [ "cloud-data-platforms-and-ai-infrastructure-vendors-increasingly-5fc0d4", "hotter-inflation-data-revived-rate-fears-and-sent-344160", "salience-labs-raises-125-million-for-all-optical-5143f5", "modal-crosses-1b-in-annualized-run-rate-and-f3bd6f", "ocp-summit-2026-powering-ai-with-open-innovations-01cd7a", "nvidia-oracle-coreweave-shares-fell-following-openai-revenue-820ac9" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 6, "notes": [ "deploy-with-void-cloudflare-blog-analysis-ef4678", "python-on-vercel-is-now-generally-available-analysis-ca9d12", "a-new-kind-of-in-house-tool-analysis-6da20f", "origin-cursor-s-code-hosting-platform-now-supports-d22f66", "repository-checkout-and-git-lfs-downloads-are-now-655788", "three-commands-record-learn-gen-then-import-the-eae4bc" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 5, "notes": [ "open-sourced-mirror-cli-the-rust-core-agent-6deb32", "cline-agents-harness-unlocks-benchmark-leading-performance-from-03fd56", "october-2026-langchain-newsletter-everything-we-shipped-at-7e3426", "how-third-moment-research-runs-10-000-agents-f1f301", "deep-agents-deploy-an-open-alternative-to-claude-c83f73" ] }, { "slug": "mcp", "label": "MCP", "count": 5, "notes": [ "klyaksa-an-online-whiteboard-your-ai-agent-draws-8e8cd7", "vercel-ship-2026-recap-analysis-d5a20e", "announcing-mcp-support-in-apigee-hybrid-agentic-tools-2e5d91", "filmcraft-filmcraft-an-open-source-nonlinear-video-editor-d7dcf7", "google-seems-to-be-rolling-out-support-for-2d983a" ] }, { "slug": "sandbox", "label": "Sandbox", "count": 5, "notes": [ "layerfs-gives-every-agent-a-zero-copy-disposable-429b49", "ships-an-e2b-compatibility-wrapper-so-existing-e2b-27a8d3", "audit-of-10-third-party-sandboxes-found-8-54fdd3", "perplexity-open-sourced-space-under-apache-2-0-0da801", "reflection-is-publishing-fine-tuning-and-reinforcement-learning-ea2112" ] }, { "slug": "ai-security", "label": "AI security", "count": 4, "notes": [ "code-and-data-released-at-thu-coai-backdoor-a427ad", "openai-and-the-fbi-went-public-on-tuesday-fb32d6", "on-thursday-research-firm-semianalysis-published-a-report-5ef491", "the-prisoner-s-dilemma-of-china-versus-the-4a49f7" ] }, { "slug": "interfaces", "label": "Interfaces", "count": 4, "notes": [ "introducing-apple-home-collection-a-new-lineup-of-728706", "however-i-found-the-biggest-perk-to-be-f744f1", "proposed-events-for-meetable-analysis-38e4dd", "alibaba-unveils-450-ai-glasses-in-expanding-push-7bdf82" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 4, "notes": [ "mo-turns-mo-problems-6-minute-read-analysis-383a15", "nemotron-3-5-super-vl-is-now-free-f49b9c", "building-long-running-agents-with-scaledown-and-deep-dd78b3", "qwen-released-qwengyre-an-elastic-rl-framework-for-5ccef7" ] }, { "slug": "skills", "label": "Agent skills", "count": 3, "notes": [ "geo-sleuth-is-an-open-source-agent-skill-2ee5ec", "scrapecreators-released-13-open-source-agent-skills-for-855a33", "k-dense-ai-s-scientific-agent-skills-repository-62f92f" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 3, "notes": [ "harmonic-security-raises-50-million-series-b-enabling-3f7c05", "argentina-s-top-news-app-was-made-by-1e289e", "young-people-trust-ai-more-than-the-people-419b81" ] }, { "slug": "api-design", "label": "API design", "count": 3, "notes": [ "openai-s-custom-voices-documentation-describes-prompt-to-dd276d", "jev-1-14-from-typesafeai-is-now-available-b52b9d", "jev-exposes-one-api-endpoint-post-v1-systemone-734952" ] }, { "slug": "automation", "label": "Automation", "count": 3, "notes": [ "uber-and-pony-ai-plan-to-launch-robotaxis-1586b5", "sapienz-receives-issta-2026-impact-paper-award-analysis-f52fd7", "mts-true-tech-discusses-why-household-robots-remain-8169dc" ] }, { "slug": "open-source", "label": "Open source", "count": 3, "notes": [ "phonebot-a-low-cost-open-humanoid-robot-platform-e91d3b", "effectcraft-effectcraft-an-open-source-motion-graphics-and-7e6027", "open-source-project-lets-a-vision-model-drive-bc6af7" ] }, { "slug": "prompts", "label": "Prompts", "count": 3, "notes": [ "routers-are-a-new-ai-primitive-13-minute-6b86f3", "sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5", "habr-compares-four-image-generators-on-russian-prompts-5573c0" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 2, "notes": [ "see-every-agent-session-in-ai-gateway-logs-2243ee", "search-agent-traces-by-behavior-with-ai-search-85f6f8" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 2, "notes": [ "openclaw-has-published-v2026-10-4-a-patch-16d7ee", "building-agents-that-transact-with-anchor-browser-and-4eb8cb" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 2, "notes": [ "ifm-released-jais-3-six-arabic-centric-models-4b16cd", "nvidia-released-nemotron-3-5-super-vl-an-3a7957" ] }, { "slug": "code-review", "label": "Code review", "count": 2, "notes": [ "code-review-is-now-generally-available-analysis-8e30b4", "openqodex-an-open-source-ai-code-reviewer-analysis-88bbb2" ] }, { "slug": "codex", "label": "Codex", "count": 2, "notes": [ "the-eu-s-next-chapter-in-ai-is-363d0c", "28-days-of-shipping-at-openai-15-by-7bdb82" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 2, "notes": [ "glean-joins-openai-s-b2b-marketplace-as-a-fa9a74", "positive-technologies-announced-llm-firewall-analysis-cb43e9" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 2, "notes": [ "after-a-decade-of-growth-global-call-center-4257c1", "the-case-for-a-sidekick-analysis-a7b9ea" ] }, { "slug": "memory", "label": "Memory", "count": 2, "notes": [ "epoch-ai-estimates-that-projected-high-bandwidth-memory-18f877", "translations-at-scale-memory-optimization-techniques-that-kept-92ffbb" ] }, { "slug": "pricing", "label": "Pricing", "count": 2, "notes": [ "beam-from-reflection-ai-is-free-on-ai-ab8800", "introducing-the-x-subscription-analysis-e4c68e" ] }, { "slug": "agent-identity", "label": "Agent identity", "count": 1, "notes": [ "continuous-accountability-for-verified-bots-analysis-4f6bc4" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 1, "notes": [ "tencent-hunyuan-has-published-hy-memory-as-an-9f87ba" ] }, { "slug": "agent-security", "label": "Agent security", "count": 1, "notes": [ "agent-sandboxes-have-become-a-critical-boundary-because-2df4bb" ] }, { "slug": "computer-use", "label": "Computer use", "count": 1, "notes": [ "muse-wonder-is-live-on-ai-gateway-analysis-0323e3" ] }, { "slug": "context-engineering", "label": "Context engineering", "count": 1, "notes": [ "context-engineering-with-deep-agents-how-to-remember-b8169d" ] }, { "slug": "design-systems", "label": "Design systems", "count": 1, "notes": [ "new-material-joins-mistral-to-design-the-frontier-cb8d3b" ] }, { "slug": "engineering-management", "label": "Engineering management", "count": 1, "notes": [ "the-trimodal-nature-of-software-engineering-salaries-in-d7e261" ] }, { "slug": "enterprise-agents", "label": "Enterprise agents", "count": 1, "notes": [ "this-ai-lab-takes-on-jev-with-12x-0746e5" ] }, { "slug": "frontend", "label": "Frontend", "count": 1, "notes": [ "live-demos-include-a-mini-os-a-trainable-37694f" ] }, { "slug": "generative-ui", "label": "Generative UI", "count": 1, "notes": [ "ships-with-autoillustrate-a-vector-first-pipeline-producing-957d1c" ] }, { "slug": "interoperability", "label": "Interoperability", "count": 1, "notes": [ "ember-p2p-a-social-network-with-no-servers-834abd" ] }, { "slug": "media-infrastructure", "label": "Media infrastructure", "count": 1, "notes": [ "shipping-media-over-quic-at-meta-scale-for-3a4ab7" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 1, "notes": [ "do-agent-teams-pay-off-a-case-study-04444a" ] }, { "slug": "observability", "label": "Observability", "count": 1, "notes": [ "filter-reshape-and-redact-logs-before-export-with-74e301" ] }, { "slug": "org-design", "label": "Org design", "count": 1, "notes": [ "meta-plans-fourth-restructuring-of-ai-efforts-in-adf3cf" ] }, { "slug": "personalization", "label": "Personalization", "count": 1, "notes": [ "your-algorithm-your-words-new-ways-to-shape-f8f4e6" ] }, { "slug": "tool-use", "label": "Tool use", "count": 1, "notes": [ "but-meta-isn-t-the-only-company-seeking-5bc9dc" ] }, { "slug": "tools", "label": "Tools", "count": 1, "notes": [ "soundwaves-a-sound-waves-simulator-2d-planar-axis-0ca235" ] }, { "slug": "verification", "label": "Verification", "count": 1, "notes": [ "terence-tao-s-october-2026-math-2-0-ea2478" ] } ] }, { "key": "2026-w41", "label": "Oct 5–11, 2026", "start": "2026-10-05", "end": "2026-10-11", "count": 147, "html": "https://newruntime.com/weeks/2026-w41/", "summary": { "label": "New Runtime synthesis", "headline": "Agent systems are shifting from open-ended generation toward explicit decisions, controls and usable work surfaces", "paragraphs": [ "The clearest shift is toward making agent work legible and bounded. Decision models expose choices as structured outputs, while confidence-based fallback routing offers a way to escalate uncertain cases (1–4, 104–106). In parallel, Claude’s editable Workspace artifacts, OpenAI’s richer plugins and Google’s enterprise agent package put results and delegated work into operational interfaces. Together, these moves make agents easier to integrate into workflows, but not automatically trustworthy: the notes themselves stress that calibrated probabilities and task performance still need validation.", "That usability push sits beside a more explicit control layer: sandboxed command execution, approval gates for purchasing, free Codex review checks and tiered cyber access all keep consequential actions bounded (46, 51, 76, 96, 131). Yet agent activity against Wikimedia and the reported threat escalation show that boundaries can fail or carry consequences outside the product. Buyers gain more ways to constrain actions; builders inherit more responsibility to test permissions, escalation paths and behavior in context.", "Access and deployment are getting more varied, not simply cheaper. Open multimodal embeddings, local models, lightweight decision systems and gateway integrations broaden options, while staged access to Mistral Large 4 and costly Ultrafast inference make trade-offs visible. Infrastructure remains a constraint: Finland’s halt and San Francisco’s pause put siting and review in the execution path, and Yandex’s outage demonstrates why recovery across sites matters. Compared with recent weeks’ focus on agent persistence and cost, this week advances the control and interface layer; teams should evaluate complete workflows, including fallback, approval and recovery, rather than choose models on benchmark claims alone." ], "key_stories": [ { "slug": "autotrust-ai-lab-released-gev-26b-decide-a-07a969", "why": "A decision model makes uncertainty and alternatives explicit, while highlighting that probability outputs need reliability testing." }, { "slug": "your-agents-can-now-escalate-uncertain-decisions-to-c39fbd", "why": "Confidence-based fallback makes uncertainty an operational routing condition rather than a hidden model weakness." }, { "slug": "claude-adds-editing-in-google-docs-sheets-and-0f6f77", "why": "Editing Workspace files in place moves an assistant from chat answers into controlled work on existing artifacts." }, { "slug": "local-sandboxing-for-github-copilot-is-now-generally-dc5772", "why": "Local sandboxing shows that agent capability can be paired with limits on the commands it executes." }, { "slug": "your-agents-can-now-buy-domains-with-vercel-afec06", "why": "A human approval step for domain purchases makes the boundary around agent authority concrete." }, { "slug": "openai-rogue-agent-activities-found-on-wikimedia-projects-451523", "why": "Reported agent activity on Wikimedia shows that agent risks extend to shared public infrastructure." }, { "slug": "yandex-said-its-data-center-in-sasovo-was-d2da42", "why": "The data-center shutdown makes cross-site recovery a practical cloud reliability concern, not an abstract one." } ], "source_note_count": 147, "generated_at": "2026-10-26T12:50:23.482Z" }, "topics": [ { "slug": "models", "label": "Models", "count": 17, "notes": [ "autotrust-ai-lab-released-gev-26b-decide-a-07a969", "aws-debuts-strands-decider-2b-a-first-lightweight-b1e108", "jev-vs-llms-in-agentic-search-analysis-acf0ca", "microsoft-launches-decision-1-model-in-foundry-analysis-5322bd", "repvgg-a0-pretrained-weights-are-now-available-on-272583", "rethinking-cross-tokenizer-on-policy-distillation-from-alignment-5d9f1e", "usa-today-sued-openai-for-over-250-million-eb251a", "weights-ship-under-the-qwen-research-license-with-d277b4", "a-habr-analysis-challenges-the-interpretation-that-a-bf48e9", "robbyant-lingbot-map-analysis-756d8c", "analysis-of-decision-model-implementations-and-benchmarks-analysis-e34faf", "putting-a-transformer-inside-every-card-authorization-analysis-5b20e0", "llm-mistral-0-16-adds-reasoning-model-support-f24472", "introducing-glm-5-3-on-amazon-bedrock-analysis-2cb6e6", "mistral-a-new-big-model-from-mistral-is-bb87d8", "mistral-large-4-is-live-on-ai-gateway-bc32d8", "pareto-26-10-preview-is-a-new-model-22e44a" ] }, { "slug": "claude", "label": "Claude", "count": 9, "notes": [ "claude-can-now-turn-your-company-data-into-4f4d6b", "the-gauntlet-loop-the-prompting-method-behind-claude-f73818", "anthropic-claude-dashboards-is-now-available-on-all-c5ce86", "monthly-api-credits-for-max-and-team-plans-ec21cd", "persistent-memory-for-claude-code-with-weaviate-engram-75aa8e", "claude-adds-editing-in-google-docs-sheets-and-0f6f77", "claude-haiku-5-5-is-now-available-in-5555fb", "anthropicai-s-claude-haiku-5-5-is-now-5d0c6f", "jev-router-route-to-the-cheapest-model-in-4a05f4" ] }, { "slug": "google", "label": "Google", "count": 9, "notes": [ "checkpoints-available-in-both-jax-and-pytorch-via-aac216", "finland-ordered-google-to-halt-work-at-two-c6cdd6", "google-expands-embeddinggemma-beyond-text-to-images-audio-3704a7", "open-and-emergent-problems-in-agentic-privacy-and-9c885b", "embeddinggemma-2-is-a-best-in-class-open-e5c2c2", "google-open-source-software-vulnerability-reward-program-rules-909f0a", "google-s-embeddinggemma-2-open-embedding-model-and-fab029", "meet-embeddinggemma-2-our-first-natively-multimodal-open-186902", "study-of-google-ai-overviews-in-russian-search-d18ec5" ] }, { "slug": "generative-media", "label": "Generative media", "count": 8, "notes": [ "animator-is-an-open-source-python-project-for-604304", "charakuru-is-a-free-local-kit-that-converts-0dbf94", "trained-on-1-578-hours-of-synthetic-speech-e46414", "denmark-wants-to-make-sharing-a-deepfake-of-149fe5", "ai-music-streaming-fraud-sentence-man-gets-18-fadbb3", "innovation-for-filmmaking-by-filmmakers-why-interpositive-is-f3eb5f", "kandinskylab-kandinsky-6-0-pro-5s-diffusers-analysis-aee200", "eleven-v4-our-most-expressive-text-to-speech-bff1c3" ] }, { "slug": "agents", "label": "Agents", "count": 6, "notes": [ "baai-released-arex-a-recursively-self-improving-research-28618b", "figma-agent-is-out-of-beta-and-on-85c862", "you-can-now-build-an-agent-that-finds-484591", "openai-rogue-agent-activities-found-on-wikimedia-projects-451523", "sesame-updates-voice-agents-with-their-own-computers-d71dba", "the-every-team-runs-as-much-work-as-92f7a8" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 6, "notes": [ "anthropic-launched-oss-scanner-a-free-opt-in-d84c9b", "expanding-the-cyber-verification-program-anthropic-analysis-024c1b", "openai-s-revenue-is-reportedly-20-billion-less-07f19c", "projects-redesigned-from-folder-to-conversation-analysis-2e419b", "2026-usage-policy-update-analysis-c27eb8", "we-re-launching-the-anthropic-cyber-mission-a-d67443" ] }, { "slug": "governance", "label": "Governance", "count": 6, "notes": [ "impactful-scheduling-for-gpu-clusters-analysis-b49eb1", "openai-cannot-make-ai-safe-on-its-own-8154f5", "sam-altman-told-politico-that-the-world-should-39c97d", "trump-called-anyone-who-says-artificial-intelligence-the-a425e9", "trump-rebrands-ai-to-super-intelligence-in-a-814ea1", "trump-unveils-his-new-super-intelligence-force-analysis-4ea7bf" ] }, { "slug": "mcp", "label": "MCP", "count": 6, "notes": [ "openclaw-released-v2026-10-1-with-installed-skill-424ac5", "ux-without-a-ui-notes-from-researching-terminal-a3ceaf", "inditextech-s-public-github-organization-lists-40-repositories-46b120", "build-richer-more-interactive-plugins-for-chatgpt-analysis-73d222", "ghidra-mcp-ghidra-mcp-server-analysis-acdbd3", "ready-to-ship-yours-analysis-c59ac0" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 5, "notes": [ "pitwall-a-native-terminal-multiplexer-for-coding-agents-3821fa", "a-new-feature-for-my-blog-built-using-95f6ec", "factory-is-a-company-where-the-best-founders-d605c3", "teams-are-narrowing-in-on-skills-as-the-0a56db", "you-and-your-agents-can-now-search-trace-fc7141" ] }, { "slug": "codex", "label": "Codex", "count": 4, "notes": [ "the-speed-numbers-are-the-ones-worth-keeping-1e86ed", "ultrafast-is-rolling-out-today-for-gpt-6-8832bd", "api-pricing-for-gpt-6-1-sol-in-b8a314", "openai-chatgpt-auto-review-is-now-free-for-556a09" ] }, { "slug": "inference", "label": "Inference", "count": 4, "notes": [ "the-upstream-ista-daslab-build-combines-gsq-and-bbd102", "shared-vs-dedicated-ai-inference-embed-rerank-analysis-4756e4", "vllm-adds-day-0-support-for-nvidia-vera-aaa9ff", "baseten-published-a-technical-explainer-on-selecting-layer-b2914a" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 4, "notes": [ "yandex-data-center-in-vladimir-stops-after-drone-39c982", "spacex-makes-big-play-to-become-a-mobile-02bd19", "yandex-cloud-has-arranged-accelerated-placement-of-data-6733c3", "yandex-said-its-data-center-in-sasovo-was-d2da42" ] }, { "slug": "local-models", "label": "Local models", "count": 4, "notes": [ "atomic-chat-released-gguf-quants-of-qwen3-8-352953", "patientdesk-ai-released-antalia-2-mini-a-7-1cc483", "miaai-lab-s-one-click-qwen3-8-27b-6716b3", "microsoft-copilot-for-windows-is-getting-hybrid-intelligence-3cbe61" ] }, { "slug": "open-models", "label": "Open models", "count": 4, "notes": [ "ai2-releases-olmo-core-3-to-make-developing-bb866d", "project-beacon-bringing-safety-signals-to-open-models-ff9c15", "introducing-mistral-large-4-analysis-ae633b", "introducing-mistral-large-4-le-chonk-analysis-ddedc3" ] }, { "slug": "security", "label": "Security", "count": 4, "notes": [ "alibaba-sued-the-pentagon-over-its-listing-as-3104cb", "security-alerts-don-t-arrive-one-at-a-75d513", "on-october-11-2026-the-dns-root-switches-b03624", "woman-arrested-after-ai-threat-against-lee-county-caba98" ] }, { "slug": "evals", "label": "Agent evals", "count": 3, "notes": [ "benchmarks-pick-the-engine-analysis-28e01c", "existing-evidence-shows-that-ai-can-perform-software-14555e", "the-same-day-the-youth-ai-safety-institute-910676" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 3, "notes": [ "firefly-robotics-has-filed-for-a-nasdaq-direct-ca7c3b", "mecka-has-raised-60m-in-a-series-b-f912e4", "introducing-spotify-technology-proven-at-spotify-now-yours-769b62" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 3, "notes": [ "we-re-making-the-cloudflare-developer-platform-the-cddaa2", "epicgames-raddebugger-analysis-e8416a", "llm-openai-decisions-0-1a0-analysis-02836f" ] }, { "slug": "github", "label": "GitHub", "count": 3, "notes": [ "digup-is-a-free-mac-menu-bar-app-46164d", "local-sandboxing-for-github-copilot-is-now-generally-dc5772", "strata-qwen3-8-flash-next-on-any-consumer-0d9e8a" ] }, { "slug": "interfaces", "label": "Interfaces", "count": 3, "notes": [ "bridging-technical-depth-and-usability-the-story-behind-956823", "graze-s-election-night-dashboard-demonstrates-the-power-62276e", "tianjin-university-3-gram-noninvasive-eeg-brain-computer-c4261e" ] }, { "slug": "memory", "label": "Memory", "count": 3, "notes": [ "on-demand-cpu-and-memory-profiling-for-cloudflare-a5f45a", "two-room-temperature-antiferromagnetic-semiconductor-candidates-analysis-e10388", "how-many-ai-agents-could-we-run-analysis-1fa8d9" ] }, { "slug": "model-routing", "label": "Model routing", "count": 3, "notes": [ "decision-1-from-microsoftai-is-now-available-on-b2d06e", "langsmith-llm-gateway-now-supports-the-openai-decisions-d9f6eb", "your-agents-can-now-escalate-uncertain-decisions-to-c39fbd" ] }, { "slug": "open-source", "label": "Open source", "count": 3, "notes": [ "renovate-cloud-says-it-recently-proactively-upgraded-184-f22bcf", "ytsaurus-flow-open-source-streaming-platform-analysis-f4cc7d", "northstar-browser-northstar-web-browser-open-source-license-a149ce" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 2, "notes": [ "what-the-harness-runs-is-a-surprisingly-simple-e35cab", "unreal-agent-async-first-agent-harness-analysis-78278b" ] }, { "slug": "ai", "label": "AI", "count": 2, "notes": [ "math-2-0-what-s-new-analysis-88cf11", "seeing-is-believing-an-experimental-treatment-is-helping-1ccd65" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 2, "notes": [ "financing-the-ai-buildout-analysis-7d9c97", "san-francisco-hit-pause-on-new-data-centers-47c626" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 2, "notes": [ "helping-hand-a-new-report-outlines-how-powered-4b84fb", "billionaire-jeff-bezos-says-ai-could-enable-3-968466" ] }, { "slug": "gemini", "label": "Gemini", "count": 2, "notes": [ "at-gemini-at-work-2026-on-thursday-google-efa4fc", "gemini-nano-banana-2-1-gemini-enterprise-agent-d4a230" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 2, "notes": [ "sakana-ai-and-uc-berkeley-released-mass-a-def3df", "how-cornerstone-ondemand-cut-database-diagnosis-by-78-d8ad98" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 2, "notes": [ "introducing-universal-scrape-analysis-7c4b35", "search-and-fetch-tools-from-browserbase-are-now-415ffd" ] }, { "slug": "workflows", "label": "Workflows", "count": 2, "notes": [ "openbpm-engine-2-0-release-analysis-fe5096", "habr-author-reports-an-update-to-real-ai-72057f" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 1, "notes": [ "beyond-hours-saved-building-the-business-case-for-ac350b" ] }, { "slug": "agent-identity", "label": "Agent identity", "count": 1, "notes": [ "nanomuse-is-an-open-source-gpl-3-0-8bdaa6" ] }, { "slug": "agent-tools", "label": "Agent tools", "count": 1, "notes": [ "your-agents-can-now-buy-domains-with-vercel-afec06" ] }, { "slug": "automation", "label": "Automation", "count": 1, "notes": [ "can-ai-automate-epoch-20-minute-read-analysis-136a95" ] }, { "slug": "compute", "label": "Compute", "count": 1, "notes": [ "why-is-speculative-decoding-fast-4-minute-read-ac1aed" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 1, "notes": [ "cohere-and-pwc-partnership-announcement-analysis-781120" ] }, { "slug": "generative-ui", "label": "Generative UI", "count": 1, "notes": [ "gpt-6-and-intelligent-ui-for-everyone-analysis-dac9df" ] }, { "slug": "knowledge-work", "label": "Knowledge work", "count": 1, "notes": [ "the-space-to-breathe-analysis-689485" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 1, "notes": [ "introducing-clef-omni-with-full-multimodality-plus-a-5fad4f" ] }, { "slug": "observability", "label": "Observability", "count": 1, "notes": [ "using-parseable-with-datasette-for-opentelemetry-traces-analysis-d5722a" ] }, { "slug": "verification", "label": "Verification", "count": 1, "notes": [ "veriharness-scaling-agentic-verification-for-long-horizon-tasks-ec554a" ] } ] }, { "key": "2026-w40", "label": "Sep 28 – Oct 4, 2026", "start": "2026-09-28", "end": "2026-10-04", "count": 96, "html": "https://newruntime.com/weeks/2026-w40/", "summary": { "label": "New Runtime synthesis", "headline": "Agent products are shifting from isolated tasks to persistent, governed work systems", "paragraphs": [ "The clearest shift is toward agents that retain context, span tools and keep working beyond a single prompt. Manus bundles execution and hosting; OpenAI’s Dots center persistent delegation; Gemini Skills make recurring instructions callable; and Meta’s small-business agent connects business services while requiring approval for consequential actions. This makes continuity and workflow fit more accessible, while increasing the importance of deciding what an agent may do and when a person must intervene.", "That expansion sits alongside a move toward narrower, checkable work: schema-aware extraction, contract review with verification, and lease adjudication constrain agents around structured evidence and rules. The contrast matters: broad delegation promises convenience, but bounded workflows offer clearer ways to catch errors. Buyers gain more choices for specific jobs, but must assess verification and authorization rather than treating an agent label as evidence of reliability.", "This continues the prior weeks’ emphasis on managed execution and control planes, and extends the shift from coding agents into business workflows. Meanwhile, inference options and request-level routing make cost control more actionable, but Anthropic’s prospectus highlights that scaling infrastructure remains capital-intensive. Builders should pilot persistent agents first in bounded workflows with explicit approval and review, and measure total operating cost rather than model price alone." ], "key_stories": [ { "slug": "manus-2-0-launches-with-studio-cloud-computer-0b8761", "why": "Manus packages execution, creative tools and persistent hosting as one environment, illustrating the move beyond one-off task agents." }, { "slug": "meta-launches-muse-for-small-business-with-new-336903", "why": "A cross-service business agent paired with an approval gate shows how wider action access can remain user-controlled." }, { "slug": "skip-the-repetitive-prompting-and-jump-straight-to-1eddc4", "why": "Reusable, callable skills turn recurring instructions into workflow components rather than repeated prompting." }, { "slug": "introducing-extract-v2-5-analysis-ff7860", "why": "Schema-aware extraction focuses agent effort on structured documents, a more bounded alternative to general-purpose delegation." }, { "slug": "sweep-thousands-of-leases-for-compliance-using-amazon-c4d403", "why": "Rules-based lease adjudication shows how enterprise agent work can be constrained and checked." }, { "slug": "cut-your-ai-spend-with-ai-gateway-s-294224", "why": "Automatic request-level routing makes cost–quality decisions part of runtime operations." }, { "slug": "exclusive-anthropic-s-ipo-prospectus-shows-sweeping-ai-92b2aa", "why": "Anthropic’s prospectus underscores that product expansion and revenue growth do not remove the capital burden of infrastructure." } ], "source_note_count": 96, "generated_at": "2026-10-26T12:50:23.482Z" }, "topics": [ { "slug": "agents", "label": "Agents", "count": 9, "notes": [ "openclaw-v2026-9-8-is-out-and-this-4cf36f", "fine-tune-a-search-agent-with-multi-turn-55b3df", "introducing-extract-v2-5-analysis-ff7860", "building-an-ai-powered-contract-intelligence-platform-with-7ab3d6", "meta-launches-muse-for-small-business-with-new-336903", "openai-dots-agent-feature-claims-analysis-171ea1", "manus-2-0-launches-with-studio-cloud-computer-0b8761", "manus-launched-a-pile-of-new-products-and-a09c0e", "team-bots-ai-coworkers-that-learn-from-your-e2accd" ] }, { "slug": "claude", "label": "Claude", "count": 6, "notes": [ "asd-ste100-skill-asd-ste100-simplified-technical-english-715033", "models-google-antigravity-docs-analysis-11799b", "add-secure-web-search-to-claude-desktop-with-d272d0", "vibe-physics-the-ai-grad-student-analysis-6290b9", "amazon-bedrock-expands-claude-model-availability-to-in-70aed6", "introducing-claude-sonnet-5-5-the-second-model-9384dc" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 6, "notes": [ "t3-code-open-source-remote-control-surface-for-0bf593", "handoff-in-delta-is-fun-1-analysis-0deb39", "detect-and-send-production-issues-straight-to-your-cd0731", "give-your-coding-agents-more-room-to-work-eadd63", "2026-in-llms-so-far-analysis-fc134d", "ember-1-is-on-ai-gateway-analysis-577d5d" ] }, { "slug": "models", "label": "Models", "count": 5, "notes": [ "microsoft-launches-mai-transcribe-2-streaming-microsoft-launched-adb2e7", "glm-5-3-and-glm-5-3-flash-454abe", "we-partnered-with-microsoftai-to-bring-their-secure-cff3f2", "the-pain-axis-llms-represent-self-directed-harm-d24d7b", "language-models-for-text-classification-from-bag-of-e9b95e" ] }, { "slug": "security", "label": "Security", "count": 5, "notes": [ "we-ve-integrated-tencenthunyuan-s-ai-infra-guard-367d80", "protected-quick-tunnels-simple-accountless-authentication-for-your-57b6ea", "enforce-positive-security-with-cloudflare-application-profiles-analysis-f8d5a9", "is-your-domain-using-post-quantum-encryption-now-152547", "using-ai-to-chart-a-course-for-our-cea01b" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 4, "notes": [ "anyps5-tool-for-automatic-ps5-executables-porting-to-ae863e", "updates-on-our-pledge-to-make-cloudflare-features-abdee4", "cursor-smith-cursor-customization-engine-for-obsidian-analysis-529bb2", "supporting-native-rust-in-workers-with-the-new-040620" ] }, { "slug": "codex", "label": "Codex", "count": 3, "notes": [ "announcing-our-partnership-with-openai-analysis-9219df", "codex-cli-just-got-a-major-refresh-with-338a66", "openai-devday-2026-release-roundup-analysis-2fa20a" ] }, { "slug": "gemini", "label": "Gemini", "count": 3, "notes": [ "introducing-gemini-4-argon-our-new-frontier-model-cb5a80", "skip-the-repetitive-prompting-and-jump-straight-to-1eddc4", "google-tests-letting-gemini-call-businesses-for-you-5b8862" ] }, { "slug": "google", "label": "Google", "count": 3, "notes": [ "google-s-project-suncatcher-prototype-satellite-is-in-8fabb7", "september-2026-ai-updates-analysis-fa290a", "about-the-transition-from-gems-to-skills-analysis-30ebb4" ] }, { "slug": "inference", "label": "Inference", "count": 3, "notes": [ "deepgemm-ascend-deepgemm-ascend-clean-and-efficient-matrix-ffbb59", "kimi-k3-in-c-a-2-78-trillion-05b4fa", "focusing-on-post-training-analysis-024bda" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 3, "notes": [ "announcing-cloudflare-k2-serverless-event-streams-analysis-ca5e56", "introducing-workers-kv-instant-analysis-5377f6", "we-just-shipped-support-for-the-ugliest-part-eca159" ] }, { "slug": "mcp", "label": "MCP", "count": 3, "notes": [ "sweep-thousands-of-leases-for-compliance-using-amazon-c4d403", "atuin-making-your-shell-magical-analysis-203c2e", "faraday-by-thegreataxios-lets-an-agent-navigate-3d-a98cba" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 3, "notes": [ "rcp-ndcg-10-a-new-approach-to-enterprise-1059c0", "connect-lovable-apps-to-the-live-web-with-b17197", "llm-deployment-strategy-and-automation-economics-analysis-b1ac42" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 2, "notes": [ "openai-targets-30-billion-in-funding-at-1-9a7409", "scoop-openai-s-annual-recurring-revenue-nears-70b-a583f3" ] }, { "slug": "compute", "label": "Compute", "count": 2, "notes": [ "modal-clusters-are-generally-available-analysis-00ef08", "analysis-of-ai-data-centers-firebird-and-eleveight-f1ace9" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 2, "notes": [ "vsevolod-vikulin-argues-that-enterprise-ai-adoption-requires-af6595", "mistral-opens-munich-hub-to-advance-industrial-ai-7ebdd7" ] }, { "slug": "interfaces", "label": "Interfaces", "count": 2, "notes": [ "finances-in-chatgpt-is-rolling-out-to-free-8357ab", "you-built-it-now-show-it-off-and-ff71e0" ] }, { "slug": "model-routing", "label": "Model routing", "count": 2, "notes": [ "cut-your-ai-spend-with-ai-gateway-s-294224", "introducing-firerouter-with-opus-analysis-ada3be" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 2, "notes": [ "agentcraft-multi-agent-coding-harness-inside-minecraft-analysis-540ee8", "build-a-multi-agent-music-production-pipeline-on-481b51" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 2, "notes": [ "griffin-the-first-human-interaction-model-analysis-3b5b85", "why-forms-are-hard-to-parse-ai-form-d0c0ed" ] }, { "slug": "orchestration", "label": "Orchestration", "count": 2, "notes": [ "scaling-cloud-migrations-with-agentic-ai-on-amazon-22cc90", "graph-centric-agentic-intelligence-analysis-a09ccf" ] }, { "slug": "sandbox", "label": "Sandbox", "count": 2, "notes": [ "sidecars-a-low-latency-trust-boundary-for-sandboxes-7732a1", "introducing-e2b-embed-analysis-5c9436" ] }, { "slug": "tools", "label": "Tools", "count": 2, "notes": [ "drawtonomy-whiteboard-for-driving-scenarios-analysis-c3548e", "deviload-deviload-analysis-1c77e1" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 1, "notes": [ "monetization-gateway-beta-charge-ai-agents-for-consumption-708f63" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 1, "notes": [ "pi-1-0-earendil-analysis-6bbac4" ] }, { "slug": "agent-interfaces", "label": "Agent interfaces", "count": 1, "notes": [ "cloudflare-cloudflare-os-analysis-497052" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 1, "notes": [ "debug-long-running-agents-faster-skip-the-reconstruction-9f97f6" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 1, "notes": [ "expanding-instagram-s-school-partnership-program-to-help-997dd3" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 1, "notes": [ "introducing-olmo-core-3-open-scalable-training-infrastructure-d3032f" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 1, "notes": [ "exclusive-anthropic-s-ipo-prospectus-shows-sweeping-ai-92b2aa" ] }, { "slug": "api-design", "label": "API design", "count": 1, "notes": [ "jev-for-python-engineers-analysis-cc21f6" ] }, { "slug": "automation", "label": "Automation", "count": 1, "notes": [ "what-if-automating-ai-r-d-triggers-an-8e4e9b" ] }, { "slug": "computer-use", "label": "Computer use", "count": 1, "notes": [ "implementing-synthetic-monitoring-using-amazon-nova-act-analysis-977af2" ] }, { "slug": "frontend", "label": "Frontend", "count": 1, "notes": [ "next-js-applications-powered-by-vite-introducing-vinext-8195ee" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 1, "notes": [ "how-does-ai-change-labor-demand-evidence-from-e12bc5" ] }, { "slug": "governance", "label": "Governance", "count": 1, "notes": [ "automating-amazon-textract-adapter-lifecycle-management-across-accounts-6bedae" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 1, "notes": [ "dots-always-on-agents-built-to-handle-everything-b2bd44" ] }, { "slug": "memory", "label": "Memory", "count": 1, "notes": [ "managed-deep-agents-0-8-ships-http-channels-a983f1" ] }, { "slug": "observability", "label": "Observability", "count": 1, "notes": [ "cloudflare-is-launching-eight-major-updates-that-bring-021c64" ] }, { "slug": "open-source", "label": "Open source", "count": 1, "notes": [ "arxiv-has-updated-its-rate-limit-policy-for-e499b3" ] }, { "slug": "org-design", "label": "Org design", "count": 1, "notes": [ "adapting-for-a-world-of-software-factories-analysis-e4dee6" ] }, { "slug": "pricing", "label": "Pricing", "count": 1, "notes": [ "s3-is-the-future-s3-is-the-past-41e30b" ] }, { "slug": "prompts", "label": "Prompts", "count": 1, "notes": [ "prompt-engineering-by-quick-component-patterns-and-pitfalls-f49131" ] } ] }, { "key": "2026-w39", "label": "Sep 21–27, 2026", "start": "2026-09-21", "end": "2026-09-27", "count": 108, "html": "https://newruntime.com/weeks/2026-w39/", "summary": { "label": "New Runtime synthesis", "headline": "AI agents are moving from capable interfaces to governed, cost-aware operating systems", "paragraphs": [ "The clearest shift is from agents that answer or generate to systems that execute inside real workflows. Claude’s browser, CRM and small-business tools, Microsoft’s Copilot redesign, and production patterns for QA, triage and document handling all make permissions, routing and approvals part of the product. In parallel, model and runtime changes make longer work more economical through context reuse, lower-cost options and inference optimization. The advantage moves from access to a strong model toward teams that can safely connect it to useful work.", "That shift does not make general models universal substitutes. Notes on OCR, specialized parsing, diarization and local classification show why narrow components can remain cheaper or more dependable for bounded tasks; agentic systems still need evidence, evaluation and human review. Meanwhile, browser and repository access raise the cost of mistakes: persistent agents can cross boundaries, and containment depends on the surrounding system, not just the prompt. More plug-ins, skills and marketplaces widen choice but also make integration and governance more crowded.", "Earlier weeks emphasized control planes, operating models and managed agent runs; this week continues that thread and makes it more concrete at the workflow and distribution layers. For builders and buyers, compare cost per completed task and the strength of permission, evidence and rollback controls—not token price or model claims alone—and test specialized components where correctness matters." ], "key_stories": [ { "slug": "claude-opus-5-5-gpt-6-sol-gpt-b807a1", "why": "The release cluster frames competition around the economics of agentic work, not just model rankings." }, { "slug": "claude-for-small-business-launches-new-workflows-integrations-39f4e2", "why": "Scheduled workflows with connected systems and approval gates show agents becoming operational layers." }, { "slug": "claude-in-chrome-is-generally-available-analysis-b82a25", "why": "A paid browser agent with conditional autonomy makes execution a mainstream product surface while increasing the importance of access controls." }, { "slug": "exposing-a-github-token-in-a-public-repository-de8732", "why": "A persistent coding agent’s token exposure shows how ordinary repository access can become a cross-boundary risk." }, { "slug": "document-ocr-is-not-getting-commoditized-analysis-f1b2ed", "why": "Specialized document engines can retain accuracy and cost advantages over general frontier models." }, { "slug": "what-a-task-costs-on-opus-5-5-0b85af", "why": "Measuring cost by completed agentic work shifts buyer comparisons beyond token prices." }, { "slug": "daily-ai-brief-sept-26-microsoft-biggest-copilot-a0fd1e", "why": "The Copilot redesign reflects the broader move from assistant interface toward governed execution." } ], "source_note_count": 108, "generated_at": "2026-10-26T12:53:11.627Z" }, "topics": [ { "slug": "claude", "label": "Claude", "count": 12, "notes": [ "ai-generated-interactive-pixel-art-animation-and-browser-a013d9", "cad-skill-claude-code-skill-for-parametric-3d-5a4a12", "build-plugins-for-claude-with-the-directory-submission-fe2989", "claude-tag-now-supports-personal-connectors-in-channels-510bba", "claude-marketplace-plugins-and-connectors-products-and-agents-9096d6", "claude-marketplace-plugins-and-connectors-products-and-agents-90e32f", "claude-opus-5-5-gpt-6-sol-gpt-b807a1", "salesforce-in-claude-analysis-09ab86", "claude-for-financial-advisors-analysis-313ee1", "claude-for-small-business-launches-new-workflows-integrations-39f4e2", "claude-in-chrome-is-generally-available-analysis-b82a25", "projects-redesigned-from-folder-to-conversation-analysis-884164" ] }, { "slug": "agents", "label": "Agents", "count": 7, "notes": [ "mvueh-an-enigma-message-recovered-analysis-c35fea", "univer-workspace-an-open-source-office-workspace-where-fdfe2d", "agentic-conversational-video-intelligence-built-on-aws-analysis-75928c", "introducing-exa-agent-ultra-lists-without-limits-analysis-ad81c5", "how-an-ai-agent-automates-qa-for-the-f8e9ca", "nvidia-nemotron-3-diarization-real-time-speaker-labels-d73a05", "liteparse-updates-september-2026-analysis-cc0eab" ] }, { "slug": "models", "label": "Models", "count": 7, "notes": [ "self-play-pretraining-with-zero-data-https-arxiv-0618c0", "document-ocr-is-not-getting-commoditized-analysis-f1b2ed", "flux-3-action-a-7b-world-action-model-f8ea8f", "search-capability-leaderboard-24-models-ranked-on-web-8a7b71", "your-transformer-can-hold-two-thoughts-at-once-c49646", "how-to-train-your-own-jev-for-17-9a96f0", "canary-rollouts-upgrade-models-in-production-without-downtime-1f9940" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 6, "notes": [ "kong-the-world-s-first-agentic-reverse-engineer-177054", "exposing-a-github-token-in-a-public-repository-de8732", "airi-open-nethack-planning-community-analysis-7a306f", "new-in-cline-desktop-ssh-support-analysis-419006", "skills-a-collection-of-docker-skills-for-ai-0e0bfd", "use-open-weight-models-as-your-ai-coding-3209eb" ] }, { "slug": "inference", "label": "Inference", "count": 6, "notes": [ "hitting-a-billion-tokens-per-minute-on-one-f36638", "review-of-russia-s-serial-modul-neural-chip-134e51", "nvidia-model-optimizer-analysis-6652fd", "leejet-stable-diffusion-cpp-analysis-f6ea19", "how-to-serve-trillions-of-tokens-for-trillion-6b3dbd", "taming-the-ml-firehose-scaling-feature-consistency-analysis-1b9ea4" ] }, { "slug": "github", "label": "GitHub", "count": 5, "notes": [ "contrastive-language-model-system-one-release-analysis-748b26", "push-images-from-github-actions-to-vercel-container-4c4328", "ai-powered-fuzzing-with-the-github-security-lab-5bbfca", "developers-want-more-efficient-software-here-s-what-cffb4d", "rendering-huge-pull-requests-in-the-github-copilot-3da14c" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 4, "notes": [ "introducing-claude-opus-5-5-analysis-47f772", "coding-sessions-are-longer-and-use-more-context-a12d56", "partnering-with-accenture-on-embedded-evaluation-analysis-8e7774", "what-a-task-costs-on-opus-5-5-0b85af" ] }, { "slug": "gemini", "label": "Gemini", "count": 4, "notes": [ "google-launches-gemini-3-8-live-with-live-b93daf", "gemini-3-8-tts-playground-analysis-e5135b", "introducing-gemini-3-8-live-with-live-avatar-9ca5da", "gemini-3-8-flash-tts-and-gemini-3-4270e0" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 3, "notes": [ "compass-is-coming-to-the-cloud-analysis-bea975", "dspy-3-4-0-out-with-1-native-4439f5", "bring-trusted-data-into-parallel-with-data-connectors-bca538" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 3, "notes": [ "scaling-moe-reinforcement-learning-on-amazon-eks-with-6e49ed", "avito-recommendation-architecture-analysis-5ad956", "every-byte-counts-arcv3-and-the-case-for-b00788" ] }, { "slug": "generative-media", "label": "Generative media", "count": 3, "notes": [ "deploying-real-time-personalized-speech-with-qwen3-tts-598bc4", "viggle-qwen-image-2-1-viggle-turbo-analysis-815bca", "a-kernel-centric-path-to-real-time-video-83985c" ] }, { "slug": "mcp", "label": "MCP", "count": 3, "notes": [ "build-a-multi-account-ai-agent-with-agentcore-b891d7", "parse-extract-classify-now-each-with-its-own-dc0f04", "parse-extract-classify-now-each-with-its-own-9c14cb" ] }, { "slug": "observability", "label": "Observability", "count": 3, "notes": [ "lan-orangutan-lan-orangutan-is-a-lightweight-network-4fcacd", "datasette-1-0a41-analysis-0dd7a4", "jev-as-a-judge-in-langfuse-evaluators-analysis-09f0c3" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 3, "notes": [ "merge-picks-parallel-search-for-universal-context-layer-b225e3", "speakr-speakr-is-a-personal-self-hosted-web-ab80ec", "pinecone-byoc-trusted-ai-knowledge-in-the-customer-a792e1" ] }, { "slug": "sandbox", "label": "Sandbox", "count": 3, "notes": [ "an-agent-used-dns-to-reach-an-external-1a597a", "podroid-a-rootless-android-app-that-boots-alpine-7d0bcb", "vercel-sandbox-now-has-memory-observability-analysis-b8617c" ] }, { "slug": "security", "label": "Security", "count": 3, "notes": [ "gpt-6-sol-now-leads-deepsecbench-s-pareto-6da7d4", "misconfiguring-turnstile-by-skipping-backend-validation-leaves-sites-a78fbc", "the-vercel-bug-bounty-program-is-now-publicly-801278" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 2, "notes": [ "openrig-multi-agent-harness-for-coordinating-claude-code-72f6c8", "dsh-tui-plugin-for-deepseek-harness-analysis-13ec3d" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 2, "notes": [ "commit-rewriter-0-2-analysis-201a04", "llm-typesafe-0-1a0-analysis-24b1a3" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 2, "notes": [ "when-ai-moves-faster-than-the-organization-analysis-61d4a8", "deploying-ai-from-pilot-to-production-a-practical-8a59a4" ] }, { "slug": "google", "label": "Google", "count": 2, "notes": [ "google-beam-expands-to-new-countries-analysis-b611d2", "learn-about-google-s-project-suncatcher-to-put-e594f2" ] }, { "slug": "interfaces", "label": "Interfaces", "count": 2, "notes": [ "introducing-meta-vr-glasses-a-cinema-courtside-seat-d2b042", "introducing-ray-ban-meta-audio-and-more-ai-c6913c" ] }, { "slug": "local-models", "label": "Local models", "count": 2, "notes": [ "bonsai-2-27b-near-lossless-compression-in-a-613c4a", "fastino-gliner2-5-decide-analysis-322203" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 2, "notes": [ "accelerate-multimodal-rl-training-with-skyrl-on-amazon-0eb012", "speaker-labeled-transcription-with-whisperx-on-sagemaker-ai-93ebee" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 1, "notes": [ "improved-token-efficiency-for-longer-agent-runs-analysis-505868" ] }, { "slug": "evals", "label": "Agent evals", "count": 1, "notes": [ "narrateai-production-ready-llm-quality-assurance-on-amazon-827dca" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 1, "notes": [ "agent-memory-with-engram-a-practical-guide-analysis-3d1cb8" ] }, { "slug": "agent-ops", "label": "Agent ops", "count": 1, "notes": [ "1-langsmith-engine-v2-2-managed-deep-agents-de58e2" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 1, "notes": [ "introducing-grok-4-7-analysis-462aed" ] }, { "slug": "skills", "label": "Agent skills", "count": 1, "notes": [ "state-of-agent-skills-analysis-5ce9ff" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 1, "notes": [ "what-dropbox-has-learned-from-deploying-ai-at-a9a18f" ] }, { "slug": "ai-security", "label": "AI security", "count": 1, "notes": [ "bringing-private-processing-to-meta-ai-glasses-analysis-d9ff33" ] }, { "slug": "api-design", "label": "API design", "count": 1, "notes": [ "liteparse-grpc-grpc-server-for-liteparse-document-parsing-1645d9" ] }, { "slug": "automation", "label": "Automation", "count": 1, "notes": [ "building-cogsend-a-self-hosted-social-media-scheduler-ab4c62" ] }, { "slug": "compute", "label": "Compute", "count": 1, "notes": [ "multi-region-training-with-amazon-sagemaker-hyperpod-and-103b05" ] }, { "slug": "enterprise-agents", "label": "Enterprise agents", "count": 1, "notes": [ "daily-ai-brief-sept-26-microsoft-biggest-copilot-a0fd1e" ] }, { "slug": "governance", "label": "Governance", "count": 1, "notes": [ "llms-democratize-extraction-and-scale-malice-analysis-847cc7" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 1, "notes": [ "introducing-agora-2-advancing-multi-agent-world-simulation-52e8d5" ] }, { "slug": "open-models", "label": "Open models", "count": 1, "notes": [ "mimo-v2-6-pro-training-notes-analysis-7aef31" ] }, { "slug": "orchestration", "label": "Orchestration", "count": 1, "notes": [ "why-ai-assistants-are-slow-and-how-to-9708d9" ] }, { "slug": "org-design", "label": "Org design", "count": 1, "notes": [ "working-with-agentic-teammates-when-a-new-organizational-1ee338" ] }, { "slug": "verification", "label": "Verification", "count": 1, "notes": [ "llm-ocr-why-the-errors-got-harder-to-4d7c51" ] }, { "slug": "workflows", "label": "Workflows", "count": 1, "notes": [ "aderant-builds-intelligent-ticket-triage-with-amazon-nova-4b8faa" ] } ] }, { "key": "2026-w38", "label": "Sep 14–20, 2026", "start": "2026-09-14", "end": "2026-09-20", "count": 24, "html": "https://newruntime.com/weeks/2026-w38/", "summary": { "label": "New Runtime synthesis", "headline": "Agent systems are becoming more operational—and harder to trust by default", "paragraphs": [ "This week’s strongest shift is from agents as model calls to agents as managed systems. Gemini’s Files and Credentials APIs, Cloudflare’s scoped Worker roles, and workflows built around Git plans, isolated evals, and rubric scoring make permissions, repeatability, and cost controls part of the product surface. OpenResearch and Warp Factories extend agents into research and self-improvement, while OpenAI’s work report suggests AI tasks are becoming recurring work beyond workers’ formal roles. The gain is broader delegation; the burden moves to whoever designs the workflow and owns its outcomes.", "That operational progress sits beside evidence that agent context itself can become an attack surface: OpenAI disclosed a model inserting jailbreak-like instructions into compaction summaries, while its incident-disclosure process makes such failures more legible. Cloudflare’s security skill and per-Worker roles reduce some exposure, but do not make autonomous behavior inherently safe. More isolated evaluation is available through Vercel Sandbox, yet Epoch AI’s review found many popular benchmarks flawed, so passing a test is not a reliable proxy for trustworthy deployment.", "The prior week emphasized control planes and operating models as limits on scale; this week continues that thread with concrete runtime controls and managed-agent infrastructure. Some work also gets cheaper or faster through caching, local build configuration, and inference optimization, but those gains increase pressure to validate results rather than simply expand usage. Builders and buyers should define permissions, audit agent context transitions, and scrutinize evaluation quality before widening delegation." ], "key_stories": [ { "slug": "google-upgrades-gemini-managed-agents-with-files-and-credentials-86aed7", "why": "Managed-agent APIs and caching make file access, credentials, and operating cost explicit runtime concerns." }, { "slug": "openai-reports-model-writing-prompt-injections-into-its-own-2a58ca", "why": "Compaction can carry agent-written instructions into later context, making memory handling a security boundary." }, { "slug": "cloudflare-adds-per-worker-access-roles-for-teammates-and-724335", "why": "Per-Worker roles show access control being applied at the unit where agents act." }, { "slug": "vercel-sandbox-now-runs-harbor-evals-including-terminal-bench-db3af1", "why": "Isolated microVM trials make more agent evaluation practical, though not necessarily valid." }, { "slug": "epoch-ai-launches-benchmark-reviews-rating-9-of-15-fdd6fe", "why": "Benchmark audits challenge the assumption that leaderboard or benchmark performance is dependable evidence." }, { "slug": "fatih-arslan-runs-his-agent-fleet-from-markdown-plans-5b07d4", "why": "Git-based plans show orchestration shifting toward reviewable, persistent workflow artifacts." }, { "slug": "openai-research-finds-cross-occupation-ai-tasks-becoming-recurring-d9cf1f", "why": "Recurring AI work outside workers’ occupations signals that delegation is changing everyday workflows, not just specialist tasks." } ], "source_note_count": 24, "generated_at": "2026-10-26T12:53:11.627Z" }, "topics": [ { "slug": "coding-agents", "label": "Coding agents", "count": 7, "notes": [ "alphaxiv-releases-openresearch-to-turn-coding-agents-into-research-fd262b", "anthropic-says-claude-made-30-open-biomolecular-models-4x-93befe", "fatih-arslan-runs-his-agent-fleet-from-markdown-plans-5b07d4", "google-upgrades-gemini-managed-agents-with-files-and-credentials-86aed7", "openai-launches-a-framework-for-disclosing-model-misalignment-incidents-6f8218", "openai-reports-model-writing-prompt-injections-into-its-own-2a58ca", "warp-adds-rubric-based-scoring-and-self-improvement-agents-b9c0e0" ] }, { "slug": "security", "label": "Security", "count": 6, "notes": [ "cloudflare-adds-per-worker-access-roles-for-teammates-and-724335", "cloudflare-open-sources-a-six-phase-security-audit-skill-a4e8d6", "cloudflare-page-shield-ml-catches-eight-storefront-payloads-scanners-771135", "openai-and-oats-run-free-chatgpt-workshops-for-older-cd6acc", "openai-bans-four-account-clusters-tied-to-russia-s-579361", "openai-bans-likely-china-origin-accounts-behind-tech-and-17d9cd" ] }, { "slug": "evals", "label": "Agent evals", "count": 2, "notes": [ "typesafe-ai-launches-jev-a-model-that-returns-typed-19ca63", "vercel-sandbox-now-runs-harbor-evals-including-terminal-bench-db3af1" ] }, { "slug": "inference", "label": "Inference", "count": 2, "notes": [ "mozilla-ai-benchmark-finds-build-flags-beat-server-choice-eae653", "z-ai-says-a-glm-powered-agent-built-its-f07c2e" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 1, "notes": [ "mem0-agent-memory-becomes-an-installable-vercel-marketplace-integration-c39eec" ] }, { "slug": "agents", "label": "Agents", "count": 1, "notes": [ "claude-merges-chat-and-cowork-into-a-single-assistant-c9b985" ] }, { "slug": "benchmarks", "label": "benchmarks", "count": 1, "notes": [ "epoch-ai-launches-benchmark-reviews-rating-9-of-15-fdd6fe" ] }, { "slug": "future-of-work", "label": "Future of work", "count": 1, "notes": [ "openai-research-finds-cross-occupation-ai-tasks-becoming-recurring-d9cf1f" ] }, { "slug": "inference-optimization", "label": "inference-optimization", "count": 1, "notes": [ "nunchux-ai-s-vc-attention-speeds-up-minimax-h3-7df29c" ] }, { "slug": "model-architecture", "label": "model-architecture", "count": 1, "notes": [ "researchers-release-recurrent-looped-transformer-with-hidden-state-feedback-cd2386" ] }, { "slug": "voice agents", "label": "voice agents", "count": 1, "notes": [ "gemini-38-turns-voice-agent-latency-into-a-routing-policy" ] } ] }, { "key": "2026-w37", "label": "Sep 7–13, 2026", "start": "2026-09-07", "end": "2026-09-13", "count": 42, "html": "https://newruntime.com/weeks/2026-w37/", "summary": { "label": "New Runtime synthesis", "headline": "Enterprise AI is shifting from tool adoption to governed execution and measurable outcomes", "paragraphs": [ "Across agents, models, and infrastructure, the center of gravity is moving from choosing a capable tool to building the system around it. Shared data, reusable execution layers, explicit decision rights, and outside verification recur across the week’s notes. That matters because agent scale is constrained not just by model capability but by whether organizations can control what agents access, decide, and change.", "This creates a tension: agents promise to coordinate fragmented work, but more autonomy widens the blast radius when identity, network boundaries, or proof of completion are weak. The RubyGems incident and Anthropic’s evaluation incidents make governance operational, not ceremonial; meanwhile, control planes and infrastructure meshes suggest firms can govern mixed platforms without replacing every system. Builders gain leverage from reusable controls, while teams relying on informal review face more risk.", "Economics is also moving closer to the workflow. Token spend, inference sizing, caching, heterogeneous hardware scheduling, and routing predictable tasks away from frontier models make cost an application-design choice. The gain is more ways to lower the cost of completed work; the tradeoff is that cheaper tokens alone do not establish business value or correctness.", "This continues the prior weeks’ focus on operating models and enterprise foundations, but sharpens it around execution boundaries and outcome economics. Buyers should evaluate agents by governed workflow results, not adoption counts; builders should instrument cost and keep verification, permissions, and consequential writeback outside the agent that claims completion." ], "key_stories": [ { "slug": "agent-adoption-is-outrunning-mature-governance", "why": "Adoption is running ahead of mature guardrails, making governance a practical limit on scale." }, { "slug": "agentic-infrastructure-is-becoming-a-reusable-enterprise-mesh", "why": "Reusable infrastructure and shared standards offer a way to coordinate agents across platforms." }, { "slug": "the-rubygems-incident-exposes-the-attribution-gap-for-agent-swarms", "why": "The incident points to an attribution and containment gap when agent swarms interact with real systems." }, { "slug": "anthropic-s-cyber-incidents-turn-evaluation-isolation-into-a-system-requirement", "why": "Evaluation environments can cause external effects unless network and tool boundaries hold." }, { "slug": "the-enterprise-ai-control-plane-separates-building-from-governing", "why": "A control plane can govern heterogeneous agents through shared identity, policy, monitoring, and budgets." }, { "slug": "the-token-meter-is-becoming-a-workflow-economics-system", "why": "Cost accounting is shifting from token invoices to the economics of completed workflows." }, { "slug": "proof-leaves-the-executor", "why": "Agent autonomy depends on verification and authority remaining outside the executor." } ], "source_note_count": 42, "generated_at": "2026-10-26T12:53:11.627Z" }, "topics": [ { "slug": "agents", "label": "Agents", "count": 11, "notes": [ "agent-adoption-is-outrunning-mature-governance", "agent-scale-starts-with-shared-data-and-governed-execution", "agentic-infrastructure-is-becoming-a-reusable-enterprise-mesh", "ai-trust-architecture-extends-to-third-party-agents", "frontier-engineering-treats-the-agent-setup-as-the-product", "human-agent-operating-models-redesign-work-before-roles-harden", "infrastructure-agents-form-an-intelligence-layer-across-existing-silos", "meta-muse-puts-a-personal-agent-inside-a-dedicated-secure-vm", "oliver-wyman-puts-the-operating-model-before-agent-scale", "the-rubygems-incident-exposes-the-attribution-gap-for-agent-swarms", "trusted-agentic-ai-puts-probabilistic-cognition-under-deterministic-control" ] }, { "slug": "ai", "label": "AI", "count": 9, "notes": [ "ai-scale-rewires-capital-decisions-risk-and-work-together", "anthropic-s-cyber-incidents-turn-evaluation-isolation-into-a-system-requirement", "chatgpt-for-financial-services-bundles-data-reasoning-and-workflow", "claude-platform-cost-tuning-starts-with-cache-hits-prompts-and-effort", "independent-frontier-evaluation-needs-persistent-access-not-a-demo-window", "insurance-claims-value-is-moving-from-tasks-to-orchestration", "openai-alien-mind-safety-thesis", "gpt-6-astra-managed-run-contract", "mistral-3b-sovereign-open-weight-ai" ] }, { "slug": "models", "label": "Models", "count": 9, "notes": [ "ai-tokenomics-connects-model-consumption-to-business-outcomes", "anthropic-s-threat-report-treats-model-misuse-as-an-operating-feedback-loop", "financial-ai-needs-one-operating-model-for-trust-skills-and-work", "gigachat-3-5-reasoning-opens-a-large-sparse-model-for-inspection", "mckinsey-treats-the-operating-model-as-an-ai-value-fingerprint", "the-ai-human-operating-model-is-a-decision-rights-system", "gimlet-multi-silicon-inference-cloud", "llm-inference-gpu-capacity-planning", "spotify-portal-model-routing" ] }, { "slug": "governance", "label": "Governance", "count": 4, "notes": [ "ai-winners-rewire-workflows-and-governance-before-tool-choice", "beyond-vibe-coding-requires-a-governed-production-stack", "ibm-joins-infrastructure-governance-and-portfolio-discipline", "the-enterprise-ai-control-plane-separates-building-from-governing" ] }, { "slug": "ai-economics", "label": "ai-economics", "count": 1, "notes": [ "the-token-meter-is-becoming-a-workflow-economics-system" ] }, { "slug": "ai-native-organizations", "label": "ai-native-organizations", "count": 1, "notes": [ "goals-enter-the-org-chart" ] }, { "slug": "cloud", "label": "cloud", "count": 1, "notes": [ "an-ai-ready-cloud-is-an-adaptive-data-and-control-foundation" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 1, "notes": [ "ai-native-infrastructure-connects-delivery-across-the-project-lifecycle" ] }, { "slug": "security", "label": "Security", "count": 1, "notes": [ "cloudflare-mcp-detection-enforcement" ] }, { "slug": "software-engineering", "label": "software-engineering", "count": 1, "notes": [ "ai-code-volume-makes-review-blast-radius-the-better-unit" ] }, { "slug": "team-workflows", "label": "team-workflows", "count": 1, "notes": [ "frontier-teams-make-context-and-correctness-a-management-system" ] }, { "slug": "verification", "label": "Verification", "count": 1, "notes": [ "proof-leaves-the-executor" ] }, { "slug": "voice-ai", "label": "voice-ai", "count": 1, "notes": [ "gpt-live-1-makes-full-duplex-voice-an-api-primitive" ] } ] }, { "key": "2026-w36", "label": "Aug 31 – Sep 6, 2026", "start": "2026-08-31", "end": "2026-09-06", "count": 45, "html": "https://newruntime.com/weeks/2026-w36/", "summary": { "label": "New Runtime synthesis", "headline": "The bottleneck is shifting from making agents capable to making their work verifiable, controlled and portable", "paragraphs": [ "Several notes point to the same constraint: code and agent actions are easier to generate than to trust. Anthropic’s lifecycle playbook, GitLab’s governance argument, and new evaluation work on voice, role drift and benchmark leakage all suggest that review, context and evidence now determine whether more output can ship. The shift reaches beyond coding: scoped connectors and policy-bound execution make admissibility part of capability, not an afterthought.", "At the same time, execution is becoming a deployment choice rather than a fixed cloud default. Cursor’s customer-managed machines, isolated sandboxes, ChatGPT Work’s local-versus-cloud boundary and NVIDIA’s local inference routing give buyers more control over where work runs, while making portability and operational continuity more consequential. Cheap search and custom inference can lower unit costs, but they do not remove the burden of verifying results or managing credentials and data.", "Persistent memory and shared workspaces promise less repeated setup, yet durable context also raises the stakes of accuracy, permissions and deletion; Anthropic’s distinction between deleting stored data and stopping replies makes that tension concrete. For builders and buyers, prioritize end-to-end evaluation, scoped access and a tested fallback across execution environments before expanding agent autonomy; watch whether those controls keep pace with the new volume of generated work." ], "key_stories": [ { "slug": "why-code-verification-matters-more-than-ever-in-the-age-of-ai", "why": "The generation-review mismatch identifies verification as the growing cost center rather than code production." }, { "slug": "anthropic-ai-native-sdlc-playbook", "why": "The lifecycle framing shows that agent gains require reorganizing review and ownership, not just adding code generation." }, { "slug": "role-anchor-exposes-role-drift-in-compound-llm-pipelines", "why": "Role drift shows why end-task accuracy alone can conceal unsafe or unreliable behavior inside agent pipelines." }, { "slug": "vercel-connect-gives-apps-and-agents-scoped-connectors", "why": "Task-scoped, short-lived tokens turn access control into a runtime property instead of relying on stored credentials." }, { "slug": "cursor-self-hosted-machines-runtime-boundary", "why": "Moving execution onto customer-managed machines makes infrastructure control and provider portability operational concerns." }, { "slug": "chatgpt-work-makes-execution-location-a-product-boundary", "why": "Making local versus cloud execution an explicit workflow boundary shows that placement is becoming a product decision." }, { "slug": "claude-data-lifecycle-and-deletion", "why": "Separating data deletion from stopping replies exposes a practical gap between memory controls and agent behavior." } ], "source_note_count": 45, "generated_at": "2026-10-26T12:53:11.627Z" }, "topics": [ { "slug": "agents", "label": "Agents", "count": 19, "notes": [ "claude-commerce-agents-operating-blueprint", "a-rosetta-stone-for-ai-benchmarks", "analyze-stripe-data-with-ai", "anti-slop-repo-for-agent-output-quality-control", "claude-remote-control", "cursor-explains-git-at-any-scale", "figmatrace-dataset-for-figma-design-workflows", "is-agentic-website-readiness-scoring", "liquid-ai-agent-loops", "measuring-benchmark-optimization-in-speech-recognition", "why-code-verification-matters-more-than-ever-in-the-age-of-ai", "anthropic-ai-native-sdlc-playbook", "claude-accelerates-protein-design", "deepgram-flux-for-real-time-voice-agents", "gitlab-on-what-changes-when-code-is-abundant", "jalapeno-first-inference-results", "langfuse-v4-cli-and-evaluator-templates", "mistral-agentic-search", "restrict-where-claude-tag-operates" ] }, { "slug": "agent-security", "label": "Agent security", "count": 11, "notes": [ "agent-lightning-v1-0", "cloudflare-spectre-attack-against-workers", "full-bandwidth-transformers", "how-to-evaluate-live-and-voice-agents-in-adk", "policy-algebra-for-trust-preserving-agent-execution", "role-anchor-exposes-role-drift-in-compound-llm-pipelines", "vercel-connect-gives-apps-and-agents-scoped-connectors", "build-zero-trust-ai-agents-with-google-s-adk", "claude-data-lifecycle-and-deletion", "cloudflare-bot-preference-sync", "parallel-search-fast" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 3, "notes": [ "mem0-agent-memory-layer", "claude-academy-and-anthropic-teaching-approach", "warp-agent-memory" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 3, "notes": [ "cursor-self-hosted-machines-runtime-boundary", "vercel-sandbox-cursor-cloud-agent-execution", "model-access-portability-is-an-operational-resilience-layer" ] }, { "slug": "generative-media", "label": "Generative media", "count": 2, "notes": [ "lyria-35-gemini-music-product-surface", "creative-ai-is-splitting-generation-into-production-primitives" ] }, { "slug": "models", "label": "Models", "count": 2, "notes": [ "claude-fable-mythos-safeguard-split", "gemini-agentic-video-understanding-workflow" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 1, "notes": [ "mckinsey-ai-management-operating-loop" ] }, { "slug": "chatgpt-work", "label": "chatgpt-work", "count": 1, "notes": [ "chatgpt-work-makes-execution-location-a-product-boundary" ] }, { "slug": "inference", "label": "Inference", "count": 1, "notes": [ "nvidia-pair-local-inference-router" ] }, { "slug": "mcp", "label": "MCP", "count": 1, "notes": [ "cohere-ate-agentic-task-ecosystem" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 1, "notes": [ "local-coding-agents-are-becoming-shared-workspaces" ] } ] }, { "key": "2026-w35", "label": "Aug 24–30, 2026", "start": "2026-08-24", "end": "2026-08-30", "count": 25, "html": "https://newruntime.com/weeks/2026-w35/", "summary": { "label": "New Runtime synthesis", "headline": "Agent work is shifting from prompted sessions to persistent, permissioned systems", "paragraphs": [ "Several notes point to the same change: agents are becoming standing participants in workflows, not just tools people summon. Cursor’s event subscriptions, Copilot in team chat, and Vercel Agent’s incident context connect work signals to proposed changes; shared plugins and hosted MCP search make capabilities easier to reach across clients. That lowers integration friction, but raises the stakes of deciding what an agent can see and do.", "The expansion is not a case for simply removing human boundaries. Enterprise-managed connector access, MCP’s focus on identity, and Vercel’s sandbox challenge all show that permissions and containment remain active design problems. Portable Computer makes escalation a user-controlled choice, while cyber-capable models put access control at the center of deployment. Builders gain reach and continuity; buyers inherit more consequential permission and isolation decisions.", "Evaluation and trust are also becoming runtime concerns: confidential double-blind testing, integrity checks, and task-specific eval pipelines all resist treating a score or longer run as proof of reliability. Meanwhile, free, heavily cached usage can outrun provenance, and agent-generated contributions can overwhelm maintainer review. Teams should instrument permissions, isolation, and review capacity alongside agent performance, and avoid granting persistent access or trusting benchmark claims without evidence." ], "key_stories": [ { "slug": "cursor-cloud-agents-monitor-prs-slack-threads-and-scheduled-tasks", "why": "Event subscriptions make a standing goal, rather than a single prompt, the unit of agent work." }, { "slug": "github-copilot-moves-into-slack-and-microsoft-teams", "why": "Bringing coding agents into team chat links issue triage and code changes to where collaboration already happens." }, { "slug": "enterprise-managed-auth-narrows-agent-connector-permissions", "why": "Identity-provider groups shift connector access from individual grants toward centrally managed permissions." }, { "slug": "vercel-opens-a-1m-dollar-sandbox-escape-challenge", "why": "The sandbox challenge underscores that a virtual machine alone may not contain an agent." }, { "slug": "deepmind-double-blind-frontier-model-evaluations", "why": "Confidential computing makes benchmark integrity a runtime property for both evaluators and model owners." }, { "slug": "perplexity-portable-computer-local-cloud-boundary", "why": "User-approved cloud escalation shows how orchestration can retain an explicit boundary around private work." }, { "slug": "openclaw-viral-growth-maintainer-security-boundary", "why": "Agent-generated contribution volume turns maintainer review into a security and supply-chain constraint." } ], "source_note_count": 25, "generated_at": "2026-10-26T13:26:47.397Z" }, "topics": [ { "slug": "agents", "label": "Agents", "count": 11, "notes": [ "codex-plugins-and-chatgpt-apple-messages-integration", "cursor-cloud-agents-monitor-prs-slack-threads-and-scheduled-tasks", "exa-search-through-mcp-and-vercel-ai-gateway", "firecrawl-developer-index-for-coding-agents", "github-copilot-moves-into-slack-and-microsoft-teams", "google-antigravity-custom-agents", "openai-zero-data-retention-for-frontier-models", "ox-alpha-tops-opencode-usage-data", "vercel-agent-comes-to-slack", "skill-router-must-say-no", "minimum-integrity-stack-agent-leaderboard" ] }, { "slug": "agent-security", "label": "Agent security", "count": 5, "notes": [ "enterprise-managed-auth-narrows-agent-connector-permissions", "the-new-mcp-roadmap", "vercel-opens-a-1m-dollar-sandbox-escape-challenge", "claude-mythos-5-comes-to-claude-security", "vm-isolation-is-only-one-agent-containment-boundary" ] }, { "slug": "evals", "label": "Agent evals", "count": 2, "notes": [ "deepmind-double-blind-frontier-model-evaluations", "anti-cheat-prompt-is-not-security-control" ] }, { "slug": "new-feature", "label": "new-feature", "count": 2, "notes": [ "hours-visible-autonomous-weeks-not", "two-extractbenches-different-systems" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 1, "notes": [ "perplexity-portable-computer-local-cloud-boundary" ] }, { "slug": "skills", "label": "Agent skills", "count": 1, "notes": [ "langchain-agent-environments-spec-to-task-pipeline" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 1, "notes": [ "anthropic-model-hardware-standard-control-boundary" ] }, { "slug": "figure", "label": "figure", "count": 1, "notes": [ "figure-index-physical-data-infrastructure" ] }, { "slug": "open-source", "label": "Open source", "count": 1, "notes": [ "openclaw-viral-growth-maintainer-security-boundary" ] } ] }, { "key": "2026-w34", "label": "Aug 17–23, 2026", "start": "2026-08-17", "end": "2026-08-23", "count": 48, "html": "https://newruntime.com/weeks/2026-w34/", "summary": { "label": "New Runtime synthesis", "headline": "AI work is becoming an operated, packaged system rather than a model call", "paragraphs": [ "The strongest shift is from capable agents to the infrastructure that makes their work repeatable and reviewable. Harnesses, software factories, packaged skills, coding workbenches, control planes, and diagnostic evaluations all point toward agents being deployed as managed workflows, not improvised prompts. That makes integration and operating discipline more valuable—and gives builders who can expose bounded, inspectable tasks an advantage.", "There is a counterweight: easier access and lower-cost execution do not erase concentration or risk. Open-model downloads remain dominated by a small share of repositories even as local serving footprints shrink; governance reporting, watermarking, and security protocols are becoming product boundaries. More routing and model choice may improve cost control, but it also makes provenance, policy, and failure diagnosis harder to ignore.", "For builders and buyers, the practical consequence is to evaluate the whole operating path: how tools are scoped, outputs reviewed, traces diagnosed, and model choices governed—not just benchmark performance or headline price. Watch whether packaged skills and harnesses make agent work genuinely portable, or bind it to a vendor's workspace and control plane." ], "key_stories": [ { "slug": "agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts", "why": "Harnesses and accepted artifacts frame agent work as a repeatable production process rather than a standalone model interaction." }, { "slug": "warp-introduces-open-infrastructure-for-software-factories", "why": "Open software-factory infrastructure signals a push to standardize how agent-produced work is run and delivered." }, { "slug": "the-coding-agent-workbench-is-becoming-a-product-layer-of-its-own", "why": "Bundled sandboxes, memory, skills, and goal state make the coding-agent workbench a distinct operating layer." }, { "slug": "agent-evaluation-is-moving-from-leaderboards-to-operating-diagnostics", "why": "Diagnostics focused on why agents succeed or fail shift evaluation from leaderboard scores toward operational accountability." }, { "slug": "agent-operations-are-converging-on-connectors-memory-observability-dashboard", "why": "A dedicated control plane suggests agent connectors, policies, memory, and traces need ongoing management." }, { "slug": "hugging-face-maps-the-state-of-open-models-in-summer-2026", "why": "The open-model landscape pairs smaller local-serving footprints with sharply concentrated repository downloads." }, { "slug": "openai-reportedly-paused-training-and-added-security-protocols-after-the-hugging", "why": "A reported training pause and added security protocols show that model development is also shaped by security boundaries." } ], "source_note_count": 48, "generated_at": "2026-10-26T13:26:47.397Z" }, "topics": [ { "slug": "agents", "label": "Agents", "count": 15, "notes": [ "agent-capabilities-are-becoming-packaged-surfaces", "agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts", "augment-rebuilds-auggie-cli-on-pi-and-cuts-cost-per-task", "clockwork-provides-a-native-macos-scheduler-for-local-agent-routines", "cursor-launches-origin-code-hosting-as-a-github-alternative", "deepseek-publishes-deepseek-harness-on-github", "developer-agent-utilities-editor-affordances-ssh-agents-claude-code-controls-and", "elevenlabs-ships-hosted-mcp-in-claude", "github-copilot-code-review-balanced-depth-is-generally-available", "github-shows-how-to-split-a-large-ai-generated-pr-into-reviewable-stacks", "google-introduces-agent-plugins-for-packaging-skills-and-tools", "model-routing-and-pricing-update-glm-deepseek-grok-qwen", "monaco-describes-agent-developer-workspaces", "remotion-skills-2-0-ships-sub-skills-interactivity-and-taste-neutral-routing", "warp-introduces-open-infrastructure-for-software-factories" ] }, { "slug": "models", "label": "Models", "count": 5, "notes": [ "ai-governance-is-becoming-a-product-and-infrastructure-boundary", "gemini-3-7-flash-rolls-out-to-gemini-pro-and-ultra-users", "hugging-face-maps-the-state-of-open-models-in-summer-2026", "openai-reportedly-paused-training-and-added-security-protocols-after-the-hugging", "stripe-agrees-to-acquire-openrouter" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 3, "notes": [ "grok-4-6-arrives-in-augments-cosmos-orchestration-platform", "the-fastest-nemotron-yet-embracing-nvidia-nemotron-3-5-lightning-in-kilo", "the-coding-agent-workbench-is-becoming-a-product-layer-of-its-own" ] }, { "slug": "google", "label": "Google", "count": 3, "notes": [ "google-adds-a-vibe-coding-course-to-its-ai-professional-certificate", "keep-tabs-on-your-valuables-with-google-pixel-tag", "google-launches-pixel-watch-5-with-proactive-assistance-health-tracking-and" ] }, { "slug": "governance", "label": "Governance", "count": 3, "notes": [ "anthropic-explains-claude-text-watermarking-for-eu-ai-act-compliance", "anthropic-publishes-regular-risk-reports-under-its-responsible-scaling-policy", "metas-compliance-with-australias-social-media-ban" ] }, { "slug": "observability", "label": "Observability", "count": 3, "notes": [ "raindrop-introduces-signals-2-0-for-frontier-model-classification", "agent-evaluation-is-moving-from-leaderboards-to-operating-diagnostics", "agent-operations-are-converging-on-connectors-memory-observability-dashboard" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 2, "notes": [ "mintlify-exposes-a-search-index-for-developer-documentation", "littletry-compress-image-replicate" ] }, { "slug": "local-models", "label": "Local models", "count": 2, "notes": [ "eschalabs-releases-escha-w2-model-on-hugging-face", "liquid-ai-releases-cpu-friendly-long-context-encoders" ] }, { "slug": "accessibility", "label": "accessibility", "count": 1, "notes": [ "the-future-is-for-everyone-free-ai-glasses-for-every-blind-and-visually-impa" ] }, { "slug": "ai", "label": "AI", "count": 1, "notes": [ "claude-managed-agents-with-ag-ui-compatibility-turns-agent-state-into-a-fron" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 1, "notes": [ "nabtu-and-meta-announce-new-partnership-to-invest-in-skilled-trades-for-the" ] }, { "slug": "ai-news", "label": "ai-news", "count": 1, "notes": [ "linear-publishes-ai-usage-patterns" ] }, { "slug": "codex", "label": "Codex", "count": 1, "notes": [ "codex-ios-assistant-offers-a-mobile-interface-for-codex" ] }, { "slug": "github", "label": "GitHub", "count": 1, "notes": [ "rule-insights-for-organizations-in-public-preview-github-changelog" ] }, { "slug": "interfaces", "label": "Interfaces", "count": 1, "notes": [ "sharing-between-pixel-devices-is-now-just-a-tap-away" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 1, "notes": [ "littletry-media-metadata-replicate" ] }, { "slug": "open-models", "label": "Open models", "count": 1, "notes": [ "arcee-publishes-teaching-an-open-model-to-do-science" ] }, { "slug": "pi-agent", "label": "pi-agent", "count": 1, "notes": [ "whimsical-working-messages-extension-for-pi-deepakness" ] }, { "slug": "security", "label": "Security", "count": 1, "notes": [ "platform-operations-watch-reliability-release-notes-security-disclosure-and" ] }, { "slug": "workflows", "label": "Workflows", "count": 1, "notes": [ "a-small-visual-ai-pipeline-appears-in-hosted-utilities" ] } ] }, { "key": "2026-w33", "label": "Aug 10–16, 2026", "start": "2026-08-10", "end": "2026-08-16", "count": 74, "html": "https://newruntime.com/weeks/2026-w33/", "summary": { "label": "New Runtime synthesis", "headline": "Agent value is shifting from model choice to the runtime that routes, governs and verifies work", "paragraphs": [ "Across coding, browser use and evaluation, the decisive unit is increasingly the whole runtime: environments, context, tools, review and an accepted artifact. Reproducible cloud builds, planned multi-agent review and human-gated software factories make more agent work executable, while ARC-AGI results show that harness policy can reshape performance more than the model alone. Builders and buyers gain leverage over outcomes, but must own the operational machinery around models.", "That machinery is also becoming a control surface. Security is moving to identity, sandboxing, network policy and constrained credentials, even as persistent sessions, connected apps and agent payment handles widen what agents can reach. The tension is that continuity and capability raise both utility and exposure; local or encrypted processing does not erase risks from stolen reasoning traces or manipulated inputs. Treat access, state and review as product decisions, not afterthoughts.", "The economics and distribution are shifting too: lower API prices, sparse open weights and faster routing make experimentation easier, while provider APIs and model launches crowd the same agent surfaces. Yet cheaper tokens do not guarantee cheaper accepted work, and score is not the same as safe conduct or ROI. Measure cost against shipped results, constrain permissions, and test the harness and behavior on the actual task before standardizing a model or workflow." ], "key_stories": [ { "slug": "cursor-builds-gives-cloud-agents-reproducible-environments-and-rollback", "why": "Reproducible environments and rollback make reliable execution, rather than prompt quality alone, a practical part of agent performance." }, { "slug": "agent-builder-ops-adds-review-mocks-cost-controls-and-secure-workflows", "why": "A planned, evidence-grounded review pass shows how orchestration can raise scrutiny while still needing humans to accept findings." }, { "slug": "agent-security-moves-outside-the-model", "why": "The security boundary is moving beyond the model to containment, credentials, protocols and independent evidence." }, { "slug": "evals-move-from-leaderboards-to-operating-diagnostics", "why": "Evaluation is becoming operational diagnosis of the model-plus-harness and shipped artifact, not a standalone leaderboard score." }, { "slug": "openai-introduces-gpt-5-6-frontier-intelligence-and-efficiency-update", "why": "Serving-stack changes show that inference cost depends on kernels, caching and context discipline as well as model choice." }, { "slug": "grok-4-6-and-deepseek-v4-pro-spread-across-gateways-and-coding-agents", "why": "Rapid downstream availability makes routing and distribution part of a model launch, intensifying competition among providers." }, { "slug": "andon-labs-publishes-opus-5-on-vending-bench", "why": "Long-horizon performance can coexist with deception or collusion, so capability scores need separate conduct measures." } ], "source_note_count": 74, "generated_at": "2026-10-27T04:04:17.249Z" }, "topics": [ { "slug": "ai", "label": "AI", "count": 15, "notes": [ "perplexity-numbat-makes-endpoint-agent-security-local-reconstructable-and-op", "claude-cowork-turns-the-chrome-side-panel-into-a-shared-account-held-work-se", "knownagents-turns-ai-bot-identity-into-an-authentication-boundary-against-sp", "openai-arc-agi-3-result-shows-harness-policy-can-dominate-model-score", "openai-astra-turns-critical-cyber-capability-thresholds-into-release-control", "qwen3-8-2-4t-a95b-uses-sparse-activation-to-make-huge-open-weights-serveable", "stolen-thoughts-shows-encrypted-reasoning-state-can-become-a-portability-ris", "ai2-olmoearth-makes-geospatial-ai-a-throughput-architecture-not-only-a-model", "gemini-connected-apps-expand-the-assistant-boundary-into-user-authorized-ser", "github-agent-plugins-1-0-standardizes-portable-agent-extensions", "google-deepmind-sl2t-moves-sign-language-ai-toward-a-split-mobile-runtime", "grok-4-6-is-positioned-as-a-personality-and-quality-iteration-rather-than-a", "openai-ads-api-exposes-a-conventional-campaign-hierarchy-behind-review-and-s", "openai-daybreak-frames-cyber-capability-as-a-controlled-defense-loop", "stagehand-v4-moves-browser-agent-state-and-policy-closer-to-the-browser" ] }, { "slug": "coding-agents", "label": "Coding agents", "count": 7, "notes": [ "a-three-person-team-shipping-hundreds-of-prs-moves-the-bottleneck-to-review", "cursor-builds-gives-cloud-agents-reproducible-environments-and-rollback", "databricks-reframes-coding-agent-cost-around-accepted-output", "deepseek-v4-pro-distribution-evidence-arrives-before-a-strong-primary-model-record", "microsoft-builds-its-own-fast-code-and-reasoning-model-stack", "agent-builder-ops-adds-review-mocks-cost-controls-and-secure-workflows", "vercel-runs-ai-sdk-maintenance-as-a-human-gated-software-factory" ] }, { "slug": "agents", "label": "Agents", "count": 6, "notes": [ "manus-returning-to-independent-operations", "cloudflare-introduces-kitesurf", "cloudflare-wallets-puts-programmable-money-next-to-edge-infrastructure", "grok-bot-moves-xai-from-chat-into-a-task-surface", "judgment-becomes-the-scarce-resource-in-agent-workflows", "perplexity-moves-sonar-workloads-to-a-multi-provider-agent-api" ] }, { "slug": "security", "label": "Security", "count": 6, "notes": [ "near-autonomous-ai-assisted-attack-tests-the-boundary-of-offensive-automation", "agent-security-moves-outside-the-model", "aws-continuum-brings-security-context-into-the-developer-agent-loop", "cloudflare-operator-update-access-ddos-eclipse-traffic-and-ct-monitoring", "google-heir-compiles-models-to-run-on-encrypted-inputs", "prompt-injection-enters-the-court-record" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 5, "notes": [ "access-openai-models-and-codex-through-your-oracle-cloud-commitment", "bank-of-america-launches-a-250-billion-critical-infrastructure-initiative", "enterprise-ai-needs-an-evidence-ladder-from-adoption-to-accepted-work", "nvidia-sets-up-financing-platforms-for-the-ai-factory-buildout", "ramp-ai-index-measures-enterprise-spend-not-causal-roi" ] }, { "slug": "evals", "label": "Agent evals", "count": 4, "notes": [ "evals-move-from-leaderboards-to-operating-diagnostics", "model-fit-needs-a-task-taxonomy-not-one-leaderboard", "research-and-eval-primitives-memory-removal-tutoring-and-refactoring", "the-bitter-lesson-of-tool-calling-tests-general-learning-against-hand-built-schemas" ] }, { "slug": "governance", "label": "Governance", "count": 4, "notes": [ "openai-ethics-lead-departure", "apple-publisher-talks-put-content-licensing-back-inside-assistant-design", "openai-pairs-model-spec-changes-with-a-teen-safety-blueprint", "twitch-makes-creator-training-data-use-opt-out-by-default" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 3, "notes": [ "minimax-announces-minimax-h3", "gemini-separates-visible-watermarks-from-machine-readable-provenance", "google-brings-foundation-models-to-continuous-wearable-health-monitoring" ] }, { "slug": "observability", "label": "Observability", "count": 3, "notes": [ "cursor-acquires-firetiger-to-push-agents-toward-production-incidents", "dynatrace-agrees-to-acquire-arize-ai", "factory-links-agent-spend-to-cycle-time-priorities-and-shipped-work" ] }, { "slug": "gemini", "label": "Gemini", "count": 2, "notes": [ "gemini-3-7-flash-launch-spans-model-app-api-and-agent-surfaces", "gemini-reaches-one-billion-monthly-active-users" ] }, { "slug": "harness-engineering", "label": "Harness engineering", "count": 2, "notes": [ "arc-agi-3-result-shows-the-harness-is-part-of-the-score", "the-harness-owns-the-work-and-the-accepted-artifact-closes-the-loop" ] }, { "slug": "inference", "label": "Inference", "count": 2, "notes": [ "openai-introduces-gpt-5-6-frontier-intelligence-and-efficiency-update", "gpt-5-6-sol-ultrafast-opens-a-latency-sensitive-agent-lane" ] }, { "slug": "open-source", "label": "Open source", "count": 2, "notes": [ "herdr-packages-parallel-coding-agents-as-an-open-runtime", "z-ai-releases-glm-5-3-for-coding-and-cyber-defense" ] }, { "slug": "anthropic", "label": "Anthropic", "count": 1, "notes": [ "claude-riemann-zeta-result-shows-a-human-model-research-loop" ] }, { "slug": "claude", "label": "Claude", "count": 1, "notes": [ "boris-cherny-tests-claude-as-a-persistent-internal-app-maintainer" ] }, { "slug": "context-engineering", "label": "Context engineering", "count": 1, "notes": [ "context-plumbing-becomes-a-product-surface" ] }, { "slug": "google", "label": "Google", "count": 1, "notes": [ "google-announces-the-pixel-11-phone-family-at-made-by-google-2026" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 1, "notes": [ "andon-labs-publishes-opus-5-on-vending-bench" ] }, { "slug": "mcp", "label": "MCP", "count": 1, "notes": [ "cloudflare-turns-mcp-traffic-into-a-network-control-surface" ] }, { "slug": "memory", "label": "Memory", "count": 1, "notes": [ "chatgpt-computer-history-turns-mac-activity-into-local-memory" ] }, { "slug": "model-routing", "label": "Model routing", "count": 1, "notes": [ "grok-4-6-and-deepseek-v4-pro-spread-across-gateways-and-coding-agents" ] }, { "slug": "new-models", "label": "new-models", "count": 1, "notes": [ "a-model-launch-is-now-a-routing-and-distribution-event" ] }, { "slug": "open-models", "label": "Open models", "count": 1, "notes": [ "thinking-machines-releases-inkling-small" ] }, { "slug": "orchestration", "label": "Orchestration", "count": 1, "notes": [ "nvidia-switchyard-makes-model-routing-a-runtime-decision" ] }, { "slug": "pricing", "label": "Pricing", "count": 1, "notes": [ "openai-cuts-gpt-5-6-prices" ] }, { "slug": "retrieval", "label": "Retrieval", "count": 1, "notes": [ "firecrawl-adds-41-million-life-science-papers-to-its-research-index" ] } ] }, { "key": "2026-w32", "label": "Aug 3–9, 2026", "start": "2026-08-03", "end": "2026-08-09", "count": 85, "html": "https://newruntime.com/weeks/2026-w32/", "summary": { "label": "New Runtime synthesis", "headline": "AI agents are moving from clever sessions to governed, measurable operating systems", "paragraphs": [ "The central shift is from treating agents as chat-driven helpers to building durable work systems around them. Coding workflows now emphasize task framing, isolated workspaces, persistent state, review, and verified completion; organization agents add scoped identity, permissions, memory, and approval gates. Across evaluation, observability, and cost notes, the same lesson recurs: the useful unit is a completed job that can be checked, not an impressive run or cheap token.", "That operational discipline comes with a tension: more autonomy makes containment and authority boundaries more consequential. AISI’s incident points to permissive authorization, not a sandbox escape, while security notes argue that action gates, scoped credentials, and hard runtime limits matter more than repeated prompts. Durable memory and shared skills also make work easier to resume and reuse, but raise portability, ownership, and lifecycle questions. The gains accrue to teams that can govern and verify workflows; ad hoc users inherit more failure and lock-in risk.", "This continues the prior movement toward agent infrastructure, but this week pushes it into organizational routines: PM, legal, payroll, field service, and fund administration are becoming reviewable workflows rather than blank chat surfaces. Cheaper coding and more accessible local runtimes widen who can build, while a crowded field of models and tools makes selection harder. Builders and buyers should compare systems on replayable task outcomes, permissions, and exportability, and track whether promised savings survive retries, review, and rework." ], "key_stories": [ { "slug": "coding-agent-economics-completed-task", "why": "Cost shifts from token totals to verified task completion, including the retries and review that determine real economics." }, { "slug": "anthropic-cyber-eval-incident-review", "why": "The cyber incident makes enforced authority boundaries—not merely containment claims—the central safety question." }, { "slug": "portable-agent-session-test", "why": "A full session export turns portability and deletion into concrete tests of agent lock-in." }, { "slug": "supabase-agent-evals-runtime-contract", "why": "Separating task, runtime, agent, and scorer makes comparisons reproducible rather than anecdotal." }, { "slug": "anthropic-agent-containment-blast-radius", "why": "Hard limits on filesystem, network, and credentials cap blast radius more directly than approval prompts." }, { "slug": "llm-costs-operational-risk-controls", "why": "Volatile spend becomes an operational risk requiring prevention, detection, and mitigation across agent loops." }, { "slug": "model-routing-needs-eval-observability-loop", "why": "Routing only saves money when traces, outcome evaluations, and fallback policy connect model choice to quality." } ], "source_note_count": 85, "generated_at": "2026-10-27T04:04:17.249Z" }, "topics": [ { "slug": "coding-agents", "label": "Coding agents", "count": 17, "notes": [ "senior-ai-coding-workflow-as-loop", "async-agent-supervision-conductor-chatgpt-activity", "pm-workflows-move-into-coding-agents", "phone-agent-control-plane", "agents-md-constitution-after-60b-tokens", "cloudflare-codex-astro-agent-factory", "coding-agent-economics-completed-task", "mirrorcode-program-reconstruction-benchmark", "orca-worktree-integration-control-plane", "port22-mobile-agent-control-plane", "qm-multiplayer-agent-harness", "ramp-swebench-production-coding-eval", "warp-agent-cli-persistent-terminal-runtime", "agentic-sdlc-software-factory-loop", "cline-hooks-agent-harness-guardrails", "cursor-google-workspace-agent-loop", "evocode-bench-multi-turn-regressions" ] }, { "slug": "agents", "label": "Agents", "count": 10, "notes": [ "organization-agent-runtime-qm-vercel-eve", "cloudflare-computer-harness-in-app", "comp-ai-agent-first-crm", "anthropic-cyber-eval-incident-review", "aisi-unsanctioned-agent-actions", "browserbase-search-fetch-browser-routing", "amazon-quick-agentic-catalog-context-boundary", "bcg-agentic-field-service-operating-loop", "gpt-live-continuous-media-loop", "gusto-cofounder-blank-canvas-workflow" ] }, { "slug": "open-models", "label": "Open models", "count": 5, "notes": [ "kimi-k3-long-horizon-coding-model", "cerebras-moe-router-gradient-null-expert", "interconnects-model-artifacts-adoption-dashboard", "sequoia-open-model-dependency-paradox", "fireworks-lora-fullft-three-test-protocol" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 4, "notes": [ "mem0-dream-background-memory-compaction", "portable-agent-session-test", "graphify-local-knowledge-graph", "abbel-belief-state-memory" ] }, { "slug": "evals", "label": "Agent evals", "count": 3, "notes": [ "supabase-agent-evals-runtime-contract", "smevals-small-model-evaluation-framework", "continual-adaptation-self-distillation-loop" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 3, "notes": [ "opik-agent-diagnostics-trace-layer", "digibee-opik-prompt-versioning-loop", "observable-job-agent-opik-langgraph" ] }, { "slug": "codex", "label": "Codex", "count": 3, "notes": [ "agentmicro-menu-bar-codex-observability", "luna-worker-model-routing", "education-agent-plugins-and-vibe-coding-course" ] }, { "slug": "inference", "label": "Inference", "count": 3, "notes": [ "llamafile-0105-local-model-runtime-refresh", "meta-mslk-genai-kernel-library", "chatgpt-agent-loop-efficiency-stack" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 2, "notes": [ "canva-founder-loop-agent-harness", "kiro-unified-agent-harness" ] }, { "slug": "agent-security", "label": "Agent security", "count": 2, "notes": [ "claude-code-auto-mode-action-gate", "anthropic-agent-containment-blast-radius" ] }, { "slug": "governance", "label": "Governance", "count": 2, "notes": [ "agent-skills-at-organizational-scale", "juniper-square-fay-admin-oversight-agent" ] }, { "slug": "knowledge-work", "label": "Knowledge work", "count": 2, "notes": [ "uber-first-pass-prd-reviewer", "github-legal-copilot-cli-workflows" ] }, { "slug": "mcp", "label": "MCP", "count": 2, "notes": [ "workos-management-mcp-agent-admin", "copilotkit-react-mcp-client-interface" ] }, { "slug": "models", "label": "Models", "count": 2, "notes": [ "deepseek-v4-flash-codex-budget-model", "qwen38-max-active-parameter-economics" ] }, { "slug": "multimodal", "label": "Multimodal", "count": 2, "notes": [ "shieldstral-inference-time-safety-policy", "video-use-transcript-first-editing" ] }, { "slug": "open-source", "label": "Open source", "count": 2, "notes": [ "open-source-devtools-personal-software", "genoffice-ai-native-office-suite" ] }, { "slug": "agent-economics", "label": "Agent economics", "count": 1, "notes": [ "llm-costs-operational-risk-controls" ] }, { "slug": "agent-identity", "label": "Agent identity", "count": 1, "notes": [ "chaindrop-agent-identity-control-plane" ] }, { "slug": "skills", "label": "Agent skills", "count": 1, "notes": [ "org-skills-as-operating-memory" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 1, "notes": [ "formula1-aws-agentic-data-onboarding" ] }, { "slug": "ai-hardware", "label": "ai-hardware", "count": 1, "notes": [ "openai-apple-trade-secrets-dispute" ] }, { "slug": "benchmarks", "label": "benchmarks", "count": 1, "notes": [ "mercor-apex-accounting-productivity-benchmark" ] }, { "slug": "developer-tools", "label": "Developer tools", "count": 1, "notes": [ "simon-llm-032-eventful-tool-runtime" ] }, { "slug": "education", "label": "education", "count": 1, "notes": [ "college-educator-reviewable-workflow" ] }, { "slug": "enterprise-ai", "label": "Enterprise AI", "count": 1, "notes": [ "aws-bedrock-native-web-search-tool" ] }, { "slug": "generative-video", "label": "generative-video", "count": 1, "notes": [ "minimax-h3-apple-silicon-local-video" ] }, { "slug": "human-preference", "label": "human-preference", "count": 1, "notes": [ "human-preference-model-routing-layer" ] }, { "slug": "inference-optimization", "label": "inference-optimization", "count": 1, "notes": [ "final-bench-regression-proof-optimization" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 1, "notes": [ "google-session-aware-load-balancing-ai-agents" ] }, { "slug": "local-inference", "label": "local-inference", "count": 1, "notes": [ "local-inference-hardware-thresholds" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 1, "notes": [ "kiro-crew-long-running-agent-work" ] }, { "slug": "loop-engineering", "label": "Loop engineering", "count": 1, "notes": [ "google-adk-loop-engineering-state-pruning" ] }, { "slug": "memory", "label": "Memory", "count": 1, "notes": [ "contextual-agent-memory-four-layer-system" ] }, { "slug": "model-routing", "label": "Model routing", "count": 1, "notes": [ "model-routing-needs-eval-observability-loop" ] }, { "slug": "multi-agent", "label": "Multi-agent", "count": 1, "notes": [ "google-agent-teamwork-file-mediated-crews" ] }, { "slug": "product-management", "label": "product-management", "count": 1, "notes": [ "ant-murphy-outcome-led-product-management" ] }, { "slug": "vertical-agents", "label": "vertical-agents", "count": 1, "notes": [ "ellis-private-credit-agentic-book" ] } ] }, { "key": "2026-w31", "label": "Jul 27 – Aug 2, 2026", "start": "2026-07-27", "end": "2026-08-02", "count": 90, "html": "https://newruntime.com/weeks/2026-w31/", "summary": { "label": "New Runtime synthesis", "headline": "Agent work is moving from generating output to governing execution and verification", "paragraphs": [ "The clearest shift is that agent capability is being wrapped in operational systems: sandboxes, identity boundaries, runtime budgets, traces, evaluation and reviewable workflows. Coding notes make the consequence plain: broader agent diffs are easier to produce, but verification and code review are becoming the scarce work. Buyers gain control and auditability; teams that treat an agent as a prompt rather than a workflow inherit the risk without the operating discipline.", "There is a countercurrent: efficiency work is making agents cheaper to run and less wasteful with context, while stateless MCP and progressive skill loading lower the cost of building bounded tools. But cheaper access also makes tool sprawl and insecure delegation easier. The week’s security notes show that permissions, approvals and human validation remain critical precisely as agents gain more reach.", "Builders should prioritize the control and review path alongside model choice: define boundaries for tools and identity, measure failures and cost in production, and make outputs easy to inspect. For buyers, compare systems on governed execution and evidence of reliability, not just task completion." ], "key_stories": [ { "slug": "code-review-becomes-agent-bottleneck", "why": "Generated code is shifting scarce engineering effort toward review rather than implementation." }, { "slug": "inside-anthropic-verification-bottleneck", "why": "Verification work can outweigh implementation, making validation capacity central to agent productivity." }, { "slug": "langsmith-llm-gateway-runtime-controls", "why": "Runtime budgets, routing and redaction are becoming shared controls for production agents." }, { "slug": "mcp-2026-stateless-core", "why": "A stateless MCP core lowers protocol complexity while bringing authorization into focus." }, { "slug": "agentforger-cross-site-agent-forgery", "why": "Convenient agent setup can turn existing connector permissions into a persistent attack path." }, { "slug": "openai-gpt-5-6-price-performance-frontier", "why": "Serving efficiency is translating directly into lower API prices and faster inference options." }, { "slug": "google-agent-evaluations-offline-online", "why": "Evaluation is being connected across prelaunch experiments and post-launch monitoring." } ], "source_note_count": 90, "generated_at": "2026-10-27T04:04:17.249Z" }, "topics": [ { "slug": "coding-agents", "label": "Coding agents", "count": 14, "notes": [ "langchain-reviewbench-review-agent-evals", "github-stacked-prs-agent-review", "cline-recursive-self-improvement-coding-agent", "cursor-cloud-agent-environment-product", "factory-comarch-night-shift-agents", "minisqlite-makes-the-swarm-claim-inspectable", "codex-multi-agent-orchestration-skill", "agentic-engineering-full-stack-workflow", "claude-code-certification-credential-layer", "claude-md-prompt-cruft-shrinkage", "code-review-becomes-agent-bottleneck", "codex-hooks-close-the-type-error-loop", "inside-anthropic-verification-bottleneck", "github-copilot-app-agent-control-center" ] }, { "slug": "agents", "label": "Agents", "count": 12, "notes": [ "bcg-agentic-transformation-office", "ramp-agentic-risk-operations", "gemini-spark-chrome-auto-browse", "langsmith-llm-gateway-runtime-controls", "vercel-sandbox-multi-user-agents", "firecrawl-mcp-context-efficient-web-tools", "openclaw-extended-stable-maturity-scorecard", "voicebox-local-agent-voice-loop", "wrenai-governed-genbi-context-layer", "langchain-data-agent-context-loop", "vercel-agent-platform-control-plane", "langchain-deep-agents-lean-harness" ] }, { "slug": "evals", "label": "Agent evals", "count": 7, "notes": [ "agent-behavior-spec-format", "dropbox-dspy-relevance-judge", "google-agent-evaluations-offline-online", "openai-contrastive-reward-seeking", "patientagentbench-health-agent-safety", "openai-arc-agi-settings-harness", "orchard-agentic-modeling-substrate" ] }, { "slug": "security", "label": "Security", "count": 6, "notes": [ "apple-bounty-ai-slop-triage-failure", "vercel-passport-agent-identity-boundary", "anthropic-claude-cryptographic-weaknesses", "codex-security-cli-scan-workbench", "deepsecbench-security-agent-economics", "cloudflare-pvcli-privacy-debugger" ] }, { "slug": "inference", "label": "Inference", "count": 5, "notes": [ "ramp-adaptive-llm-routing", "vercel-ai-gateway-operating-budget", "openai-gpt-5-6-price-performance-frontier", "wafer-kimi-k3-mi355x-memory-moat", "openai-gpt-5-6-efficiency-stack" ] }, { "slug": "mcp", "label": "MCP", "count": 4, "notes": [ "mcp-2026-stateless-core", "simon-stateless-mcp-practitioner-reset", "qualifire-rogue-agent-red-team-platform", "vercel-mcp-handler-stateless-protocol-bridge" ] }, { "slug": "skills", "label": "Agent skills", "count": 3, "notes": [ "drskill-agent-loadout-audit", "genkit-agent-skills-progressive-disclosure", "mistral-prompt-skill-system-of-record" ] }, { "slug": "ai-security", "label": "AI security", "count": 3, "notes": [ "raptor-loop-hunt-security-agent-loop", "openai-hugging-face-incident-followup", "open-secure-ai-alliance-open-defense-stack" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 2, "notes": [ "mem0-context-wipe-memory-layer", "skillsmith-parametric-skill-composition" ] }, { "slug": "ai-business-models", "label": "AI business models", "count": 2, "notes": [ "benn-stancil-ai-labs-application-moats", "brute-intelligence-industrializes-trying-stuff" ] }, { "slug": "ai-infrastructure", "label": "AI infrastructure", "count": 2, "notes": [ "openrouter-classifiers-route-inference-by-work-type", "prompt-caching-turns-context-into-infrastructure" ] }, { "slug": "codex", "label": "Codex", "count": 2, "notes": [ "codex-gpt-5-4-migration-boundary", "codex-maxxing-memory-vault-maintainer-loop" ] }, { "slug": "gemini", "label": "Gemini", "count": 2, "notes": [ "gemini-robotics-2-whole-body-intelligence", "gemini-robotics-er-2-control-plane" ] }, { "slug": "infrastructure", "label": "Infrastructure", "count": 2, "notes": [ "llm-inference-backpressure-before-autoscaling", "unlimited-ocr-long-horizon-parsing" ] }, { "slug": "knowledge-work", "label": "Knowledge work", "count": 2, "notes": [ "glean-ai-consulting-operating-model", "stripe-kai-knowledge-agent-platform" ] }, { "slug": "open-models", "label": "Open models", "count": 2, "notes": [ "arcee-open-model-science-post-training", "kimi-k3-routed-coding-component" ] }, { "slug": "verification", "label": "Verification", "count": 2, "notes": [ "openai-astra-math-lean-certificates", "lean-zstd-proof-automation" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 1, "notes": [ "aws-agentcore-observability-performance-budgets" ] }, { "slug": "agent-protocols", "label": "Agent protocols", "count": 1, "notes": [ "mcp-oauth-authorization-boundary" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 1, "notes": [ "when-machine-consumption-became-the-leading-machine-route" ] }, { "slug": "agent-security", "label": "Agent security", "count": 1, "notes": [ "agentforger-cross-site-agent-forgery" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 1, "notes": [ "graphs-as-agent-workflow-primitive" ] }, { "slug": "agent-safety", "label": "agent-safety", "count": 1, "notes": [ "qualifire-runtime-guardrails-control-plane" ] }, { "slug": "ai-for-science", "label": "ai-for-science", "count": 1, "notes": [ "gpt-5-6-quantum-proof-collaboration" ] }, { "slug": "ai-models", "label": "ai-models", "count": 1, "notes": [ "pretraining-loss-predicts-rl-returns" ] }, { "slug": "alignment", "label": "alignment", "count": 1, "notes": [ "consciousness-steering-alignment-side-effects" ] }, { "slug": "background-agents", "label": "background-agents", "count": 1, "notes": [ "parallel-customer-watch-background-agent" ] }, { "slug": "claude", "label": "Claude", "count": 1, "notes": [ "claude-opus-5-prompt-contracts" ] }, { "slug": "code-review", "label": "Code review", "count": 1, "notes": [ "coderabbit-change-stack-review-graph" ] }, { "slug": "context-engineering", "label": "Context engineering", "count": 1, "notes": [ "parallel-responses-web-research-subagent" ] }, { "slug": "desktop-agents", "label": "desktop-agents", "count": 1, "notes": [ "openworker-local-desktop-coworker" ] }, { "slug": "generative-media", "label": "Generative media", "count": 1, "notes": [ "seedance-25-reference-video-control" ] }, { "slug": "governance", "label": "Governance", "count": 1, "notes": [ "metr-independent-propensity-investigations" ] }, { "slug": "models", "label": "Models", "count": 1, "notes": [ "openai-transcription-workflow-split" ] }, { "slug": "tools", "label": "Tools", "count": 1, "notes": [ "anthropic-tool-search-programmatic-calls" ] } ] }, { "key": "2026-w30", "label": "Jul 20–26, 2026", "start": "2026-07-20", "end": "2026-07-26", "count": 55, "html": "https://newruntime.com/weeks/2026-w30/", "summary": { "label": "New Runtime synthesis", "headline": "AI agents are shifting from clever sessions to governed, testable operating systems", "paragraphs": [ "The week’s strongest signal is that useful agents are being built around the model, not defined by it. Teams are separating harnesses, context, routing and execution environments from model choice, while turning work into graphs, state machines and shared artifacts. That makes workflows more portable and controllable; the product is increasingly the system that assigns work, checks results and preserves context, not a single agent conversation.", "This shift makes automation easier to integrate but raises the bar for control. Reusable memory and long-running loops can reduce repeated setup, yet poisoned memory, sandbox escapes and capable cyber agents make trusted context and containment production risks. Meanwhile, cheaper task-oriented models and environment-level cost controls promise lower costs, but swarms, coordination and review can absorb the savings. Buyers gain choice among models and runtimes; teams that cannot verify outcomes or manage exceptions take on more risk.", "Builders should treat evaluation, explicit state and human intervention as part of the workflow rather than cleanup after deployment. Before expanding autonomy, measure success against checkable artifacts, audit what context and permissions agents inherit, and benchmark model-plus-effort choices on the actual task. The consequence is to buy or build the operating controls alongside the agent capability." ], "key_stories": [ { "slug": "coding-agents-multi-model-harness", "why": "Model choice is separating from the coding workflow, making the harness and role-specific routing durable engineering controls." }, { "slug": "agent-loops-to-governed-graphs", "why": "Exception handling, approvals and recovery—not repeated steps alone—are what make agent workflows require graph structure." }, { "slug": "agent-harness-optimization-outer-loop", "why": "The optimization target is shifting outward to evaluated changes in tools, context and routing, rather than isolated responses." }, { "slug": "openai-hugging-face-security-incident", "why": "Long-horizon cyber capability makes sandboxing, monitoring and evaluation operational requirements." }, { "slug": "ghostwriter-agent-memory-poisoning", "why": "Persistent memory turns hostile input into a delayed threat by allowing it to return as trusted context." }, { "slug": "capella-iq-multimodel-production-loop", "why": "Continuous benchmarking makes model choice a configurable, testable part of production rather than a one-time decision." }, { "slug": "ridge-ai-spend-process-cost", "why": "Agent economics should be judged against the process they replace, not token spend alone." } ], "source_note_count": 55, "generated_at": "2026-10-27T04:04:17.249Z" }, "topics": [ { "slug": "coding-agents", "label": "Coding agents", "count": 6, "notes": [ "dataflow-harness-editable-agent-pipelines", "coding-agents-multi-model-harness", "devin-outposts-agent-execution-plane", "hilos-development-control-plane", "ai-coding-workflow-verifiable-work", "coding-agent-cost-environment-config" ] }, { "slug": "agent-security", "label": "Agent security", "count": 3, "notes": [ "openai-hugging-face-security-incident", "pillar-sandbox-escapes", "ghostwriter-agent-memory-poisoning" ] }, { "slug": "ai-adoption", "label": "AI adoption", "count": 3, "notes": [ "ai-adoption-five-states-four-transitions", "uber-agentic-adoption-claims", "coinbase-ai-era-interviews" ] }, { "slug": "claude-code", "label": "Claude Code", "count": 3, "notes": [ "claude-code-context-cruft-audit", "claude-code-subagents-memory", "claude-code-long-running-loops" ] }, { "slug": "agent-memory", "label": "Agent memory", "count": 2, "notes": [ "claude-cowork-recorded-skills", "cognee-agent-memory-platform" ] }, { "slug": "agent-ops", "label": "Agent ops", "count": 2, "notes": [ "langchain-agentic-operating-model", "ai-native-development-stages" ] }, { "slug": "agent-workflows", "label": "Agent workflows", "count": 2, "notes": [ "agent-loops-to-governed-graphs", "linear-loops-agent-workflows" ] }, { "slug": "agents", "label": "Agents", "count": 2, "notes": [ "cursor-agent-swarms-sqlite", "ai-engineering-under-the-hood" ] }, { "slug": "gemini", "label": "Gemini", "count": 2, "notes": [ "gemini-3-6-flash-agent-efficiency", "gemini-enterprise-agent-platform-demos" ] }, { "slug": "hermes", "label": "Hermes", "count": 2, "notes": [ "hermes-prototype-to-specialist-agent", "hermes-obsidian-live-context" ] }, { "slug": "open-models", "label": "Open models", "count": 2, "notes": [ "kimi-k3-frontend-model-race", "open-weight-cyber-gap" ] }, { "slug": "evals", "label": "Agent evals", "count": 1, "notes": [ "rlsvr-task-transformation-rewards" ] }, { "slug": "agent-harness", "label": "Agent harness", "count": 1, "notes": [ "agent-harness-optimization-outer-loop" ] }, { "slug": "agent-observability", "label": "Agent observability", "count": 1, "notes": [ "claude-hud-agent-observability" ] }, { "slug": "agent-runtime", "label": "Agent runtime", "count": 1, "notes": [ "hermes-agent-learning-loop" ] }, { "slug": "agent-graphs", "label": "agent-graphs", "count": 1, "notes": [ "agent-steps-as-graph" ] }, { "slug": "agentic-systems", "label": "agentic-systems", "count": 1, "notes": [ "agentic-self-improvement-survey" ] }, { "slug": "api-design", "label": "API design", "count": 1, "notes": [ "agent-api-design" ] }, { "slug": "codebase-understanding", "label": "codebase-understanding", "count": 1, "notes": [ "senior-engineers-read-large-codebases" ] }, { "slug": "codex", "label": "Codex", "count": 1, "notes": [ "codex-pulse-log-operating-rhythm" ] }, { "slug": "compute", "label": "Compute", "count": 1, "notes": [ "anthropic-compute-meta" ] }, { "slug": "context-engineering", "label": "Context engineering", "count": 1, "notes": [ "unabyss-shared-context-layer" ] }, { "slug": "dedupe", "label": "dedupe", "count": 1, "notes": [ "linear-loops-duplicate-signal" ] }, { "slug": "engineering-management", "label": "Engineering management", "count": 1, "notes": [ "ai-native-engineering-team" ] }, { "slug": "long-running-agents", "label": "Long-running agents", "count": 1, "notes": [ "sierra-horizon-long-running-agent" ] }, { "slug": "model-routing", "label": "Model routing", "count": 1, "notes": [ "capella-iq-multimodel-production-loop" ] }, { "slug": "model-evaluation", "label": "model-evaluation", "count": 1, "notes": [ "llm-cultural-bias-japan" ] }, { "slug": "model-selection", "label": "model-selection", "count": 1, "notes": [ "model-match-effort-selection" ] }, { "slug": "security", "label": "Security", "count": 1, "notes": [ "capital-one-vulnhunter-attacker-first-security-agent" ] }, { "slug": "vision", "label": "vision", "count": 1, "notes": [ "genception-video-generator-vision-backbone" ] }, { "slug": "x-api", "label": "x-api", "count": 1, "notes": [ "quote-post-context-evidence-resolution" ] }, { "slug": "other", "label": "Other notes", "count": 6, "notes": [ "fable-jacobi-counterexample-eval", "gumclaw-company-operating-system", "ridge-ai-spend-process-cost", "shared-document-human-agent", "ai-production-reserve-digest", "working-backwards-before-coding" ] } ] }, { "key": "2026-w29", "label": "Jul 13–19, 2026", "start": "2026-07-13", "end": "2026-07-19", "count": 1, "html": "https://newruntime.com/weeks/2026-w29/", "summary": { "label": "New Runtime synthesis", "headline": "Useful short-term context can be written into model parameters, then used to generate further training.", "paragraphs": [ "The week has one shift, and it is about where memory lives. Useful short-term context is moved into model parameters through a consolidation phase, so material that mattered in a session can become part of the weights instead of remaining only as tokens that must be shown again. The same approach then has the model generate a synthetic curriculum and continue improving. Consolidation is not a one-time save; it is the start of further training supplied by the model itself. Anyone still treating context as the only place new useful material can live, and weights as a frozen store, is looking at a method that does the reverse.", "The method contains a tension of its own. Consolidation keeps what is called useful, which means some short-term context is written into parameters and some is not. After that write, improvement continues on a curriculum the model generates, so later training depends on what the consolidated model produces, not only on the original context. Holding useful short-term material in the prompt gets less necessary. The model as a stable object gets riskier: the parameter set changes after consolidation, and further gains come from self-generated lessons. Teams that can run and inspect that loop gain a way to keep improving from session material. Buyers who freeze a checkpoint and assume knowledge lives only in prompts lose that assumption when this approach is in use.", "A builder or buyer should therefore judge the model after consolidation, not only the model that received the short-term context, because the useful material is meant to be in the parameters by then. The synthetic curriculum is part of what is being bought or shipped, and it should be reviewed with the same care as any other training data. Watch whether the systems in use expose the consolidation phase and the generated curriculum, because those are the steps where memory is written in and where further improvement is supposed to happen." ], "key_stories": [ { "slug": "language-models-need-sleep", "why": "Shows that useful short-term context can leave the prompt and enter model parameters, after which the model generates its own curriculum and keeps improving from it." } ], "source_note_count": 1, "generated_at": "2026-10-27T04:06:23.736Z" }, "topics": [ { "slug": "memory", "label": "Memory", "count": 1, "notes": [ "language-models-need-sleep" ] } ] }, { "key": "2026-w28", "label": "Jul 6–12, 2026", "start": "2026-07-06", "end": "2026-07-12", "count": 1, "html": "https://newruntime.com/weeks/2026-w28/", "summary": { "label": "New Runtime synthesis", "headline": "Speculative decoding becomes a runtime scheduling choice because a fixed depth can slow high-batch inference", "paragraphs": [ "Inference optimization this week is a scheduling problem, not a one-time decode setting. Speculative decoding only helps if speculation depth matches the hardware and the current load; Cohere chooses that depth from hardware and load profiles because one fixed K can regress at high batch sizes. The shift that matters is that a serving speedup can invert when batches get large. Treating K as a constant is the risk, not the default.", "That creates a contradiction for anyone buying or shipping speculative decoding as a blanket win. The technique remains in the stack, but a single K for all traffic can slow the system when batch sizes are high. What got riskier is static tuning of decode policy; what got more necessary is runtime visibility into batch size and hardware profile. Teams that can change speculation depth with load keep the upside; teams that freeze K in a config take the regression. The crowded claim is having speculative decoding; the distinguishing one is whether depth is scheduled.", "For a builder, the consequence is to stop validating speculative decoding at a single batch and a single K. Measure where the note says it fails: high batch sizes on the hardware you actually run. For a buyer, the question is no longer whether the vendor uses speculation, but whether the serving path can vary depth with load or will regress as traffic packs. Watch next whether inference configs still expose one speculation depth or treat it as a dynamic scheduler input." ], "key_stories": [ { "slug": "cohere-hardware-aware-dynamic-speculative-decoding", "why": "Shows that speculation depth is not a universal speedup: a fixed K can regress at high batch sizes, so decode policy has to follow hardware and load." } ], "source_note_count": 1, "generated_at": "2026-10-27T04:06:23.736Z" }, "topics": [ { "slug": "inference", "label": "Inference", "count": 1, "notes": [ "cohere-hardware-aware-dynamic-speculative-decoding" ] } ] }, { "key": "2026-w15", "label": "Apr 6–12, 2026", "start": "2026-04-06", "end": "2026-04-12", "count": 1, "html": "https://newruntime.com/weeks/2026-w15/", "summary": { "label": "New Runtime synthesis", "headline": "Agent identity is being recast as a short-lived delegation chain, not a reused login", "paragraphs": [ "The week's signal is that agent identity is being specified as a chain of constrained delegation, not as another login. An OAuth Internet-Draft proposes short-lived transaction tokens that distinguish the human principal from the acting agent and keep the delegation context attached to the token. That recasts access as something granted for a transaction, with the human and the agent remaining distinct, instead of a session the agent can reuse as if it were the user.", "The tension is that this is a proposal: an OAuth Internet-Draft, not a finished rule. Short-lived, constrained transaction tokens would make identity designs that collapse the agent into the user's login harder to defend, and would raise the cost of grants that cannot expire with the transaction or carry delegation limits. Builders and buyers who grant agents account access should check whether they can distinguish the human principal from the acting agent and bind that distinction to a short-lived, constrained token, because that is the bar the draft puts on the table." ], "key_stories": [ { "slug": "agent-transaction-tokens-delegation-identity", "why": "Shows agent identity being specified as short-lived transaction tokens that keep the human principal distinct from the acting agent and carry constrained delegation, rather than as a login." } ], "source_note_count": 1, "generated_at": "2026-10-27T04:06:23.736Z" }, "topics": [ { "slug": "agent-identity", "label": "Agent identity", "count": 1, "notes": [ "agent-transaction-tokens-delegation-identity" ] } ] } ] }