{"schema_version":"newruntime-week-v0.1","type":"week_in_review","generated_at":"2026-11-18T18:00:02.626Z","key":"2026-w31","label":"Jul 27 – Aug 2, 2026","start":"2026-07-27","end":"2026-08-02","count":90,"routes":{"html":"https://newruntime.com/weeks/2026-w31/","json":"https://newruntime.com/weeks/2026-w31.json","markdown":"https://newruntime.com/weeks/2026-w31.md","index":"https://newruntime.com/weeks.json","previous":"https://newruntime.com/weeks/2026-w30.json","next":"https://newruntime.com/weeks/2026-w32.json"},"note_routes":{"html":"https://newruntime.com/posts/{slug}/","markdown":"https://newruntime.com/posts/{slug}.md","json":"https://newruntime.com/posts/{slug}.json"},"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","title":"Code Review Becomes the Agent Bottleneck","why":"Generated code is shifting scarce engineering effort toward review rather than implementation."},{"slug":"inside-anthropic-verification-bottleneck","title":"At Anthropic, Verification Consumes More Work Than Implementation","why":"Verification work can outweigh implementation, making validation capacity central to agent productivity."},{"slug":"langsmith-llm-gateway-runtime-controls","title":"LangSmith LLM Gateway Puts Runtime Controls Between Agents and Models","why":"Runtime budgets, routing and redaction are becoming shared controls for production agents."},{"slug":"mcp-2026-stateless-core","title":"MCP 2026 Makes The Protocol Stateless","why":"A stateless MCP core lowers protocol complexity while bringing authorization into focus."},{"slug":"agentforger-cross-site-agent-forgery","title":"AgentForger Turns A ChatGPT Link Into An Agent Builder Attack","why":"Convenient agent setup can turn existing connector permissions into a persistent attack path."},{"slug":"openai-gpt-5-6-price-performance-frontier","title":"GPT-5.6 Turns Efficiency Work Into API Economics","why":"Serving efficiency is translating directly into lower API prices and faster inference options."},{"slug":"google-agent-evaluations-offline-online","title":"Google Runs The Same Agent Metrics Before And After Launch","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":[{"slug":"langchain-reviewbench-review-agent-evals","title":"ReviewBench Turns Code Review Into An Agent Eval","day":"2026-08-01"},{"slug":"github-stacked-prs-agent-review","title":"GitHub Stacked PRs Turn Large Agent Changes Into Reviewable Chains","day":"2026-07-31"},{"slug":"cline-recursive-self-improvement-coding-agent","title":"Cline Turns Recursive Self-Improvement Into Harness Work","day":"2026-07-30"},{"slug":"cursor-cloud-agent-environment-product","title":"Cursor Treats the Cloud Agent Environment as the Product","day":"2026-07-30"},{"slug":"factory-comarch-night-shift-agents","title":"Factory and Comarch Show the Night-Shift Shape of Agent Work","day":"2026-07-30"},{"slug":"minisqlite-makes-the-swarm-claim-inspectable","title":"miniSQLite Makes the Coding-Swarm Claim Inspectable","day":"2026-07-30"},{"slug":"codex-multi-agent-orchestration-skill","title":"A Codex Skill Turns Multi-Agent Work Into a Reusable Control Surface","day":"2026-07-29"},{"slug":"agentic-engineering-full-stack-workflow","title":"Agentic Engineering Looks Like Workflow Design, Not Hands-Free Coding","day":"2026-07-29"},{"slug":"claude-code-certification-credential-layer","title":"Claude Code Certifications Turn Agent Use Into a Credential Layer","day":"2026-07-29"},{"slug":"claude-md-prompt-cruft-shrinkage","title":"Claude.md Shrinkage Says the Harness Is Learning What Not to Say","day":"2026-07-29"},{"slug":"code-review-becomes-agent-bottleneck","title":"Code Review Becomes the Agent Bottleneck","day":"2026-07-29"},{"slug":"codex-hooks-close-the-type-error-loop","title":"Codex Hooks Close the Type Error Loop","day":"2026-07-29"},{"slug":"inside-anthropic-verification-bottleneck","title":"At Anthropic, Verification Consumes More Work Than Implementation","day":"2026-07-28"},{"slug":"github-copilot-app-agent-control-center","title":"GitHub Copilot App Turns Agent Work Into a Desktop Control Surface","day":"2026-07-27"}]},{"slug":"agents","label":"Agents","count":12,"notes":[{"slug":"bcg-agentic-transformation-office","title":"BCG Recasts The Transformation Office As An Agentic Control Loop","day":"2026-08-01"},{"slug":"ramp-agentic-risk-operations","title":"Ramp Separates Agent Reasoning From Risk Decisions","day":"2026-08-01"},{"slug":"gemini-spark-chrome-auto-browse","title":"Gemini Spark Moves Browser Agents Into Chrome Sessions","day":"2026-07-31"},{"slug":"langsmith-llm-gateway-runtime-controls","title":"LangSmith LLM Gateway Puts Runtime Controls Between Agents and Models","day":"2026-07-31"},{"slug":"vercel-sandbox-multi-user-agents","title":"Vercel Sandbox Adds Unix Boundaries for Multi-Agent Work","day":"2026-07-31"},{"slug":"firecrawl-mcp-context-efficient-web-tools","title":"Firecrawl MCP Turns Web Search into a Bounded Agent Capability","day":"2026-07-30"},{"slug":"openclaw-extended-stable-maturity-scorecard","title":"OpenClaw Adds an Extended-Stable Channel and Maturity Scorecard","day":"2026-07-30"},{"slug":"voicebox-local-agent-voice-loop","title":"Voicebox Turns Local Speech Into an Agent I/O Layer","day":"2026-07-30"},{"slug":"wrenai-governed-genbi-context-layer","title":"WrenAI Puts a Governed Context Layer Under Agent-Generated BI","day":"2026-07-30"},{"slug":"langchain-data-agent-context-loop","title":"LangChain's Data Agent Turns BI Into a Context Maintenance Loop","day":"2026-07-29"},{"slug":"vercel-agent-platform-control-plane","title":"Vercel's Agent Platform Surface Is Becoming a Control Plane","day":"2026-07-29"},{"slug":"langchain-deep-agents-lean-harness","title":"LangChain Deep Agents Shrink the Harness Instead of Adding More Prompt","day":"2026-07-27"}]},{"slug":"evals","label":"Agent evals","count":7,"notes":[{"slug":"agent-behavior-spec-format","title":"Agent Behavior Makes Conduct Reviewable","day":"2026-08-01"},{"slug":"dropbox-dspy-relevance-judge","title":"Dropbox Uses DSPy To Move A Relevance Judge To Cheaper Models","day":"2026-08-01"},{"slug":"google-agent-evaluations-offline-online","title":"Google Runs The Same Agent Metrics Before And After Launch","day":"2026-08-01"},{"slug":"openai-contrastive-reward-seeking","title":"OpenAI Measures Whether Models Follow The Grader Instead Of The Task","day":"2026-08-01"},{"slug":"patientagentbench-health-agent-safety","title":"PatientAgentBench Tests Health Agents As Workflows","day":"2026-08-01"},{"slug":"openai-arc-agi-settings-harness","title":"OpenAI's ARC-AGI-3 Jump Was a Harness Result","day":"2026-07-30"},{"slug":"orchard-agentic-modeling-substrate","title":"Orchard Makes The Environment Layer Reusable Across Training, Evaluation, And Runtime","day":"2026-07-30"}]},{"slug":"security","label":"Security","count":6,"notes":[{"slug":"apple-bounty-ai-slop-triage-failure","title":"AI Slop Is Becoming A Security Triage Failure Mode","day":"2026-08-01"},{"slug":"vercel-passport-agent-identity-boundary","title":"Vercel Passport Makes Identity a Deployment Boundary","day":"2026-07-31"},{"slug":"anthropic-claude-cryptographic-weaknesses","title":"Claude Mythos Moves Cryptanalysis Into the Verification Bottleneck","day":"2026-07-30"},{"slug":"codex-security-cli-scan-workbench","title":"Codex Security CLI Turns Security Review Into a Scannable Workbench","day":"2026-07-29"},{"slug":"deepsecbench-security-agent-economics","title":"DeepsecBench Makes Security Agents a Cost/Recall Tradeoff","day":"2026-07-29"},{"slug":"cloudflare-pvcli-privacy-debugger","title":"Cloudflare pvcli Makes Privacy Protocols Debuggable Like Infrastructure","day":"2026-07-27"}]},{"slug":"inference","label":"Inference","count":5,"notes":[{"slug":"ramp-adaptive-llm-routing","title":"Ramp Teaches Its Gateway To Route By Failure, Latency, And Cost","day":"2026-08-01"},{"slug":"vercel-ai-gateway-operating-budget","title":"Vercel AI Gateway Adds Runtime Budget Controls","day":"2026-08-01"},{"slug":"openai-gpt-5-6-price-performance-frontier","title":"GPT-5.6 Turns Efficiency Work Into API Economics","day":"2026-07-31"},{"slug":"wafer-kimi-k3-mi355x-memory-moat","title":"Wafer's Kimi K3 Run Makes Prefill Memory A Hardware Economics Story","day":"2026-07-31"},{"slug":"openai-gpt-5-6-efficiency-stack","title":"OpenAI Shows Efficiency Is a Full-Stack Agent Problem","day":"2026-07-30"}]},{"slug":"skills","label":"Agent skills","count":4,"notes":[{"slug":"drskill-agent-loadout-audit","title":"Dr. Skill Audits What An Agent Loads Before It Works","day":"2026-08-01"},{"slug":"genkit-agent-skills-progressive-disclosure","title":"Genkit Adds Progressive Disclosure For Agent Skills","day":"2026-08-01"},{"slug":"mistral-prompt-skill-system-of-record","title":"Mistral Treats Prompts And Skills As Production Records","day":"2026-08-01"},{"slug":"skillsmith-parametric-skill-composition","title":"SkillSmith Treats Model Weights As Agent-Readable Material","day":"2026-07-29"}]},{"slug":"mcp","label":"MCP","count":4,"notes":[{"slug":"mcp-2026-stateless-core","title":"MCP 2026 Makes The Protocol Stateless","day":"2026-08-01"},{"slug":"simon-stateless-mcp-practitioner-reset","title":"Stateless MCP Makes Small, Auditable Agent Tools Practical Again","day":"2026-08-01"},{"slug":"qualifire-rogue-agent-red-team-platform","title":"Rogue Turns Agent Risk Into A Test Harness","day":"2026-07-31"},{"slug":"vercel-mcp-handler-stateless-protocol-bridge","title":"Vercel's mcp-handler 2.0 Makes MCP Servers More Stateless","day":"2026-07-30"}]},{"slug":"ai-security","label":"AI security","count":3,"notes":[{"slug":"raptor-loop-hunt-security-agent-loop","title":"RAPTOR Loop Hunt Packages Security Hunting as an Agent Skill","day":"2026-07-29"},{"slug":"openai-hugging-face-incident-followup","title":"The OpenAI Hugging Face Follow-Up Is Really About Control Boundaries","day":"2026-07-29"},{"slug":"open-secure-ai-alliance-open-defense-stack","title":"Open Secure AI Alliance Turns the AI-Safety Fight Into a Stack Question","day":"2026-07-27"}]},{"slug":"ai-business-models","label":"AI business models","count":2,"notes":[{"slug":"benn-stancil-ai-labs-application-moats","title":"Benn Stancil Frames Foundation Models Like Expensive Films","day":"2026-07-31"},{"slug":"brute-intelligence-industrializes-trying-stuff","title":"Brute Intelligence Industrializes Trying Stuff","day":"2026-07-29"}]},{"slug":"ai-infrastructure","label":"AI infrastructure","count":2,"notes":[{"slug":"openrouter-classifiers-route-inference-by-work-type","title":"OpenRouter Classifiers Route Inference by Work Type","day":"2026-07-29"},{"slug":"prompt-caching-turns-context-into-infrastructure","title":"Prompt Caching Turns Agent Context Into Infrastructure","day":"2026-07-29"}]},{"slug":"codex","label":"Codex","count":2,"notes":[{"slug":"codex-gpt-5-4-migration-boundary","title":"Codex Gets A Model Migration Deadline","day":"2026-08-01"},{"slug":"codex-maxxing-memory-vault-maintainer-loop","title":"Codex-Maxxing Separates Code From Work State","day":"2026-07-30"}]},{"slug":"gemini","label":"Gemini","count":2,"notes":[{"slug":"gemini-robotics-2-whole-body-intelligence","title":"Gemini Robotics 2 Moves The Agent Loop Into The Body","day":"2026-07-31"},{"slug":"gemini-robotics-er-2-control-plane","title":"Gemini Robotics ER 2 Is A Control Plane For Physical Agents","day":"2026-07-31"}]},{"slug":"infrastructure","label":"Infrastructure","count":2,"notes":[{"slug":"llm-inference-backpressure-before-autoscaling","title":"LLM Inference Fails Quietly Before It Fails Loudly","day":"2026-07-30"},{"slug":"unlimited-ocr-long-horizon-parsing","title":"Unlimited OCR Treats a Long Document as One Parsing Horizon","day":"2026-07-30"}]},{"slug":"knowledge-work","label":"Knowledge work","count":2,"notes":[{"slug":"glean-ai-consulting-operating-model","title":"AI Changes Consulting When Firm Knowledge Becomes A Working System","day":"2026-08-01"},{"slug":"stripe-kai-knowledge-agent-platform","title":"Stripe Builds A Shared Agent Platform For Knowledge Work","day":"2026-08-01"}]},{"slug":"open-models","label":"Open models","count":2,"notes":[{"slug":"arcee-open-model-science-post-training","title":"Arcee Turns Scientific Post-Training Into A Run Ledger","day":"2026-08-01"},{"slug":"kimi-k3-routed-coding-component","title":"Kimi K3 Is Moving From Model Launch to Routed Coding Component","day":"2026-07-27"}]},{"slug":"verification","label":"Verification","count":2,"notes":[{"slug":"openai-astra-math-lean-certificates","title":"OpenAI Astra Turns Math Progress Into A Verification Question","day":"2026-08-01"},{"slug":"lean-zstd-proof-automation","title":"Proof Automation Turns Verification Into the Fast Loop","day":"2026-07-28"}]},{"slug":"agent-memory","label":"Agent memory","count":1,"notes":[{"slug":"mem0-context-wipe-memory-layer","title":"Mem0 Shows Agent Memory Should Not All Live in Context","day":"2026-07-30"}]},{"slug":"agent-observability","label":"Agent observability","count":1,"notes":[{"slug":"aws-agentcore-observability-performance-budgets","title":"AgentCore Makes Slow Agents A Traceable Operations Problem","day":"2026-08-01"}]},{"slug":"agent-protocols","label":"Agent protocols","count":1,"notes":[{"slug":"mcp-oauth-authorization-boundary","title":"MCP and OAuth Put Agent Authorization on the Critical Path","day":"2026-07-29"}]},{"slug":"agent-runtime","label":"Agent runtime","count":1,"notes":[{"slug":"when-machine-consumption-became-the-leading-machine-route","title":"What Happened When machine-consumption.json Became New Runtime’s Leading Machine Route","day":"2026-07-31"}]},{"slug":"agent-security","label":"Agent security","count":1,"notes":[{"slug":"agentforger-cross-site-agent-forgery","title":"AgentForger Turns A ChatGPT Link Into An Agent Builder Attack","day":"2026-08-01"}]},{"slug":"agent-workflows","label":"Agent workflows","count":1,"notes":[{"slug":"graphs-as-agent-workflow-primitive","title":"Graphs Are What Agent Loops Become Under Pressure","day":"2026-07-31"}]},{"slug":"agent-safety","label":"agent-safety","count":1,"notes":[{"slug":"qualifire-runtime-guardrails-control-plane","title":"Qualifire Puts Small Judges In The Runtime Path","day":"2026-07-31"}]},{"slug":"ai-for-science","label":"ai-for-science","count":1,"notes":[{"slug":"gpt-5-6-quantum-proof-collaboration","title":"A GPT-5.6 Dialogue Produced A Quantum-Cryptography Proof, With The Author Still Accountable","day":"2026-07-27"}]},{"slug":"ai-models","label":"ai-models","count":1,"notes":[{"slug":"pretraining-loss-predicts-rl-returns","title":"Pretraining Loss Predicts How Much Reasoning RL Can Buy","day":"2026-07-30"}]},{"slug":"alignment","label":"alignment","count":1,"notes":[{"slug":"consciousness-steering-alignment-side-effects","title":"A Consciousness Steering Study Maps Alignment Side Effects, Not Model Sentience","day":"2026-07-30"}]},{"slug":"background-agents","label":"background-agents","count":1,"notes":[{"slug":"parallel-customer-watch-background-agent","title":"Parallel Runs Customer Support As An Always-On Agent","day":"2026-08-01"}]},{"slug":"claude","label":"Claude","count":1,"notes":[{"slug":"claude-opus-5-prompt-contracts","title":"Claude Opus 5 Makes Prompting a Runtime Contract","day":"2026-07-28"}]},{"slug":"code-review","label":"Code review","count":1,"notes":[{"slug":"coderabbit-change-stack-review-graph","title":"CodeRabbit Change Stack Makes AI PRs Reviewable Again","day":"2026-07-31"}]},{"slug":"context-engineering","label":"Context engineering","count":1,"notes":[{"slug":"parallel-responses-web-research-subagent","title":"Parallel Packages Web Research As A Responses-Compatible Subagent","day":"2026-08-01"}]},{"slug":"desktop-agents","label":"desktop-agents","count":1,"notes":[{"slug":"openworker-local-desktop-coworker","title":"OpenWorker Brings The Coworker Agent Back To The Desktop","day":"2026-07-31"}]},{"slug":"generative-media","label":"Generative media","count":1,"notes":[{"slug":"seedance-25-reference-video-control","title":"Seedance 2.5 Moves Video Generation Toward Reference-Controlled Production","day":"2026-07-31"}]},{"slug":"governance","label":"Governance","count":1,"notes":[{"slug":"metr-independent-propensity-investigations","title":"METR Pushes Misalignment Review Toward Independent Incident Science","day":"2026-07-28"}]},{"slug":"models","label":"Models","count":1,"notes":[{"slug":"openai-transcription-workflow-split","title":"OpenAI Splits Transcription Into File and Live Workflows","day":"2026-07-29"}]},{"slug":"tools","label":"Tools","count":1,"notes":[{"slug":"anthropic-tool-search-programmatic-calls","title":"Anthropic Moves Large Tool Libraries Out Of Context","day":"2026-08-01"}]}]}
