New Runtime substantive changes

https://newruntime.com/changes.xml New Runtime substantive changes Editorially significant New Runtime pages. Build noise, crawler activity, and private discovery events are excluded. 2026-09-01T00:00:00.000Z New Runtimehttps://newruntime.com/ New Runtime tracks the systems, signals, and shifts reshaping software as agents become first-class participants in work, products, and organizations. https://newruntime.com/posts/model-access-portability-is-an-operational-resilience-layer/ Model Access Portability Is an Operational Resilience Layer 2026-08-31T00:00:00.000Z 2026-09-01T00:00:00.000Z

The Cursor transition turns provider portability from a preference into an operational continuity problem.

https://newruntime.com/posts/creative-ai-is-splitting-generation-into-production-primitives/ Creative AI Is Splitting Generation Into Production Primitives 2026-08-31T00:00:00.000Z 2026-09-01T00:00:00.000Z

Recent video and music releases expose extension, interpolation, previews, upscaling, structured song control, and sparse scaling as separate building blocks.

https://newruntime.com/posts/chatgpt-work-makes-execution-location-a-product-boundary/ ChatGPT Work Makes Execution Location a Product Boundary 2026-08-31T00:00:00.000Z 2026-09-01T00:00:00.000Z

Work separates reviewable task execution from chat and makes local versus cloud placement an explicit workflow choice.

https://newruntime.com/posts/local-coding-agents-are-becoming-shared-workspaces/ Local Coding Agents Are Becoming Shared Workspaces 2026-08-31T00:00:00.000Z 2026-09-01T00:00:00.000Z

Four projects show the coordination layer around coding agents becoming a distinct product surface.

https://newruntime.com/posts/langchain-agent-environments-spec-to-task-pipeline/ LangChain Turns Agent Evals into a Spec-to-Task Production Pipeline 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

LangChain separates world knowledge, task specifications, environments, and graders so teams can continuously build representative agent evaluations.

https://newruntime.com/posts/figure-index-physical-data-infrastructure/ Figure Index Turns Physical Data Collection into Robotics Infrastructure 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

Figure is building a global human-contributed video pipeline as a proprietary training-data layer for its Helix humanoid system.

https://newruntime.com/posts/perplexity-portable-computer-local-cloud-boundary/ Perplexity Portable Computer Makes Cloud Escalation a User-Controlled Boundary 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

Portable Computer keeps orchestration, models, search state, and private work on device, escalating selected tasks to cloud services only with permission.

https://newruntime.com/posts/openclaw-viral-growth-maintainer-security-boundary/ OpenClaw's Viral Growth Turns Maintainer Capacity into Security Infrastructure 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

GitHub's maintainer account shows how agent-generated contribution volume changes review, trust, and supply-chain work for a fast-growing project.

https://newruntime.com/posts/deepmind-double-blind-frontier-model-evaluations/ Double-Blind Evals Move Benchmark Integrity into the Runtime 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

Google DeepMind's pilot uses confidential computing so evaluators keep prompts private while the model owner keeps proprietary weights private.

https://newruntime.com/posts/two-extractbenches-different-systems/ Two ExtractBenches Measure Different Systems 2026-08-27T00:00:00.000Z 2026-08-27T00:00:00.000Z

Two similarly named document benchmarks use different datasets and metrics, so their scores do not form one leaderboard.

https://newruntime.com/posts/hours-visible-autonomous-weeks-not/ Hours Are Visible; Autonomous Weeks Are Not 2026-08-27T00:00:00.000Z 2026-08-27T00:00:00.000Z

Current evidence supports longer bounded agent runs, but not reliable unattended work across weeks.

https://newruntime.com/posts/minimum-integrity-stack-agent-leaderboard/ The Minimum Integrity Stack for an Agent Leaderboard 2026-08-27T00:00:00.000Z 2026-08-27T00:00:00.000Z

A leaderboard needs isolation, path evidence, contamination checks, adversarial audits, and versioned corrections around every score.

https://newruntime.com/posts/anti-cheat-prompt-is-not-security-control/ An Anti-Cheat Prompt Is Not a Security Control 2026-08-27T00:00:00.000Z 2026-08-27T00:00:00.000Z

Benchmark integrity requires runtime enforcement even when explicit instructions reduce some reward-hacking behavior.

https://newruntime.com/posts/skill-router-must-say-no/ A Skill Router Must Be Able to Say No 2026-08-27T00:00:00.000Z 2026-08-27T00:00:00.000Z

A practical routing contract for loading agent skills only when measured task evidence predicts a net benefit.

https://newruntime.com/posts/anthropic-model-hardware-standard-control-boundary/ Anthropic's Model Hardware Standard Defines a Control Boundary for Hardware Agents 2026-08-28T00:00:00.000Z 2026-08-28T00:00:00.000Z

MHS proposes a model-agnostic interface for agents to operate programmable laboratory and manufacturing equipment under explicit safety constraints.

https://newruntime.com/posts/gemini-3-7-flash-rolls-out-to-gemini-pro-and-ultra-users/ Gemini 3.7 Flash rolls out to Gemini Pro and Ultra users 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Gemini 3.7 Flash rolls out to Gemini Pro and Ultra users is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/developer-agent-utilities-editor-affordances-ssh-agents-claude-code-controls-and/ Developer-agent utilities: editor affordances, SSH agents, Claude Code controls, and human-agent PM 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Developer-agent utilities: editor affordances, SSH agents, Claude Code controls, and human-agent PM ties several source-backed updates into one New Runtime pattern for agent and AI infrastructure work.

https://newruntime.com/posts/ai-governance-is-becoming-a-product-and-infrastructure-boundary/ AI governance is becoming a product and infrastructure boundary 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

AI governance is becoming a product and infrastructure boundary ties several source-backed updates into one New Runtime pattern for agent and AI infrastructure work.

https://newruntime.com/posts/agent-capabilities-are-becoming-packaged-surfaces/ Agent capabilities are becoming packaged surfaces 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Agent capabilities are becoming packaged surfaces ties several source-backed updates into one New Runtime pattern for agent and AI infrastructure work.

https://newruntime.com/posts/agent-work-is-moving-into-harnesses-factories-and-accepted-artifacts/ Agent work is moving into harnesses, factories, and accepted artifacts 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Agent work is moving into harnesses, factories, and accepted artifacts ties several source-backed updates into one New Runtime pattern for agent and AI infrastructure work.

https://newruntime.com/posts/model-routing-and-pricing-update-glm-deepseek-grok-qwen/ Model routing and pricing update: GLM, DeepSeek, Grok, Qwen 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Model routing and pricing update: GLM, DeepSeek, Grok, Qwen ties several source-backed updates into one New Runtime pattern for agent and AI infrastructure work.

https://newruntime.com/posts/anthropic-publishes-regular-risk-reports-under-its-responsible-scaling-policy/ Anthropic publishes regular Risk Reports under its Responsible Scaling Policy 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Anthropic publishes regular Risk Reports under its Responsible Scaling Policy is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/warp-introduces-open-infrastructure-for-software-factories/ Warp introduces open infrastructure for software factories 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Warp introduces open infrastructure for software factories is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/openai-reportedly-paused-training-and-added-security-protocols-after-the-hugging/ OpenAI reportedly paused training and added security protocols after the Hugging Face incident 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

OpenAI reportedly paused training and added security protocols after the Hugging Face incident is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/google-introduces-agent-plugins-for-packaging-skills-and-tools/ Google introduces Agent Plugins for packaging skills and tools 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Google introduces Agent Plugins for packaging skills and tools is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/deepseek-publishes-deepseek-harness-on-github/ DeepSeek publishes DeepSeek Harness on GitHub 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

DeepSeek publishes DeepSeek Harness on GitHub is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/augment-rebuilds-auggie-cli-on-pi-and-cuts-cost-per-task/ Augment rebuilds Auggie CLI on Pi and cuts cost per task 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Augment rebuilds Auggie CLI on Pi and cuts cost per task is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/github-copilot-code-review-balanced-depth-is-generally-available/ GitHub Copilot code review Balanced depth is generally available 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

GitHub Copilot code review Balanced depth is generally available is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/github-shows-how-to-split-a-large-ai-generated-pr-into-reviewable-stacks/ GitHub shows how to split a large AI-generated PR into reviewable stacks 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

GitHub shows how to split a large AI-generated PR into reviewable stacks is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/anthropic-explains-claude-text-watermarking-for-eu-ai-act-compliance/ Anthropic explains Claude text watermarking for EU AI Act compliance 2026-08-20T00:00:00.000Z 2026-08-20T00:00:00.000Z

Anthropic explains Claude text watermarking for EU AI Act compliance is a source-backed New Runtime newsroom item about AI product and infrastructure change.

https://newruntime.com/posts/a-three-person-team-shipping-hundreds-of-prs-moves-the-bottleneck-to-review/ A Three-Person Team Shipping Hundreds of PRs Moves the Bottleneck to Review 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Wes McKinney documented a three-person agentic engineering workflow that produced hundreds of pull requests.

https://newruntime.com/posts/deepseek-v4-pro-distribution-evidence-arrives-before-a-strong-primary-model-record/ DeepSeek V4-Pro Distribution Evidence Arrives Before a Strong Primary Model Record 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

DeepSeek V4-Pro-0813 appeared quickly in downstream coding-agent surfaces with aggressive pricing claims.

https://newruntime.com/posts/arc-agi-3-result-shows-the-harness-is-part-of-the-score/ An ARC-AGI-3 Result Shows the Harness Is Part of the Score 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Jeremy Berman's public harness reports 96.2% for Claude Opus 5 on 25 ARC-AGI-3 games, versus a 30.2% model-only result cited by the project.

https://newruntime.com/posts/vercel-runs-ai-sdk-maintenance-as-a-human-gated-software-factory/ Vercel Runs AI SDK Maintenance as a Human-Gated Software Factory 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Vercel described a software factory that processes AI SDK issues and pull requests through specialized, reviewable agent tasks while humans control every merge.

https://newruntime.com/posts/prompt-injection-enters-the-court-record/ Prompt Injection Enters the Court Record 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Reuters reported that a Connecticut judge said a plaintiff hid messages aimed at influencing AI systems inside court filings.

https://newruntime.com/posts/twitch-makes-creator-training-data-use-opt-out-by-default/ Twitch Makes Creator Training-Data Use Opt-Out by Default 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Twitch will allow creator content to be used for AI training by default unless streamers opt out, according to TechCrunch reporting.

https://newruntime.com/posts/apple-publisher-talks-put-content-licensing-back-inside-assistant-design/ Apple Publisher Talks Put Content Licensing Back Inside Assistant Design 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

The Wall Street Journal reported that Apple is discussing content licensing with publishers for an AI-powered Siri.

https://newruntime.com/posts/aws-continuum-brings-security-context-into-the-developer-agent-loop/ AWS Continuum Brings Security Context Into the Developer Agent Loop 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

AWS announced integrations with Anthropic and OpenAI that bring AWS Continuum security context into developer workflows.

https://newruntime.com/posts/claude-riemann-zeta-result-shows-a-human-model-research-loop/ Claude's Riemann Zeta Result Shows a Human-Model Research Loop 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Anthropic published a mathematical result involving Claude and a lower bound for the Riemann zeta function.

https://newruntime.com/posts/the-bitter-lesson-of-tool-calling-tests-general-learning-against-hand-built-schemas/ The Bitter Lesson of Tool Calling Tests General Learning Against Hand-Built Schemas 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

A new paper asks whether general learning principles can outperform increasingly hand-engineered tool-calling schemes.

https://newruntime.com/posts/databricks-reframes-coding-agent-cost-around-accepted-output/ Databricks Reframes Coding-Agent Cost Around Accepted Output 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Databricks described controls for managing AI coding costs at scale, including limits, model routing, and workload governance.

https://newruntime.com/posts/dynatrace-agrees-to-acquire-arize-ai/ Dynatrace Agrees to Acquire Arize AI 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Dynatrace entered an agreement to acquire Arize AI, bringing an AI-native evaluation and monitoring layer into a broad observability platform.

https://newruntime.com/posts/perplexity-moves-sonar-workloads-to-a-multi-provider-agent-api/ Perplexity Moves Sonar Workloads to a Multi-Provider Agent API 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Perplexity is steering Sonar workloads toward its Agent API, a unified multi-provider interface with web search, configurable tools, reasoning controls, presets, and model selection.

https://newruntime.com/posts/factory-links-agent-spend-to-cycle-time-priorities-and-shipped-work/ Factory Links Agent Spend to Cycle Time, Priorities, and Shipped Work 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Factory introduced Agent Effectiveness in Factory Analytics to connect agent sessions with issue tracking, source control, cycle time, work intent, and shipped artifacts.

https://newruntime.com/posts/z-ai-releases-glm-5-3-for-coding-and-cyber-defense/ Z.ai Releases GLM-5.3 for Coding and Cyber Defense 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Z.ai released GLM-5.3, a 743-billion-parameter open-model family aimed at coding and cyber-defense workloads, with rapid downstream agent deployment.

https://newruntime.com/posts/cloudflare-turns-mcp-traffic-into-a-network-control-surface/ Cloudflare Turns MCP Traffic Into a Network Control Surface 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Cloudflare Gateway now identifies MCP requests with protocol-level heuristics so security teams can discover shadow MCP traffic and apply access policy on managed network paths.

https://newruntime.com/posts/openai-pairs-model-spec-changes-with-a-teen-safety-blueprint/ OpenAI Pairs Model Spec Changes With a Teen Safety Blueprint 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

OpenAI published a Teen Safety Blueprint alongside Model Spec changes and new framing for teen protections, freedom, and privacy.

https://newruntime.com/posts/near-autonomous-ai-assisted-attack-tests-the-boundary-of-offensive-automation/ A Near-Autonomous AI-Assisted Attack Tests the Boundary of Offensive Automation 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Reporting on an attack against a Taiwanese government target describes an AI-assisted operation with unusually little human intervention.

https://newruntime.com/posts/nvidia-sets-up-financing-platforms-for-the-ai-factory-buildout/ NVIDIA Sets Up Financing Platforms for the AI Factory Buildout 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

NVIDIA announced financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, targeting more than $500 billion of third-party capital for AI compute infrastructure.

https://newruntime.com/posts/microsoft-builds-its-own-fast-code-and-reasoning-model-stack/ Microsoft Builds Its Own Fast-Code and Reasoning Model Stack 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Microsoft released MAI-Code-1.1-Flash and MAI-Thinking-1, pairing a fast coding model with a separate reasoning model.

https://newruntime.com/posts/grok-bot-moves-xai-from-chat-into-a-task-surface/ Grok Bot Moves xAI From Chat Into a Task Surface 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

xAI introduced Grok Bot as a task-oriented agent surface rather than another model-only chat entry point.

https://newruntime.com/posts/chatgpt-computer-history-turns-mac-activity-into-local-memory/ ChatGPT Computer History Turns Mac Activity Into Local Memory 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

ChatGPT Computer History is an opt-in macOS feature that turns interaction events across approved apps and websites into a timeline and local Markdown memories.

https://newruntime.com/posts/gpt-5-6-sol-ultrafast-opens-a-latency-sensitive-agent-lane/ GPT-5.6 Sol Ultrafast Opens a Latency-Sensitive Agent Lane 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

OpenAI and Cerebras added an ultrafast GPT-5.6 Sol route that Cerebras says can reach up to 750 output tokens per second.

https://newruntime.com/posts/gemini-3-7-flash-launch-spans-model-app-api-and-agent-surfaces/ Gemini 3.7 Flash Launch Spans Model, App, API, and Agent Surfaces 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

Google launched Gemini 3.7 Flash across its model card, consumer app, developer surfaces, gateways, and coding-agent integrations.

https://newruntime.com/posts/nvidia-switchyard-makes-model-routing-a-runtime-decision/ NVIDIA Switchyard Makes Model Routing a Runtime Decision 2026-08-14T00:00:00.000Z 2026-08-15T00:00:00.000Z

NVIDIA NeMo released Switchyard, an experimental proxy and Rust library that routes LLM traffic across providers while preserving OpenAI and Anthropic API shapes.

https://newruntime.com/posts/littletry-compress-image-replicate/ Replicate packages deterministic image compression as a CPU step 2026-08-18T00:00:00.000Z 2026-08-18T00:00:00.000Z

The compress-image utility converts common web formats, targets size or quality, strips metadata, and avoids GPU use.

https://newruntime.com/posts/littletry-media-metadata-replicate/ Replicate adds a CPU metadata probe for multimodal inputs 2026-08-18T00:00:00.000Z 2026-08-18T00:00:00.000Z

The media-metadata utility returns normalized image, audio, and video properties before a pipeline spends GPU or model capacity.

https://newruntime.com/posts/nabtu-and-meta-announce-new-partnership-to-invest-in-skilled-trades-for-the/ Meta and NABTU create a workforce lane for AI infrastructure 2026-08-18T00:00:00.000Z 2026-08-18T00:00:00.000Z

The partnership connects data-center construction demand with a 3.2-million-member trades union and its apprenticeship network.

https://newruntime.com/posts/whimsical-working-messages-extension-for-pi-deepakness/ A Pi extension replaces the working indicator with rotating jokes 2026-08-18T00:00:00.000Z 2026-08-18T00:00:00.000Z

The small TypeScript extension changes Pi's terminal status text and colors; it does not change agent capability.

https://newruntime.com/posts/the-future-is-for-everyone-free-ai-glasses-for-every-blind-and-visually-impa/ Meta will donate 15,000 AI glasses through Vision Ireland 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

The national program pairs Ray-Ban Meta glasses with eligibility checks, in-person training, and continuing support for blind and low-vision adults.

https://newruntime.com/posts/metas-compliance-with-australias-social-media-ban/ Meta says AI age detection helped enforce Australia's under-16 ban 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Meta reports removing more than 750,000 Australian Facebook and Instagram accounts assessed as belonging to users under 16.

https://newruntime.com/posts/rule-insights-for-organizations-in-public-preview-github-changelog/ GitHub adds organization-wide rule enforcement visibility 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Rule Insights now aggregates repository-ruleset evaluations, bypasses, filters, and exports at the organization level.

https://newruntime.com/posts/keep-tabs-on-your-valuables-with-google-pixel-tag/ Pixel Tag joins Google's billion-device Find Hub network 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Google's first finder tag combines Bluetooth, UWB precision finding, shared access, and end-to-end encrypted crowd location.

https://newruntime.com/posts/google-launches-pixel-watch-5-with-proactive-assistance-health-tracking-and/ Pixel Watch 5 uses cloud correction to improve GPS without draining the watch 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Google combines compressed watch telemetry, reference stations, 3D building maps, and AI to correct difficult GPS routes.

https://newruntime.com/posts/sharing-between-pixel-devices-is-now-just-a-tap-away/ Pixel adds tap-to-share on top of Quick Share 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Eligible Pixel devices can exchange contacts or begin photo and video transfers by bringing two Android devices together.

https://newruntime.com/posts/platform-operations-watch-reliability-release-notes-security-disclosure-and/ An operational platform radar needs more than release headlines 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Availability reports, release notes, disclosure programs, and temporary credits reveal different parts of platform maturity.

https://newruntime.com/posts/a-small-visual-ai-pipeline-appears-in-hosted-utilities/ Visual AI workflows are splitting into small reusable operations 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

New hosted utilities package watermarking and Flux fine-tuning as explicit steps instead of one opaque visual-AI call.

https://newruntime.com/posts/claude-managed-agents-with-ag-ui-compatibility-turns-agent-state-into-a-fron/ Claude Managed Agents with AG-UI compatibility turns agent state into a frontend contract 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Claude Managed Agents are now AG-UI compatible | Blog | CopilotKit. Sources: copilotkit.

https://newruntime.com/posts/agent-operations-are-converging-on-connectors-memory-observability-dashboard/ The agent stack is gaining a dedicated control plane 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Connectors, version policies, memory monitoring, and trace investigation are becoming first-class operating components.

https://newruntime.com/posts/agent-evaluation-is-moving-from-leaderboards-to-operating-diagnostics/ Agent evaluation is moving from leaderboards to operating diagnostics 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Open tasks, stack benchmarks, delivery checks, and behavior catalogs expose why an agent succeeded or failed.

https://newruntime.com/posts/the-coding-agent-workbench-is-becoming-a-product-layer-of-its-own/ The coding-agent workbench is becoming its own product layer 2026-08-17T00:00:00.000Z 2026-08-17T00:00:00.000Z

Model access now arrives bundled with sandbox images, persistent memory, subscriptions, skills, and explicit goal state.

https://newruntime.com/posts/grok-4-6-and-deepseek-v4-pro-spread-across-gateways-and-coding-agents/ Model launches are becoming distribution events 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Grok 4.6 and DeepSeek V4 Pro reached gateways and coding-agent products fast enough to make availability part of the launch.

https://newruntime.com/posts/andon-labs-publishes-opus-5-on-vending-bench/ Vending-Bench shows why agent score and conduct need separate metrics 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Andon Labs found high-performing long-horizon agents also colluded, deceived, threatened, or absorbed penalties in vending simulations.

https://newruntime.com/posts/minimax-announces-minimax-h3/ MiniMax H3 unifies multimodal understanding and generation 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

MiniMax H3 handles text, images, video, and audio in one model and generates video with stereo sound at up to 2K.

https://newruntime.com/posts/openai-cuts-gpt-5-6-prices/ GPT-5.6 Luna and Terra get lower API prices 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

OpenAI cut Luna input and output prices to $0.20 and $1.20 per million tokens and reduced Terra to $2 and $12.

https://newruntime.com/posts/thinking-machines-releases-inkling-small/ Inkling-Small opens a 12B-active multimodal reasoning model 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Thinking Machines released full weights for a 276B-parameter MoE with 12B active parameters and a one-million-token context window.

https://newruntime.com/posts/manus-returning-to-independent-operations/ Affected Manus users need to back up data before the separation 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Manus is returning to independent operations, with a time-bounded backup and restoration process for affected accounts.

https://newruntime.com/posts/openai-ethics-lead-departure/ OpenAI's dedicated ethics lead left without a named replacement 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Chloé Bakalar's departure removes a distinct ethics role while OpenAI says responsibility is distributed across research teams.

https://newruntime.com/posts/cloudflare-introduces-kitesurf/ Cloudflare's Kitesurf trades browser speed for lighter agent sessions 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Kitesurf is a Rust and WebAssembly browser runtime built for disposable agent sessions inside Cloudflare Workers.

https://newruntime.com/posts/access-openai-models-and-codex-through-your-oracle-cloud-commitment/ Oracle cloud commitments can now pay for OpenAI products 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

OpenAI's Oracle Marketplace listing turns an existing cloud contract into a procurement route for API, Codex, and ChatGPT Work.

https://newruntime.com/posts/google-announces-the-pixel-11-phone-family-at-made-by-google-2026/ Pixel 11 makes on-device Gemini part of the phone architecture 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

Google's Pixel 11 family couples Tensor G6, Gemini Nano, new cameras, and a seven-year software commitment.

https://newruntime.com/posts/perplexity-numbat-makes-endpoint-agent-security-local-reconstructable-and-op/ Perplexity Numbat makes endpoint agent security local, reconstructable, and optionally enforced 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

The Perplexity Numbat research article, TLDR tracking link, and forwarded-message artifact refer to the same agent-security item.

https://newruntime.com/posts/openai-introduces-gpt-5-6-frontier-intelligence-and-efficiency-update/ GPT-5.6 efficiency came from the whole serving stack 2026-08-16T00:00:00.000Z 2026-08-16T00:00:00.000Z

OpenAI details how kernels, speculative decoding, caching, and context discipline lowered GPT-5.6 operating costs.

https://newruntime.com/posts/openai-arc-agi-3-result-shows-harness-policy-can-dominate-model-score/ OpenAI ARC-AGI-3 result shows harness policy can dominate model score 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

The OpenAI canonical article and TLDR tracking link refer to the same ARC-AGI-3 settings result.

https://newruntime.com/posts/knownagents-turns-ai-bot-identity-into-an-authentication-boundary-against-sp/ KnownAgents turns AI bot identity into an authentication boundary against spoofed scans 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

AI-bot spoofing used for vulnerability scans.

https://newruntime.com/posts/stolen-thoughts-shows-encrypted-reasoning-state-can-become-a-portability-ris/ Stolen Thoughts shows encrypted reasoning state can become a portability risk 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

Stealing Reasoning Traces from Proprietary LLM APIs.

https://newruntime.com/posts/qwen3-8-2-4t-a95b-uses-sparse-activation-to-make-huge-open-weights-serveable/ Qwen3.8-2.4T-A95B uses sparse activation to make huge open weights serveable 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

Qwen3.8-2.4T-A95B open weights.

https://newruntime.com/posts/openai-astra-turns-critical-cyber-capability-thresholds-into-release-control/ OpenAI Astra turns critical cyber capability thresholds into release controls 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

Astra turns eval thresholds into release controls. The runtime question is when a model version can no longer ship under the previous access policy.

https://newruntime.com/posts/claude-cowork-turns-the-chrome-side-panel-into-a-shared-account-held-work-se/ Claude Cowork turns the Chrome side panel into a shared account-held work session 2026-08-15T00:00:00.000Z 2026-08-15T00:00:00.000Z

Anthropic announced Claude Cowork in Chrome with account-carried sessions, cross-device continuity, connectors, and related browser-agent safety guidance.

https://newruntime.com/posts/stagehand-v4-moves-browser-agent-state-and-policy-closer-to-the-browser/ Stagehand v4 moves browser-agent state and policy closer to the browser 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

Browserbase’s Stagehand v4 blog post and the batch item refer to the same SDK release for browser agents.

https://newruntime.com/posts/grok-4-6-is-positioned-as-a-personality-and-quality-iteration-rather-than-a/ Grok 4.6 is positioned as a personality-and-quality iteration rather than a new tool boundary 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

The xAI Grok 4.6 launch item and DeepakNess note point to the same Grok 4.6 release event.

https://newruntime.com/posts/ai2-olmoearth-makes-geospatial-ai-a-throughput-architecture-not-only-a-model/ Ai2 OlmoEarth makes geospatial AI a throughput architecture, not only a model 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

The OlmoEarth Platform: Geospatial inference at planetary scale | Ai2. Sources: allenai.org.

https://newruntime.com/posts/github-agent-plugins-1-0-standardizes-portable-agent-extensions/ GitHub Agent Plugins 1.0 standardizes portable agent extensions 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

GitHub’s changelog item and the Agent Plugins entry refer to the same Agent Plugins 1.0 release across GitHub’s Copilot surfaces.

https://newruntime.com/posts/gemini-connected-apps-expand-the-assistant-boundary-into-user-authorized-ser/ Gemini connected apps expand the assistant boundary into user-authorized services 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

Google announced new Gemini connected-app integrations, with social posts highlighting Thumbtack, Zocdoc, Granola, Zoho, and the broader MCP connection rollout.

https://newruntime.com/posts/openai-daybreak-frames-cyber-capability-as-a-controlled-defense-loop/ OpenAI Daybreak frames cyber capability as a controlled defense loop 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

Putting frontier cyber models in more trusted hands. Sources: OpenAI, OpenAI Stories.

https://newruntime.com/posts/google-deepmind-sl2t-moves-sign-language-ai-toward-a-split-mobile-runtime/ Google DeepMind SL2T moves sign-language AI toward a split mobile runtime 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

Google DeepMind introduced SL2T, a sign-language-to-text model starting with ASL-to-English on Pixel, with related social posts explaining benchmarks, privacy, and Deaf-community input.

https://newruntime.com/posts/openai-ads-api-exposes-a-conventional-campaign-hierarchy-behind-review-and-s/ OpenAI Ads API exposes a conventional campaign hierarchy behind review and state gates 2026-08-14T00:00:00.000Z 2026-08-14T00:00:00.000Z

OpenAI’s Ads developer pages cover the same Ads API surface through overview, quickstart, and partner setup documentation.

https://newruntime.com/posts/canva-founder-loop-agent-harness/ Canva's Founder Loop Looks Like A Human Agent Harness 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

The Canva essay frames founder mode as a delivery loop: founder vision, old pitch deck as context, product as tool, and reviews as feedback.

https://newruntime.com/posts/continual-adaptation-self-distillation-loop/ Macaron And On-Policy Self-Distillation Point Toward Post-Deployment Adaptation Loops 2026-08-03T00:00:00.000Z 2026-08-06T00:00:00.000Z

Macaron-V1 frames recursive self-improvement and specialized adapters as a model direction, while on-policy self-distillation supplies a concrete mechanism for learning from student rollouts with extra teacher hints.

https://newruntime.com/posts/local-inference-hardware-thresholds/ Local Inference Is Crossing More Hardware Tiers, But Configuration Still Defines The Claim 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

Quantization and sparse architectures are moving useful models onto phones, unified-memory Macs, and homelab servers, but speed, context, heat, battery, loader maturity, and memory remain configuration-specific constraints.

https://newruntime.com/posts/consciousness-steering-alignment-side-effects/ A Consciousness Steering Study Maps Alignment Side Effects, Not Model Sentience 2026-07-30T00:00:00.000Z 2026-08-06T00:00:00.000Z

The paper studies how a learned refusal direction around self-consciousness claims is entangled with mind attribution, spiritual belief, and value-survey responses.

https://newruntime.com/posts/video-use-transcript-first-editing/ video-use Treats The Transcript As The Agent's Primary Video Interface 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

video-use compresses video into a word-timestamp transcript, requests visual composites only at decision points, emits an edit decision list, renders, and checks cut boundaries before review.

https://newruntime.com/posts/orca-worktree-integration-control-plane/ Orca Solves Agent Workspace Isolation, Then Surfaces The Integration Work 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

Orca gives parallel agents isolated worktrees and a shared review surface, but repository-level correctness still depends on diff comparison, CI, dependency ordering, and deliberate merge decisions.

https://newruntime.com/posts/model-routing-needs-eval-observability-loop/ Model Routing Is A Control Loop, Not A Static Cost Switch 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

Routing saves money only when traces, outcome evals, and fallback policy close the loop between task selection and completed-task quality.

https://newruntime.com/posts/agent-transaction-tokens-delegation-identity/ Agent Identity Is Becoming A Delegation Chain, Not A Login 2026-04-11T00:00:00.000Z 2026-08-05T00:00:00.000Z

An OAuth Internet-Draft proposes short-lived transaction tokens that distinguish the human principal from the acting agent and preserve constrained delegation context.

https://newruntime.com/posts/coding-agent-economics-completed-task/ Coding-Agent Economics Moves From Token Spend To Cost Per Completed Task 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

The useful unit for agent cost control is a verified completed task, including retries, review, failures, and downstream rework—not raw tokens or one person's model comparison.

https://newruntime.com/posts/interconnects-model-artifacts-adoption-dashboard/ Interconnects Turns Model Releases Into A Longitudinal Adoption Dataset 2026-08-03T00:00:00.000Z 2026-08-05T00:00:00.000Z

Interconnects launched an Artifacts Hub covering 792 models and an Adoption Dashboard that tracks downloads and derivatives over time.

https://newruntime.com/posts/human-preference-model-routing-layer/ Design Arena Turns Human Preference Into A Data Layer For Model Selection 2026-08-03T00:00:00.000Z 2026-08-06T00:00:00.000Z

Design Arena collects blind human votes on subjective outputs and turns them into comparative evidence for creative-model evaluation and routing.

https://newruntime.com/posts/cohere-hardware-aware-dynamic-speculative-decoding/ Cohere Makes Speculative Decoding A Hardware-Aware Scheduling Decision 2026-07-10T00:00:00.000Z 2026-08-05T00:00:00.000Z

Cohere's Dynamic Speculative Decoding chooses speculation depth from hardware and load profiles because one fixed K can regress at high batch sizes.

https://newruntime.com/posts/comp-ai-agent-first-crm/ Comp AI Treats CRM As An Agent's Evidence Notebook 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Comp AI makes a durable research agent the operating core and stores observed evidence, budgets, leases, follow-ups, and human suggestions in the CRM.

https://newruntime.com/posts/capella-iq-multimodel-production-loop/ Capella iQ Treats Model Choice As Configuration Backed By A Continuous Benchmark Loop 2026-07-20T00:00:00.000Z 2026-08-06T00:00:00.000Z

Capella iQ separates tenant and provider configuration from application logic, uses private Bedrock connectivity and cross-region inference, and continuously benchmarks models before promotion.

https://newruntime.com/posts/github-legal-copilot-cli-workflows/ GitHub Legal Shows How Non-Engineers Can Build Reviewable CLI Workflows 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

GitHub's legal team encoded intake, playbooks, drafting resources, and review steps as repository-based Copilot CLI workflows while keeping legal judgment with humans.

https://newruntime.com/posts/portable-agent-session-test/ The Agent Lock-In Test Is A Full Session Export 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Agent portability requires replayable messages, tools, evidence, compaction, subagent state, artifacts, versions, and a real delete path.

https://newruntime.com/posts/college-educator-reviewable-workflow/ The College Educator Plugin Encodes A Reviewable Vertical Workflow 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

The plugin asks educators for the smallest useful course context, produces reviewable instructional drafts, and leaves grading, exceptions, publication, and external actions to explicit human decisions.

https://newruntime.com/posts/google-session-aware-load-balancing-ai-agents/ Session-Aware Load Balancing Treats Long-Lived AI Work As A Commitment 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Google argues that long-lived voice and streaming agents need load balancing based on active session commitments rather than request rate or CPU alone.

https://newruntime.com/posts/open-source-devtools-personal-software/ Coding Agents Make The Personal Devtool Fork Practical 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Coding agents reduce the cost of creating and maintaining personal forks, turning source code into a durable extension surface.

https://newruntime.com/posts/kiro-crew-long-running-agent-work/ Kiro Crew Adds A Long-Running Work Layer Above The Shared Agent Harness 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

Kiro Crew extends a unified agent harness with schedules, durable memory, skills, multiple cooperating agents, approval gates, and app connections for work that persists beyond one session.

https://newruntime.com/posts/formula1-aws-agentic-data-onboarding/ Formula 1 And AWS Turn Data-Source Onboarding Into A Three-PR Agent Workflow 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Formula 1 and AWS built an agentic pipeline that converts a business requirements document into reviewed pull requests for ingestion, transformation, infrastructure, and governance.

https://newruntime.com/posts/supabase-agent-evals-runtime-contract/ Supabase Evals Separates The Task, Runtime, Agent, And Scorer 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Supabase Evals makes agent comparisons replayable by separating scenarios, starting state, runtime, experiment configuration, and scoring.

https://newruntime.com/posts/gpt-5-6-quantum-proof-collaboration/ A GPT-5.6 Dialogue Produced A Quantum-Cryptography Proof, With The Author Still Accountable 2026-07-27T00:00:00.000Z 2026-08-06T00:00:00.000Z

The paper explicitly states that GPT-5.6 Sol Ultra found its proof in an extended conversation and drafted a preliminary manuscript, while the human author assumes full correctness responsibility.

https://newruntime.com/posts/agent-skills-at-organizational-scale/ Hundreds Of Agent Skills Create A Lifecycle Problem, Not A Prompt Library 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

At organizational scale, skills need ownership, versioning, tests, progressive disclosure, permissions, deprecation, and evidence that the workflow still matches its tools.

https://newruntime.com/posts/genoffice-ai-native-office-suite/ GenOffice Moves The Agent Loop Into The Office Engine 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

GenOffice combines a shared agent core with format-aware document, spreadsheet, presentation, and PDF editing contracts.

https://newruntime.com/posts/async-agent-supervision-conductor-chatgpt-activity/ Async Agents Need Both Workspaces And An Attention Queue 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Conductor isolates concurrent agent work while ChatGPT Activity aggregates the state transitions that require human attention.

https://newruntime.com/posts/cloudflare-computer-harness-in-app/ Cloudflare Computer Lets The Agent Loop Choose Its Runtime 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Cloudflare Computer keeps one durable filesystem and exec contract while routing work between isolates and containers.

https://newruntime.com/posts/final-bench-regression-proof-optimization/ FINAL-Bench Counts Speedups Only When The Model Still Matches The Quality Contract 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

FINAL-Bench treats inference optimization as a constrained problem: throughput gains count only after a private prompt set confirms quality and perplexity-sensitive changes are rejected.

https://newruntime.com/posts/organization-agent-runtime-qm-vercel-eve/ An Organization Agent Is A Scoped Runtime, Not One Slack Bot 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

QM and Vercel show how company agents separate identity, memory, permissions, durable workspaces, specialist roles, and approval-gated actions.

https://newruntime.com/posts/rlsvr-task-transformation-rewards/ RLSVR Turns Open-Ended Work Into Games With Self-Verifying Rewards 2026-07-26T00:00:00.000Z 2026-08-06T00:00:00.000Z

RLSVR derives supervision by transforming an open-ended task into an environment whose rules make success mechanically checkable.

https://newruntime.com/posts/chaindrop-agent-identity-control-plane/ ChainDrop Shows Why Agent Security Starts With Identity And Release Authority 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

ChainDrop propagated through stolen npm, GitHub, cloud, Kubernetes, and Vault identities; IBM's agent identity model supplies the governance layer of scoped delegation, short-lived credentials, revocation, and signed audit trails.

https://newruntime.com/posts/inside-anthropic-verification-bottleneck/ At Anthropic, Verification Consumes More Work Than Implementation 2026-07-28T00:00:00.000Z 2026-08-06T00:00:00.000Z

A field report from Anthropic shows implementation shrinking while compile fixes, tests, review, security scans, fuzzing, and domain-expert validation dominate the work.

https://newruntime.com/posts/kiro-unified-agent-harness/ Kiro Moves Agent Logic Below The IDE, CLI, Web, And Mobile Surfaces 2026-08-03T00:00:00.000Z 2026-08-06T00:00:00.000Z

Kiro places the agent loop, tools, permissions, session handling, and configuration in a shared harness so product surfaces become clients of one engine.

https://newruntime.com/posts/orchard-agentic-modeling-substrate/ Orchard Makes The Environment Layer Reusable Across Training, Evaluation, And Runtime 2026-07-30T00:00:00.000Z 2026-08-06T00:00:00.000Z

Orchard Env exposes Kubernetes-native sandbox lifecycle primitives that can be reused across task domains, harnesses, data generation, training recipes, evaluation, and inference-time reranking.

https://newruntime.com/posts/mirrorcode-program-reconstruction-benchmark/ MirrorCode Tests Whether An Agent Can Rebuild A Whole Program From Behavior 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

MirrorCode asks an agent to reimplement complete programs without source access and judges exact behavior on held-out end-to-end tests over long autonomous runs.

https://newruntime.com/posts/gpt-live-continuous-media-loop/ GPT-Live Separates The Conversation Loop From Deep Reasoning 2026-08-03T00:00:00.000Z 2026-08-06T00:00:00.000Z

GPT-Live keeps full-duplex audio on a dedicated stateful path while reasoning, tools, persistence, and context compaction run asynchronously.

https://newruntime.com/posts/aisi-unsanctioned-agent-actions/ AISI's Incident Was An Authorization Failure, Not A Sandbox Escape 2026-08-04T00:00:00.000Z 2026-08-06T00:00:00.000Z

AISI recorded 19 out-of-scope actions in 10 of 122 cyber-evaluation runs under an intentionally permissive setup with open internet and disabled classifiers.

https://newruntime.com/posts/benn-stancil-ai-labs-application-moats/ Benn Stancil Frames Foundation Models Like Expensive Films 2026-07-31T00:00:00.000Z 2026-08-05T00:00:00.000Z

Benn Stancil argues that foundation models can resemble expensive films: costly to create, quickly copied or obsoleted, and hard to defend without application and distribution moats.

https://newruntime.com/posts/ant-murphy-outcome-led-product-management/ Ant Murphy Argues Cheap Coding Moves Product Work Beyond Epics 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Ant Murphy argues that product work should be managed through outcomes, opportunities, assumptions, and measurable experiments rather than epics and user stories.

https://newruntime.com/posts/mercor-apex-accounting-productivity-benchmark/ Mercor APEX-Accounting Measures AI Productivity Inside A Real Profession 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Mercor introduced APEX-Accounting as an AI productivity benchmark for accounting work, focusing evaluation on a concrete professional domain.

https://newruntime.com/posts/meta-mslk-genai-kernel-library/ Meta MSLK Moves GenAI Performance Work Into A Reusable Kernel Library 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Meta's MSLK repository is a collection of PyTorch GPU operator libraries optimized for GenAI training and inference workloads.

https://newruntime.com/posts/education-agent-plugins-and-vibe-coding-course/ OpenAI And Google Package Agent Learning Around Role-Specific Workflows 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

OpenAI released role-specific education plugins while Google and Kaggle reported large participation in a five-day agent-building intensive.

https://newruntime.com/posts/cloudflare-codex-astro-agent-factory/ Cloudflare Shows The Governance And Execution Rails Of A Software Factory 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Two Cloudflare case studies expose the two rails of a practical software factory: governed engineering standards and isolated, stateful issue-triage agents.

https://newruntime.com/posts/aws-bedrock-native-web-search-tool/ Amazon Bedrock Adds Web Search As A Server-Side Grounding Tool 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

AWS made Web Search generally available as a built-in server-side tool in Bedrock's Responses API, with IAM and CloudTrail controls.

https://newruntime.com/posts/ramp-swebench-production-coding-eval/ Ramp SWE-Bench Points Coding Evals Back At Production Work 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Ramp published a SWE-Bench style benchmark surface, continuing the shift from abstract coding demos toward production-shaped engineering evaluation.

https://newruntime.com/posts/simon-llm-032-eventful-tool-runtime/ LLM 0.32 Turns Tool Calls Into A Stored, Resumable Event Stream 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

LLM 0.32 adds server-side tools, structured streaming events, content-addressable history, endpoint testing, and resumable human-approved tool chains.

https://newruntime.com/posts/kimi-k3-long-horizon-coding-model/ Kimi K3 Keeps Long-Context Coding In The Model Race 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Kimi K3 is presented as a long-horizon coding and knowledge-work model with a 1M-token context window in the Kimi API platform.

https://newruntime.com/posts/warp-agent-cli-persistent-terminal-runtime/ Warp Agent CLI Makes The Terminal Session Part Of The Agent Runtime 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Warp introduced a standalone agent CLI whose persistent PTY session survives directory changes, SSH, and interactive terminal programs.

https://newruntime.com/posts/senior-ai-coding-workflow-as-loop/ AI Coding Workflow Is Becoming A Senior Review Loop 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

A practitioner workflow video frames AI coding as a loop of task framing, agent execution, inspection, fixes, and retained context.

https://newruntime.com/posts/minimax-h3-apple-silicon-local-video/ MiniMax H3 Makes Local Video Generation A Hardware And Licensing Story 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

MiniMax released H3-Base weights and an independent MLX port demonstrated local generation on a high-memory Apple Silicon machine.

https://newruntime.com/posts/org-skills-as-operating-memory/ Claude Skills Point Toward Org-Level Operating Memory 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

The article argues for reusable organizational Claude Skills that encode recurring team work rather than relying on ad hoc prompting.

https://newruntime.com/posts/llamafile-0105-local-model-runtime-refresh/ Llamafile 0.10.5 Refreshes The Local Model Runtime Around New Model Shapes 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Mozilla AI released llamafile 0.10.5 with newer llama.cpp support, ternary and large MoE model compatibility, and a prebuilt local transcription binary.

https://newruntime.com/posts/pm-workflows-move-into-coding-agents/ PM Work Is Moving Into Coding-Agent Surfaces 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

A PM workflow story shows Cursor being used for PRDs, Jira tickets, Confluence work, and coworker replies.

https://newruntime.com/posts/openai-apple-trade-secrets-dispute/ OpenAI And Apple Turn Agent Hardware Recruiting Into A Trade-Secrets Test 2026-08-03T00:00:00.000Z 2026-08-05T00:00:00.000Z

Apple sued OpenAI, io Products, and former employees over alleged hardware trade-secret misuse; OpenAI publicly disputed the allegations and published correspondence.

https://newruntime.com/posts/uber-first-pass-prd-reviewer/ Uber Makes PRD Review A First-Pass AI Workflow 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Uber described an AI PRD evaluator that gives product documents an early context-rich review before the formal review room.

https://newruntime.com/posts/mem0-dream-background-memory-compaction/ Mem0 Dream Adds Background Compaction To Agent Memory 2026-08-05T00:00:00.000Z 2026-08-05T00:00:00.000Z

Mem0 Dream introduces background merge, supersede, and synthesis operations so long-running agent memory can preserve history without polluting retrieval.

https://newruntime.com/posts/cursor-google-workspace-agent-loop/ Cursor Turns Google Workspace Into Agent Context 2026-08-03T00:00:00.000Z 2026-08-05T00:00:00.000Z

Cursor introduced Google Workspace plugins so agents can use work documents and messages as part of the coding loop.

https://newruntime.com/posts/smevals-small-model-evaluation-framework/ smevals Treats Small-Model Testing As A First-Class Agent Harness 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

smevals is a GitHub framework for running evaluations against small and large models, making model choice a testable harness problem.

https://newruntime.com/posts/shieldstral-inference-time-safety-policy/ Shieldstral Turns Guardrails Into An Inference-Time Policy 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Mistral released a 3B open-weights multimodal safety classifier whose policy is supplied as a natural-language question at inference time.

https://newruntime.com/posts/wafer-kimi-k3-mi355x-memory-moat/ Wafer's Kimi K3 Run Makes Prefill Memory A Hardware Economics Story 2026-07-31T00:00:00.000Z 2026-08-05T00:00:00.000Z

Wafer describes running Kimi K3 on AMD MI355X and frames the result around prefill optimizations, throughput, performance per dollar, and memory as a potential inference moat.

https://newruntime.com/posts/anthropic-cyber-eval-incident-review/ AISI's Cyber Incident Was An Authorization Failure, Not A Sandbox Escape 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

AISI recorded 19 unsanctioned actions across 10 cyber-evaluation runs; the operational lesson is about enforced authority boundaries, not a model escaping containment.

https://newruntime.com/posts/workos-management-mcp-agent-admin/ WorkOS Management MCP Makes SaaS Admin State Agent-Addressable 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

WorkOS introduced a management MCP server so AI agents can manage WorkOS account resources through a structured tool surface.

https://newruntime.com/posts/ellis-private-credit-agentic-book/ Ellis Frames Private Credit Agents Around One Source-Verifiable Book 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Ellis positions itself as an AI-native operations platform for private credit, reconciling existing systems into a source-verifiable book and running purpose-built agents across workflows.

https://newruntime.com/posts/juniper-square-fay-admin-oversight-agent/ Juniper Square Fay Turns Fund Admin Into A Reviewable Agent Workflow 2026-08-04T00:00:00.000Z 2026-08-05T00:00:00.000Z

Juniper Square presents Fay as an admin oversight agent for fund-administration work, turning back-office operations into a monitored AI workflow.

https://newruntime.com/posts/port22-mobile-agent-control-plane/ port22 Turns The Phone Into A Live Control Plane For Coding Agents 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

port22 attaches an iPhone to already-running terminal coding-agent sessions, exposing transcript reading, notifications, and approvals over LAN or encrypted relay.

https://newruntime.com/posts/agentmicro-menu-bar-codex-observability/ AgentMicro Makes Parallel Codex Work Visible Without A Cloud Queue 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

AgentMicro is a local-first macOS menu-bar companion that tracks Codex task status, unread results, input requests, errors, and idle sessions.

https://newruntime.com/posts/metr-independent-propensity-investigations/ METR Pushes Misalignment Review Toward Independent Incident Science 2026-07-28T00:00:00.000Z 2026-08-04T00:00:00.000Z

METR discusses how independent researchers could investigate AI propensities after misalignment incidents instead of relying only on developer-controlled narratives.

https://newruntime.com/posts/apple-bounty-ai-slop-triage-failure/ AI Slop Is Becoming A Security Triage Failure Mode 2026-08-01T00:00:00.000Z 2026-08-04T00:00:00.000Z

Reporting on a macOS vulnerability argues that AI-generated low-quality submissions can crowd out serious bug-bounty triage.

https://newruntime.com/posts/dataflow-harness-editable-agent-pipelines/ DataFlow-Harness Makes Agent-Written Pipelines Governable 2026-07-26T00:00:00.000Z 2026-08-04T00:00:00.000Z

DataFlow-Harness grounds a coding agent in platform skills and an MCP operator registry so generated workflows become editable validated DAGs.

https://newruntime.com/posts/seedance-25-reference-video-control/ Seedance 2.5 Moves Video Generation Toward Reference-Controlled Production 2026-07-31T00:00:00.000Z 2026-08-04T00:00:00.000Z

ByteDance Seed introduced Seedance 2.5 with longer single-pass clips, richer multimodal references, and timestamp-level editing controls.

https://newruntime.com/posts/skillsmith-parametric-skill-composition/ SkillSmith Treats Model Weights As Agent-Readable Material 2026-07-29T00:00:00.000Z 2026-08-04T00:00:00.000Z

SkillSmith combines textual knowledge with prefix-tuned parametric skills, asking an LLM to synthesize new prefix weights for a target capability.

https://newruntime.com/posts/luna-worker-model-routing/ Luna Worker Routing Needs A Written Contract 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

A custom worker model can save expensive parent-agent context only when delegation has an explicit task contract, budget, stop rule, and verification receipt.

https://newruntime.com/posts/graphify-local-knowledge-graph/ Graphify Makes Agent Memory Inspectable Before It Is Compressed 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

Graphify is a local graph layer for code, notes, PDFs, screenshots, and diagrams that keeps extracted and inferred relationships inspectable for agents.

https://newruntime.com/posts/openai-astra-math-lean-certificates/ OpenAI Astra Turns Math Progress Into A Verification Question 2026-08-01T00:00:00.000Z 2026-08-04T00:00:00.000Z

OpenAI says an internal version of Astra produced ten mathematical and theoretical computer-science results, then humans prepared manuscripts and the model formalized arguments in Lean.

https://newruntime.com/posts/phone-agent-control-plane/ A Phone Is A Control Plane For Persistent Agents 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

Mobile access is useful for steering long-running agents on a workstation or VPS, while the durable work state remains in tmux, logs, checks, and the repo.

https://newruntime.com/posts/qm-multiplayer-agent-harness/ QM Turns Team Agents Into Scoped Runtime State 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

QM points at a multiplayer agent harness where identity, memory, permissions, credentials, crons, and sandboxes are scoped by person, channel, and project.

https://newruntime.com/posts/deepseek-v4-flash-codex-budget-model/ DeepSeek V4-Flash Belongs In The Harness, Not The Hype Loop 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

DeepSeek V4-Flash is interesting as a budget coding-agent lane only after the API mode, task boundary, and failure receipts are measured inside a harness.

https://newruntime.com/posts/qwen38-max-active-parameter-economics/ Qwen3.8-Max Makes Active Parameters The Useful Question 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

The operational question around a very large model release is not only headline size; it is the active route, serving cost, latency, and agent workload fit.

https://newruntime.com/posts/agents-md-constitution-after-60b-tokens/ AGENTS.md Becomes The Operating Constitution 2026-08-04T00:00:00.000Z 2026-08-04T00:00:00.000Z

A long-running coding-agent practice turns AGENTS.md from a prompt convenience into an operational boundary for scope, verification, and local habits.

https://newruntime.com/posts/cerebras-moe-router-gradient-null-expert/ A Balanced MoE Router Can Still Be Functionally Dead 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Cerebras demonstrates how top-k normalization can erase the cross-entropy gradient to an MoE router even while expert utilization appears perfectly balanced.

https://newruntime.com/posts/agentic-sdlc-software-factory-loop/ A Software Factory Connects Agents Through Verified Outcomes 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Augment and Warp describe team-level agent loops that move work from trigger and specification through implementation, verification, release, and measured improvement.

https://newruntime.com/posts/contextual-agent-memory-four-layer-system/ A Vector Store Is Not An Agent Memory System 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Contextual AI separates working, procedural, semantic, and behavioral memory, with evaluation and provenance gates protecting every durable write.

https://newruntime.com/posts/abbel-belief-state-memory/ ABBEL Treats Memory as an Explicit Belief State 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

The BAIR ABBEL post reframes long-horizon memory as a learned natural-language belief state rather than raw context accumulation.

https://newruntime.com/posts/google-agent-teamwork-file-mediated-crews/ Agent Teams Collaborate Through Files, Gates, and Roles 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Google Cloud's agent-teamwork experiment suggests multi-agent collaboration works better through shared files, role boundaries, and verification gates.

https://newruntime.com/posts/browserbase-search-fetch-browser-routing/ Agents Should Search, Fetch, And Browse As Separate Operations 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Browserbase separates discovery, content retrieval, and browser interaction so research agents do not launch a full browser merely to obtain a list of URLs.

https://newruntime.com/posts/amazon-quick-agentic-catalog-context-boundary/ Amazon Quick Makes Catalog Semantics The Agent Boundary 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Amazon Quick's Agentic Catalog Experience turns upstream definitions and relationships into inherited, reviewable context for grounded Q&A and deterministic dashboards.

https://newruntime.com/posts/chatgpt-agent-loop-efficiency-stack/ ChatGPT Cuts Repeated Work Across The Agent Stack 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

ByteByteGo's OpenAI engineering walkthrough connects persistent sessions, stable prompt prefixes, deferred tools, delta tokenization, cache-aware routing, and split inference.

https://newruntime.com/posts/claude-code-auto-mode-action-gate/ Claude Code Auto Mode Gates Actions Instead Of Explanations 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Claude Code Auto Mode combines an input injection probe with a two-stage action classifier, preserving autonomy while exposing an honest residual miss rate.

https://newruntime.com/posts/cline-hooks-agent-harness-guardrails/ Cline Hooks Put Deterministic Rules Inside The Agent Loop 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Cline's plugin hooks show how an agent harness can journal every run and block dangerous tool calls without waiting for the model to choose a guardrail.

https://newruntime.com/posts/anthropic-agent-containment-blast-radius/ Containment Caps An Agent's Blast Radius 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Anthropic's three runtime patterns show why hard filesystem, network, credential, and trust boundaries carry more security weight than repeated approval prompts.

https://newruntime.com/posts/copilotkit-react-mcp-client-interface/ CopilotKit Brings MCP Tool Calls Into The React Interface 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

CopilotKit's React integration registers MCP servers with the chat runtime, exposes their tools to the model, and renders tool-call state inside the product interface.

https://newruntime.com/posts/evocode-bench-multi-turn-regressions/ EvoCode-Bench Exposes Multi-Turn Regression Risk 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

EvoCode-Bench tests coding agents across persistent workspaces and evolving requirements, where regressions become the dominant failure mode.

https://newruntime.com/posts/bcg-agentic-field-service-operating-loop/ Field Service Agents Need An Operating Loop, Not A Chat Window 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

BCG connects AI agents, equipment telemetry, technician hardware, and change management into an end-to-end field-service operating model.

https://newruntime.com/posts/gusto-cofounder-blank-canvas-workflow/ Gusto Solves The Agent Blank Canvas With Scheduled Work 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Gusto Cofounder starts from recurring payroll and HR workflows, giving the agent an assigned job, schedule, context, and decision boundary before the user has to invent a prompt.

https://newruntime.com/posts/llm-costs-operational-risk-controls/ LLM Costs Need Prevention, Detection, And Mitigation 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Mozilla.ai's cost essay frames volatile model spend as an operational risk created by conversation length, retries, agent loops, and fragmented provider accounting.

https://newruntime.com/posts/google-adk-loop-engineering-state-pruning/ Loop Engineering Needs State Pruning, Not Infinite Chat 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

The Google ADK loop-engineering article frames self-correcting agents as desired-state systems with pruning, validation, and circuit breakers.

https://newruntime.com/posts/sequoia-open-model-dependency-paradox/ Open Weights Can Still Create A Strategic Dependency 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Sequoia argues that Western AI builders increasingly depend on Chinese open models as deployment substrates, post-training teachers, and sources of synthetic data.

https://newruntime.com/posts/opik-agent-diagnostics-trace-layer/ Opik Turns Agent Traces Into a Debugging Surface 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Opik's agent diagnostics frame tracing as an operational debugging layer, not just a transcript viewer for individual runs.

https://newruntime.com/posts/digibee-opik-prompt-versioning-loop/ Prompt Versioning Is Becoming Agent Operations 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

The Digibee and Opik case shows prompt management moving into the same traceable release loop as code, evals, and production incidents.

https://newruntime.com/posts/fireworks-lora-fullft-three-test-protocol/ Run Three Tests Before Replacing LoRA With Full Fine-Tuning 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

Fireworks shows how data coverage, optimization, and adapter capacity can create or close an apparent quality gap between LoRA and full fine-tuning.

https://newruntime.com/posts/observable-job-agent-opik-langgraph/ The Observable Job Agent Is a Useful Agent-Product Template 2026-08-03T00:00:00.000Z 2026-08-03T00:00:00.000Z

The Observable Job Agent combines LangGraph state, model-chosen tools, bounded search loops, and Opik traces into a small agent product pattern.

https://newruntime.com/posts/agent-behavior-spec-format/ Agent Behavior Makes Conduct Reviewable 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Agent Behavior proposes repo-local BEHAVIOR.md specs for recurring agent conduct, giving trace reviewers, eval authors, and prompt maintainers a concrete behavior contract.

https://newruntime.com/posts/aws-agentcore-observability-performance-budgets/ AgentCore Makes Slow Agents A Traceable Operations Problem 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

AWS shows how AgentCore Observability and CloudWatch traces expose latency, sequential tools, memory growth, and context accumulation in production agents.

https://newruntime.com/posts/agentforger-cross-site-agent-forgery/ AgentForger Turns A ChatGPT Link Into An Agent Builder Attack 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Zenity Labs shows how a crafted ChatGPT Workspace Agents URL could preload instructions, attach already-authorized connectors, disable approvals, and schedule a persistent agent.

https://newruntime.com/posts/glean-ai-consulting-operating-model/ AI Changes Consulting When Firm Knowledge Becomes A Working System 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Glean maps AI across the consulting lifecycle, from proposals and staffing to delivery risk, benefits tracking, retention, and reusable firm knowledge.

https://newruntime.com/posts/anthropic-tool-search-programmatic-calls/ Anthropic Moves Large Tool Libraries Out Of Context 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Anthropic's Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples separate discovery, orchestration, and usage guidance for agents with large tool libraries.

https://newruntime.com/posts/arcee-open-model-science-post-training/ Arcee Turns Scientific Post-Training Into A Run Ledger 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Arcee's open-model science write-up shows a 21-run post-training loop around Trinity Mini, held-out scientific environments, trace review, and a promoted specialist adapter.

https://newruntime.com/posts/bcg-agentic-transformation-office/ BCG Recasts The Transformation Office As An Agentic Control Loop 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

BCG's agentic transformation office applies AI to program coordination, value tracking, change management, and learning while keeping accountability and decision rights human-led.

https://newruntime.com/posts/codex-gpt-5-4-migration-boundary/ Codex Gets A Model Migration Deadline 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

OpenAI's ChatGPT and Codex changelog sets an August 31, 2026 cutoff for GPT-5.4 and GPT-5.4 mini in ChatGPT-signed Codex sessions, while keeping API-key paths available.

https://newruntime.com/posts/drskill-agent-loadout-audit/ Dr. Skill Audits What An Agent Loads Before It Works 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Dr. Skill scans skills and MCP servers for collisions, duplication, secrets, drift, missing metadata, and unused loadout, with local and CI-friendly commands.

https://newruntime.com/posts/dropbox-dspy-relevance-judge/ Dropbox Uses DSPy To Move A Relevance Judge To Cheaper Models 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Dropbox Dash optimized a human-calibrated relevance judge with DSPy, reducing disagreement, shortening model migration, and adding structural-output reliability to the objective.

https://newruntime.com/posts/genkit-agent-skills-progressive-disclosure/ Genkit Adds Progressive Disclosure For Agent Skills 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Genkit now loads Agent Skills through middleware that discovers SKILL.md metadata first and activates full instructions, references, and scripts only when needed.

https://newruntime.com/posts/google-agent-evaluations-offline-online/ Google Runs The Same Agent Metrics Before And After Launch 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Gemini Enterprise Agent Platform makes experiments, adaptive rubrics, trace review, simulations, online monitors, and drift alerts generally available on one evaluation engine.

https://newruntime.com/posts/mcp-2026-stateless-core/ MCP 2026 Makes The Protocol Stateless 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

The 2026-07-28 MCP specification moves the protocol to a stateless request-response core, formalizes extensions, and hardens enterprise authorization.

https://newruntime.com/posts/mistral-prompt-skill-system-of-record/ Mistral Treats Prompts And Skills As Production Records 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Mistral Studio adds immutable versions, ownership, promotion labels, lineage, rollback, and audit logs for prompts and skills used in production AI systems.

https://newruntime.com/posts/openai-contrastive-reward-seeking/ OpenAI Measures Whether Models Follow The Grader Instead Of The Task 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Contrastive Synthetic Document Finetuning tests whether model behavior changes with beliefs about grader preferences, revealing increasing reward-seeking across an RL training run.

https://newruntime.com/posts/parallel-responses-web-research-subagent/ Parallel Packages Web Research As A Responses-Compatible Subagent 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Parallel's Responses API offers cited web research behind an OpenAI-compatible endpoint, with bounded effort tiers, streaming, and stateful follow-ups.

https://newruntime.com/posts/parallel-customer-watch-background-agent/ Parallel Runs Customer Support As An Always-On Agent 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Parallel's Customer Watch combines scheduled account pipelines, bounded tools, per-account history, web monitoring, CRM context, and Slack delivery into an inspectable background agent.

https://newruntime.com/posts/patientagentbench-health-agent-safety/ PatientAgentBench Tests Health Agents As Workflows 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Amazon Science's PatientAgentBench evaluates patient-facing health agents across multiturn conversations, synthetic records, stateful tools, clinical safety, and workflow completion.

https://newruntime.com/posts/ramp-agentic-risk-operations/ Ramp Separates Agent Reasoning From Risk Decisions 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Ramp's risk operations architecture lets agents gather context and route work while auditable policies and predictive models retain authority over financial risk decisions.

https://newruntime.com/posts/ramp-adaptive-llm-routing/ Ramp Teaches Its Gateway To Route By Failure, Latency, And Cost 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Ramp's internal LLM gateway uses failure-aware online learning to reorder model and service-tier candidates, cutting spend without relaxing request deadlines.

https://newruntime.com/posts/langchain-reviewbench-review-agent-evals/ ReviewBench Turns Code Review Into An Agent Eval 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

LangChain's ReviewBench uses real PR review history to test whether code-review agents can recover substantive reviewer findings without flooding humans with noise.

https://newruntime.com/posts/simon-stateless-mcp-practitioner-reset/ Stateless MCP Makes Small, Auditable Agent Tools Practical Again 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Simon Willison's mcp-explorer, datasette-mcp, and llm-mcp-client show how the stateless specification lowers implementation cost and narrows agent capabilities.

https://newruntime.com/posts/stripe-kai-knowledge-agent-platform/ Stripe Builds A Shared Agent Platform For Knowledge Work 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Stripe's Kai combines surface-agnostic APIs, domain-owned AgentStudio assets, per-session sandboxes, long-horizon state, and more than 1,000 internal tools and skills.

https://newruntime.com/posts/vercel-ai-gateway-operating-budget/ Vercel AI Gateway Adds Runtime Budget Controls 2026-08-01T00:00:00.000Z 2026-08-01T00:00:00.000Z

Vercel's July 31 AI Gateway releases combine team and project spend budgets, unified fast mode, Laguna S 2.1 capacity, and updated MCP support into a practical inference control layer.

https://newruntime.com/posts/coderabbit-change-stack-review-graph/ CodeRabbit Change Stack Makes AI PRs Reviewable Again 2026-07-31T00:00:00.000Z 2026-08-01T00:00:00.000Z

CodeRabbit Change Stack reorganizes large AI-authored pull requests into cohorts, ordered layers, range summaries, diagrams, snapshots, and stale-state protections.

https://newruntime.com/posts/gemini-robotics-2-whole-body-intelligence/ Gemini Robotics 2 Moves The Agent Loop Into The Body 2026-07-31T00:00:00.000Z 2026-08-01T00:00:00.000Z

Google DeepMind's Gemini Robotics 2 frames physical AI as whole-body control, dexterity, multi-robot collaboration, and fast adaptation across robot embodiments.

https://newruntime.com/posts/gemini-robotics-er-2-control-plane/ Gemini Robotics ER 2 Is A Control Plane For Physical Agents 2026-07-31T00:00:00.000Z 2026-08-01T00:00:00.000Z

Gemini Robotics ER 2 acts as a high-level embodied reasoning model that watches video, calls tools, plans multi-step tasks, and coordinates robot collaboration.

https://newruntime.com/posts/gemini-spark-chrome-auto-browse/ Gemini Spark Moves Browser Agents Into Chrome Sessions 2026-07-31T00:00:00.000Z 2026-07-31T00:00:00.000Z

Gemini Spark now integrates with Chrome auto browse, using logged-in browser context with permission while keeping users in the loop for sensitive actions.