SIGNALNEW DECISION API; FOUR UPGRADE EDGES|SHIPPED 4|SKIPPED 4
The one thing
01OpenAI’s Decisions API is GA. Use it for bounded choices, not open-ended work.
OpenAI made the Decisions API generally available on 22 October. A request supplies text, images or native JSON, then asks for a predicate, choice or score. OpenAI says the GPT-6 Luna-based endpoint handles up to 400 questions per request and can run up to ten times faster than Luna through the Responses API. 1
The listed price is $0.10 per million input tokens and $0.01 per million cached input tokens, with no output-token charge. Regional processing is available in the US and EU; Zero Data Retention is available to eligible customers. 1
Start with one repeated, bounded decision. Replay labeled production examples for a classifier, router or policy gate; compare accuracy, p95 latency and cost against the current path, and send low-confidence cases to the existing fallback. Keep free-form reasoning and tool execution outside this endpoint.
From the source
You only pay for input tokens: there are no output-token charges.
VERDICT · USE IT — Pilot one high-volume bounded decision against labeled production examples.
Shipped
04Breaking
AI SDK 7.0.149 changes approval and interruption handling in a patch
The patch processes only the first approval response for a tool call, forwards toolChoice: none to the Hugging Face Responses API and blocks those tool calls before local execution. It also adds interrupted to public finish-reason types. Exhaustive switches and Record<FinishReason, ...> mappings can stop compiling; add the case before upgrading, then test a dropped provider stream and conflicting approval responses. 2
MATTERS TO · AI SDK 7 users with tools, approvals, provider streams or exhaustive finish-reason handling.
nofeed.dev/issues/2026-10-23/ai-sdk-7-0-149-interrupted-approvals/Breaking
OpenAI Python 3.37.0 expands Decisions types
Live question instructions may now be either a string or structured text, image or JSON parts. Request types also accept more message roles and tool-history items. Existing strings remain supported, but code that applies string operations or exhaustively checks Decisions inputs must branch on the structured form before upgrading. 3
MATTERS TO · Python applications using Decisions or Live response models.
nofeed.dev/issues/2026-10-23/openai-python-3-37-decisions-types/Use it
Claude Code 2.1.309 repairs several sandbox and permission boundaries
The release fixes shell startup files running outside the sandbox, writes to the --mcp-config file from sandboxed commands or auto-accepted edits, a disabled MCP server reconnecting after a permission prompt, and several symlink or redirect paths around deny rules. Upgrade unattended runners, then retest custom shell startup files, MCP configuration and write-deny rules. 4
MATTERS TO · Claude Code users relying on sandboxing, deny rules, MCP policy or unattended sessions.
nofeed.dev/issues/2026-10-23/claude-code-2-1-309-sandbox-boundaries/Breaking
LangChain Core 1.7.0 changes streamed tool-call arguments
Raw tool-call arguments now arrive as tool-call-delta slices instead of block-delta snapshots. High-level .tool_calls.get() and .output remain unchanged. If your code consumes raw stream events, assemble the slices or pin below 1.7 until the consumer is updated. 5
MATTERS TO · LangChain applications that parse raw streamed tool-call events.
nofeed.dev/issues/2026-10-23/langchain-core-1-7-tool-call-deltas/The conversation
01The claim — including ours
Repository context should be designed for retrieval, not loaded as one growing file.
From the floor
- JC_builds (@JC_builds)Firsthand report · no published benchmark
The developer says repeated repository searches were consuming context, so they built a codebase map for Claude Code. The public post shows the workflow but does not publish a controlled token comparison. 6
- lombarovic on RedditFirsthand report · not independently reproduced
A developer says Claude Code wrote a new game while they maintained a
CLAUDE.md, a 279-entry decision log and visual review. They also report that thousands of tests missed motion and phone-input failures that human observation caught. 7
Our take
Store decisions with an owner, rationale, status and supersession link, then retrieve only the entries relevant to the current task. Measure a codebase map on the same task set with and without it; keep visual or behavioral acceptance checks outside the agent’s own test suite.
Both developer accounts are firsthand and were not independently reproduced; neither supplies a controlled token or quality benchmark.
Skip this
0401
Agent Skills foundation transfer The governance move does not change the current skill format or an installation decision.
02
Pydantic AI 2.60.0 Its compatibility surface is broad enough to require a dedicated migration read rather than a compressed recommendation.
03
OpenHands 1.32.0 The verified changes center on picker, menu and settings behavior without an urgent upgrade trigger.
04
Copilot global-default rollout Organization admins should inspect their policy state, but the gradual rollout is not a package migration.
Everything we saw
9191 candidates scanned · 7 used in this issue — the rest, with the reason each one was left out
The receipt. Ranking is only trustworthy if the discarded pile is visible, so here it is: everything the collectors surfaced in the window, with its signal and what we did with it.
| Item | Source | Signal | Call |
|---|---|---|---|
| @ClaudeDevs: We're donating Agent Skills to the Agentic AI Foundation (@AgenticAIFdn), the same foundation we donated MCP to. Since December 2025, ... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @RobertJBye: Here's the new Claude experience on the iPhone Duo! It's going to be amazing for multitasking, Claude Code, and creating new things. | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @JC_builds: ive been messing around with ways to make Claude Code use fewer tokens (so I stop hitting my limit so early 😅) one thing I noticed is h... | X | Firsthand or attributed public post; not independently reproduced | Conversation: firsthand codebase-map workflow. |
| @notjazii: 🚨sonnet 5.5 seems to be routing to fable 5.5 sonnet 5.5 in claude code is performing way better than opus 5.5 and knows stuff that it is... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @akshay_pachaar: Another insane Jev use case! Decision models like Jev are making it dramatically cheaper to evaluate what actually happens inside ... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @MeticulAi: How I get Claude Code to actually loop and critique its own game dev work Almost everything you see went through a build → critique → m... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @luccacerf: Vibe coding with Unreal Engine for the first time! Tripo CLI → Blender → Unreal Engine The animations and VFX are on a whole different ... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @Grummz: JUST IN: Epic Games is hosting an AI Megajam, a VIBE coding contest for Unreal. "Up to four humans" and "Unlimited Agents" in the 50k priz... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @bcherny: Agent Skills started as an experiment in Claude Code. Many across the industry have come to adopt it, so we at Anthropic found it more fi... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @theo: Harness = thing that turns an LLM into an agent Provides tools, context, capabilities etc. Claude Code and Codex are harnesses. Pi is a harn... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @svpino: People are in denial. They literally think they will wake up tomorrow and all the AI-generated code will somehow be useless, and we'll go ... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @bcherny: We’re digging into this and will report back very soon | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @supabase: Give your Supabase app its own MCP server. Your customers can use your app from ChatGPT or Claude with the account they already have. | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @nateliason: The most expensive call I ever had was with @mattshumer_. Ever since he showed me and @Avivsh his ways our token spend is up 10x. | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @swyx: imo this is the right attitude - dont blame AI for finding the bugs, just fucking lock in on clearing all the bugs asap | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @alexalbert__: One side quest that would be so much fun to try right now is running a bunch of local small businesses with AI helping you out I thi... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @hwchase17: is agent server a good name? agent os? agent? agent kernel? | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @theo: This Vivek article should clear up a lot of your specific misconceptions. You seem to think that everything interface-down is the harness. T... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @badlogicgames: recommended reading. the "harness" is hilarious. | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @Docker: Containers became the standard by solving a specific problem: making software portable and reproducible. @TheFuturumGroup frames the Kit S... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| @svpino: This is pretty smart: MemT 1.0 is a new open model that can use information from earlier interactions in later tasks. For example, it can ... | X | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. |
| Claude Code 2.1.305 to 2.1.309 (Oct 20 to 22): sed -i is now real Bash, rm auto-denied in bypass mode, ultrareview needs workspace trust, 62 bug fi... | Firsthand or attributed public post; not independently reproduced | Dropped: useful practitioner context, but no independently verified product behavior or measured migration decision. | |
| 15 years of coding by hand, then two weeks with Claude Code: crayon drawings that come alive | Firsthand or attributed public post; not independently reproduced | Conversation: firsthand decision-log and visual-review workflow. | |
| JuliusBrussee/caveman bin-v4.0.2 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| All-Hands-AI/OpenHands v1.32.0 | GitHub release | Official release | Dropped: picker and settings changes did not require an urgent migration. |
| pydantic/pydantic-ai v2.60.0 | GitHub release | Official release | Watch: broad compatibility changes warrant a dedicated migration read before upgrading. |
| akitaonrails/ai-memory v3.2.2 | GitHub release | Official release | Dropped: packaging and checksum notes did not establish a workflow change. |
| BuilderIO/agent-native @agent-native/core@0.217.0 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| anthropics/claude-code v2.1.309 | GitHub release | Official release | Shipped: sandbox and permission-boundary fixes. |
| vercel/ai ai@7.0.150 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| openai/openai-python v3.37.0 | GitHub release | Official release | Shipped: Decisions type migration. |
| akitaonrails/ai-memory v3.2.1 | GitHub release | Official release | Dropped: packaging and checksum notes did not establish a workflow change. |
| vercel/ai ai@7.0.149 | GitHub release | Official release | Shipped: approval handling and the new interrupted finish reason. |
| vercel/ai ai@6.0.313 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| strands-agents/harness-sdk harness-python/v0.3.2 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| strands-agents/harness-sdk harness-cli/v0.2.2 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| langchain-ai/langchain langchain-core==1.7.0 | GitHub release | Official release | Shipped: streamed tool-call argument migration. |
| BuilderIO/agent-native @agent-native/core@0.216.0 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| BuilderIO/agent-native @agent-native/agentkit@0.216.0 | GitHub release | Official release | Dropped: available release, but its verified notes did not outrank today’s narrower compatibility and security actions. |
| All-Hands-AI/OpenHands v1.31.0 | GitHub release | Official release | Dropped: picker and settings changes did not require an urgent migration. |
| All-Hands-AI/OpenHands v1.30.1 | GitHub release | Official release | Dropped: picker and settings changes did not require an urgent migration. |
| How Interfaze serves 50 million requests a month on Vercel | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Delays in customer support responses | Vendor update | Official changelog or status page | Dropped: incident evidence was reviewed; no continuing reader action was established. |
| Send feedback from the Vercel CLI | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| GitHub Copilot in Visual Studio — September update | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available now in AI Gateway | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| OpenAI and Poland’s NASK partner to advance AI safety and security | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Does Your Model Run on AWS Trainium? Look It Up | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Eight more TLDs now eligible for a free domain with Pro | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Custom skills, agents, and MCP servers in usage metrics API | Vendor update | Official changelog or status page | Watch: useful for Copilot organization reporting, but it does not change local runtime behavior. |
| Global default for Copilot features generally available | Vendor update | Official changelog or status page | Watch: organization admins should inspect policy state; rollout is gradual rather than a package migration. |
| New Ad Accounts failing to provision admins | Vendor update | Official changelog or status page | Dropped: incident evidence was reviewed; no continuing reader action was established. |
| Build more responsive apps with the Decisions API | Vendor update | Official changelog or status page | One Thing: bounded decision API, pricing and evaluation plan. |
| Degraded Performance of Dots on Enterprise | Vendor update | Official changelog or status page | Dropped: incident evidence was reviewed; no continuing reader action was established. |
| Meta API endpoints updated for Actions blob storage | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Public monitoring now available with data residency | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Pull requests: Saved whitespace preferences and improved split view | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Industry comparisons in the Copilot impact dashboard | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Project updates delayed | Vendor update | Official changelog or status page | Dropped: incident evidence was reviewed; no continuing reader action was established. |
| Incident with Actions Results | Vendor update | Official changelog or status page | Dropped: incident evidence was reviewed; no continuing reader action was established. |
| Ensemble forecasting with IBM Granite time-series models | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| What FDE means at OpenAI | Vendor update | Official changelog or status page | Dropped: official update, but no general install or migration action was required. |
| Every AI model has the same "controversial" take on humans | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| OpenShell – A safe, private runtime for fleets of autonomous AI agents | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies? | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| LLMs are making formal software verification feasible at a much larger scale | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| Hedra is a 3D open world game I wrote from scratch while in highschool | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| AI and the Enclosure of the Scientific Commons | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| Most Americans Say AI Is Bad for Society, Oppose More Data Centers | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| AI analysis of CT scans finds 42.9% of patients have coronary inflammation | Hacker News | Public discussion with comments | Dropped: discussion lead reviewed; it did not establish a shipped product change. |
| Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3 | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| PAUSE: Search-Free Steering of Coding Agents Before Their First Source-Code Edit | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| SWE-DIVE: Multi-Turn Coding Agent Evaluation with Dynamic User Simulation | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| From Reasoning to Action: Benchmarking Frontier AI Models on LIBERO-10 Robot Manipulation Tasks | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Automating Feature Flag Cleanup at Scale with Multi-Agent LLM Systems: An Industrial Experience at DoorDash | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| RIPPLE: Recovery Internalization through Process-Level Policy Learning and Effect-Aware Credit for Long-Horizon Agentic Coding | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| SCOPE: Compiling Natural-Language Policies into Neurosymbolic Predicate Programs | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| What the Trace Shows and What the Label Says: Auditing Behavioral Detectors in Agent Evaluation | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| The End of Consensus Research? | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Can AI Co-Design Structures? An AI Agent for Early-Stage Structural Design | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Quant-Arena: Benchmarking LLM Agents for Quantitative Trading Research | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Benchmarking Coding Agents for System-Level Performance Optimization | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Superintelligent Software Engineering: Governance That Scales with Capability | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Governance-State Corruption in Long-Term Memory for LLM Agents: A Cross-Model Stress Test | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Code Comprehensor: Hierarchical Documentation and Cross-Modal Retrieval for Large Repositories | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| UserSWE: Evaluating Coding Agents in Multi-Turn Conversations with Dynamic Users | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Secure Agentic AI and Multimodal Intelligence in 6G Communication Systems | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Low-Bit N-Gram Embeddings for Language Models | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
| Conservative Meta-Agent Search: Architecture Adaptation under Non-Stationarity | arXiv | Research paper | Dropped: research result, not a shipped tool or immediate operating change for this edition. |
EOF
End of feed. That is everything from the window worth your time.
Explore more news, analysis and practical guides on NO FEED.
Sources — each with the date it was read01 OpenAI · Decisions API generally available · 22 October 2026 (2026-10-23) · 02 Vercel AI SDK 7.0.149 release · 22 October 2026 (2026-10-23) · 03 OpenAI Python 3.37.0 release · 22 October 2026 (2026-10-23) · 04 Claude Code 2.1.309 release · 22 October 2026 (2026-10-23) · 05 LangChain Core 1.7.0 release · 22 October 2026 (2026-10-23) · 06 JC_builds · X post · 22 October 2026 (2026-10-23) · 07 lombarovic · Reddit post · 22 October 2026 (2026-10-23)