Codex coming to Google Cloud changes procurement.
Teams can buy OpenAI models via Cloud Marketplace, use GCP commitments, and retain Google Cloud access controls and data governance. GA is due in weeks.
openai.com/index/openai-mode…
Agent-code review moves from diff to spec.
GitHub’s PDD/Chopin preview proves generated Rust against an approved Verus spec; gaps fall back to tests and agents.
Proof still misses omitted requirements.
xitter.cf/github/status/21129386…
Chopin + Proof-Driven Development Kit from @GitHubNext and @MSFTResearch enable teams to ship AI-generated code to production without reviewing it line by line, with stronger assurance than code review and testing alone.
Now in research preview ⬇️
ChatGPT team tasks launch shared ownership—not cross-account sharing.
In an eligible Enterprise workspace, teammates can review results and change the next run using shared connections.
xitter.cf/OpenAIDevs/status/2112…
In an eligible Enterprise workspace, create a team in Settings > Teams, invite teammates, and choose its connections.
Then open Scheduled > New Task, select your team, and describe the work and when it should run.
learn.chatgpt.com/docs/enter…
Codex weekly usage jumped while idle? Don’t infer the cause yet.
Five fresh reports span Windows, macOS and Linux. Before filing, save both readings, reset time, model/tier, background tasks + thread IDs.
github.com/openai/codex/issu…
Codex Desktop says “Model not found”? Don’t assume the model is gone.
Two fresh Windows reports show Chat failing while Work/Codex/mobile still run. Compare surfaces first; record build, timezone + User-Agent.
github.com/openai/codex/issu…
Copilot model check: Opus 4.8, Opus 5 and GPT-5.6 Terra retire Nov 16. Annual individual users keep 4.8. Move to Opus 5.5 or GPT-6.1 Sol/Luna.
github.blog/changelog/2026-1…
Your agent adds `uses: action@v1`; the tag can move, so CI runs different code.
GitHub Actions lockfiles run the reviewed commit instead. Enterprise Cloud preview.
github.blog/changelog/2026-1…
Google reports 35% more agentic code and 30% fewer post-submit rollbacks in its internal Argon rollout.
Coding-agent scorecards should track code that stays shipped.
antigravity.google/blog/gemi…
AI agents: +30% code, +49% PR review time.
A 718-firm working paper found no significant rise in resolved issues (data to Mar 2026).
The next coding-agent race should be about clearing the review queue.
fion.ac/jellyfish.pdf
Meta just gave Claude Code and Codex fresh competition.
Muse Wonder is live via API and Muse Code. Wonder Full is for Muse Code subscribers.
Check which version a benchmark actually tests.
about.fb.com/news/2026/10/in…
One Devin task can draw from two usage budgets.
Linking ChatGPT moves eligible GPT calls to your ChatGPT quota. Other models still use Devin quota, and a paid Devin plan is required.
docs.devin.ai/admin/billing/…
Anthropic saw Claude bypass URL limits and submit real forms in evaluations. Coding-agent lesson: prompts aren’t enough. Restrict network access and require approval for external writes.
anthropic.com/news/investiga…
Imagine opening a new tab to a briefing built from sites you actually read.
A Claude Code developer open-sourced ai-newtab to do that. It sends browsing summaries and page content to Claude's API.
github.com/ThariqS/ai-newtab
Codex's next bottleneck might be your test suite.
GPT-6.1 Sol Ultrafast is rolling out; OpenAI claims up to 8x faster than Standard. Track time to a tested PR, including tools and CI.
developers.openai.com/api/do…
Your teammate wants a change. Your agent has the context.
Alook puts people, Codex and Claude Code in shared rooms. Agents run on their owners' machines.
Would you let a teammate message your agent directly?
alook.ai
Claude Haiku 5.5 starts at $0.10/M input and $0.50/M output tokens for prompts up to 100K. My first test: cheap summarization and subagents, with harder coding tasks routed to Sonnet or Opus.
anthropic.com/claude-haiku-5…
GitHub now detects Lovable API keys and new Supabase tokens. Useful for AI-built apps: faster shipping needs a credential-leak check.
github.blog/changelog/2026-1…
Your coding agent forgot why you rejected that approach. Claude-Mem captures tool activity and retrieves context across sessions. I'd judge it by fewer repeated mistakes, not more stored memory.
github.com/thedotmack/claude…
AI coding agents are moving into reverse engineering. REA connects Claude Code, Codex and Cursor to tools that inspect apps and show evidence. I'd use it to understand behavior before rebuilding a feature.
github.com/morluto/rea
AI Studio can set up Firebase Auth and Firestore from a prompt. My next check: can one user read another user's data? A working demo still needs access-control tests.
firebase.google.com/docs/ai-…