OpenAI says it met its Sept 2026 goal of an automated research intern that can run multi-day research tasks under human direction, with coding agents now logging 3.1 agent-workdays per human workday. It is aiming for a full automated AI researcher by March 2028 and says it still does not know how to do aligned full RSI safely.
Generally available Oct 22 on Docker Hub and GHCR, it gives any MCP client the running Traefik Proxy or Hub config, a bundled schema catalog for the deployed release to validate routes and middlewares before suggesting them, and matching traces, metrics, and logs from Datadog, Grafana, or Dash0. It never applies changes itself and needs a licensed Traefik Hub token (product page)…
A small daemon per computer lists every live agent session on your team (who, repo, branch, busy or idle) and gives Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI and others MCP tools like list team , send message , and ask user , so one agent can ask another a question or request your sign off via a pop up. The open source engine passed 1,000 GitHub stars this week; joining needs a free Reason sign in, and you can run your own relay (docs)…
Released Oct 21, subagents now respect your custom and MCP tool approvals instead of asking twice, p and piped stdin runs follow managed Auto model selection, and replies cut off by a provider overload retry with backoff and show the request ID. It also keeps long running sandboxed commands alive past 10 seconds and makes sandboxed traffic honor NO PROXY and ALL PROXY…
Launched Oct 20, Kiro's memory captures context that comes up while you work, like a test that only passes with a local service running, and stores it locally, scoped globally or per repo and shared between the IDE (1.4.0+) and CLI (2.29.0+, local V3 sessions). It's off by default: turn it on from the IDE's Memories view or /memories in the CLI, and note recalled memories add to context usage (CLI changelog)…
Announced Oct 21 as a research preview with Microsoft Research and built into GitHub Next's Chopin planner, PDD Kit turns your plan into a Verus spec for a human to sign off, then has the Verus verifier prove the agent's Rust code satisfies it for every input, falling back to tests and computer use agents where a proof isn't possible. It's Rust only for now, with a live demo at GitHub Universe on Oct 28–29 (Automater recap)…
Spotted something we missed? Start a thread.
