Earendil’s open-source Pi coding agent hit 1.0 with MCP support—after creator Mario Zechner once dismissed the protocol—plus Codemode sandboxing, deferred tool loading, and a separate Pi Durable layer for long-running multi-conversation agents on SQLite/JSONL.
Anthropic’s Code Review (research preview, Team/Enterprise) dispatches multi agent reviewers on GitHub PRs—parallel bug hunt, false positive filtering, severity ranking—then posts an overview plus inline comments; it never approves. Internally, substantive PR comments rose from 16% to 54%; reviews average $15–25 in token usage…
Berlin’s Restate (Apache Flink co creators) closed a $20M Series A led by Singular to scale durable execution for long running AI agents—its own distributed log resumes after crashes without an external DB. Replit already runs Replit Agent on it; the raise puts Restate up against Temporal…
Grok Build’s Agent Dashboard puts every coding session on one screen so you can peek, reply, answer approvals, and dispatch new agents in parallel—including into fresh git worktrees. Open it with grok dashboard or /dashboard ( Ctrl+\ ) from any session…
Computer use is in public preview for Copilot CLI and the Copilot app on macOS and Windows: click, type, scroll, and drag across desktop apps (including GUI only tools with no API/MCP), with per app approval before control. Turn it on with /computer on or Settings → Computer Use; orgs can disable it…
OpenAI confirmed it parted ways with three researchers—at least two in safety—for mishandling sensitive company information outside established procedures, including work tied to an external model evaluation org. The company says the exits were for policy violations, not for raising safety concerns…
Spotted something we missed? Start a thread.