Slack Code puts coding agents in channels

Slack launched Slack Code, which gives a team its own channel for building software with an AI coding agent, on any Slack plan. Someone mentions an agent, Anthropic’s Claude Code or Cognition’s Devin among them, and the agent spins up a dedicated code channel for that project. Everyone in the channel watches the same thread the agent is working from: they can inspect each proposed diff, see a live preview when the output is HTML, feed back corrections the agent folds in, and sign off when it is done. Finished channels archive themselves and retain an audit log. Slack Code requires human sign-off on high-stakes actions like merging to production, and the launch includes agents from Anthropic, GitHub, Cognition and Vercel.

What it means for you

Most agent-coding tools are single-player: one developer, one terminal, output nobody else sees until a pull request lands. Slack’s bet is that the review and hand-off problem is really a collaboration problem, so it puts the agent’s work where the team already talks. The audit log and mandatory sign-off on production merges are the details that make this usable in an org with actual change-control expectations, rather than a demo. It lands the same week Cursor added subscriptions and isolated subagents to its cloud agents and Binance opened trading to agents through MCP — the common thread is guardrails and human oversight bolted onto autonomy. Worth a look if agent output currently disappears into individual laptops and your reviewers are flying blind until the end.