Slack launches shared channels for AI coding agents

A product manager finds a bug report in a Slack channel, tags an AI coding agent and gets a dedicated code channel. The agent reads the existing conversation, screenshots and documents, proposes a fix, and lets the team inspect the result before it moves forward. Slack is making that workflow available through Slack Code, launched on August 20.

The central unit is the code channel. Its tabs separate the conversation and plan from the code diff—the exact changes made—and a live HTML preview of the output. Team members can follow the agent’s investigation, ask for a correction, pause it or bring in an engineer; after human approval, the agent can create a pull request for review and merging.

Slack Code supports Claude, Devin, GitHub Copilot and Vercel Agent at launch, with ChatGPT support planned soon. It works on every Slack plan, but access to each external agent may require its own account or subscription. Slack says it plans to open the system’s application programming interface more broadly, with possible uses beyond software, including marketing work, contract review and IT onboarding.

The idea has company-backed momentum but is not without competition. Cursor has let users mention it in Slack since 2025 and create GitHub pull requests, while OpenAI’s Codex can also be started from Slack. Microsoft announced a similar approach for Teams in June 2026, connecting chats and channels with several agents. Slack’s reported result is that more than 70% of its code channels go from an idea to a merged pull request within one day; no independent assessment of that figure is cited.

So what changes in practice? A bug report, its context and the agent’s proposed fix can stay in one visible place, where non-engineers can start the request and engineers can inspect the evidence before approval. This keeps the work visible to the team, while Slack’s permissions, audit logs and human approval requirement for important actions add controls before production changes.