What happened
On August 20, 2026, Slack published a blog post titled "Slack Code: Where Your Team and Agents Build Together," by Katie Steigman and Claire Bowman. Slack describes it as a dedicated channel type built around a single project, not a general assistant thread bolted onto an existing channel. Inside a code channel, teammates and connected AI agents work side by side: edits show up as code diffs, and changes to web pages or interfaces render as live HTML previews, so teammates watch the work happen instead of waiting for a finished pull request.
Slack says teammates can steer the agent mid-task and approve or request changes before code ships, not just review it afterward. When a project wraps up, the channel automatically archives but stays searchable, serving as an audit log of what was asked, what the agent did, and who approved it.
Slack lists Anthropic, Cognition, GitHub, OpenAI, and Vercel as launch partners; code channels are live today for Claude, Devin, GitHub Copilot, and Vercel, with ChatGPT support listed as coming soon. This differs from Slack's earlier Claude Tag integration from June 2026, which brought Claude into existing conversations as an assistant; Slack Code instead centers on project-scoped channels for multiple agents, with visible diffs, live previews, and structured approval steps.
The Verge independently confirmed the dedicated channels, code-change comparisons, live HTML previews, and automatic archiving with a searchable log. It also reports Slack Code is available on any plan; that detail comes from its reporting and was not separately checked against Slack's pricing pages.
Why it matters
For teams experimenting with AI coding agents, the appeal Slack describes is visibility: instead of an agent working in a separate tool and reporting back with a summary, the diffs and previews show up where the team already talks, and a teammate can intervene mid-task rather than only afterward. The auto-archive and audit trail could make it easier to reconstruct what changed and who signed off, though how that holds up across many concurrent channels has not been independently tested here.
Slack also says code channels inherit a workspace's existing permissions and admin controls. That is Slack's own description, not an independently verified security guarantee, and it says nothing about whether a given agent integration is scoped tightly enough for a specific codebase. Authorization settings, and how narrowly each agent's access is scoped, remain decisions a workspace admin must make deliberately; inheriting existing controls does not substitute for reviewing what they actually permit.
What to do next
- Before connecting Claude, Devin, GitHub Copilot, or a Vercel integration to a code channel, review repository scope, secrets handling, and approval controls rather than assuming existing settings already fit agent access.
- Confirm in your own admin console what inherited permissions actually grant a connected agent, rather than relying on Slack's general description.
- Keep a human reviewer in the approval step Slack describes; treat an agent's diff or live preview as something to check, not something ready to ship on its own.
- If your team relies on ChatGPT for coding, note Slack lists that as coming soon, not available, and check Slack's documentation before assuming otherwise.
This briefing is based on Slack's official blog post dated August 20, 2026, and independent reporting from The Verge dated the same day. Claims about how code channels inherit permissions, and any undemonstrated details, are attributed to Slack and unverified beyond what these two publishers report.