Slack Put Claude Code, Devin, and Copilot in One Channel

TL;DR

On August 20, Slack shipped Slack Code, which drags the coding agent out of your private terminal tab and into a channel your whole team can watch. Mention an agent on a project, and it spins up a temporary code channel carrying the plan, the diffs, and a live preview. High-stakes changes route to a human for approval inside the channel. When the work is done the channel archives itself and stays searchable as an audit log. Four agents are live on day one: Claude, Devin, GitHub Copilot, and Vercel. ChatGPT is listed as coming soon. Slack's one published number: over 70% of code channels open and close within a single day, idea to merged PR.

What actually shipped

Slack Code is not a model and not an IDE. It is a channel type with opinions.

The mechanics, from Slack's own announcement: you mention a coding agent in a project channel, the agent creates a dedicated code channel, and it pulls in whoever on your team needs to be there. Inside that channel you get code diffs rendered in place, live HTML previews, planning documents, and, through the Vercel integration, shareable preview links.

The approval gate is the part that separates this from a bot posting logs. Slack says high-stakes changes "route to a person for a fast approval, right in the channel." Nothing ships on the agent's own say-so.

Then the channel deletes itself from your sidebar without deleting itself from history. Slack's framing is that a finished code channel "archives itself but remains like an audit log in Slack for future reference." You get the ephemerality of a scratch branch and the paper trail of a compliance system, which is a genuinely nice trick.

Four rivals, one room

Slack lists five launch partners and four live integrations. Working today: Claude from Anthropic, Devin from Cognition, GitHub Copilot, and Vercel. ChatGPT from OpenAI is marked coming soon.

Take a second with that partner list. These are four products that spend the rest of the week trying to evict each other from your editor, and they have all agreed to file their work through the same Slack-branded approval button. Whoever owns the channel owns the workflow, and Slack just volunteered.

Slack's own feature page publishes no pricing tier. Coverage of the launch from The Next Web, SiliconANGLE, and Unite.AI all report it as available on every Slack plan, free workspaces included, with each agent's own subscription bought separately. That split matters: the container is free, the labor is not.

On the admin side, Slack's claim is that code channels inherit existing workspace permissions and controls, "so IT doesn't need to configure or audit anything new." Agents see only what their permissions already allow. That is the right default, and it is also the sentence a security reviewer is going to want to test rather than believe.

The one number Slack published

Slack says over 70% of code channels spin up and close within a single day, from idea to merged PR.

Treat that as a design statement rather than a benchmark. It is Slack's own figure, drawn from usage before general availability, with no sample size published. What it tells you is the unit of work the product was built around: a same-day task, small enough that a marketing lead can eyeball the preview before lunch. It was not built for your two-week migration.

The part your security team will ask about

Putting an agent in a shared channel means putting an agent in a room where other people type.

Anthropic's own documentation for Claude Code in Slack has carried the warning for a while: when @Claude is invoked, it is handed the conversation context, and "Claude may follow directions from other messages in the context, so users should make sure to only use Claude in trusted Slack conversations."

That sentence was written for a thread with three colleagues in it. Slack Code widens the room. The whole selling point is that a designer, a PM, and someone from marketing can all drop feedback the agent incorporates, which is exactly the same channel through which an instruction arrives.

The analogy: it is the difference between dictating to an assistant in a closed office and dictating in an open-plan one. The assistant is just as capable in both. It simply cannot always tell which voice in the room is the one that signs its paychecks.

The approval gate is the real mitigation, and it is a good one, but be precise about what it covers. It gates what gets merged. It does not gate what the agent reads on the way there, and reading is where an injected instruction lands.

Two ways @Claude can exist in your workspace

If Claude is the agent you tag, there is a migration happening underneath this launch that is easy to miss.

The earlier Claude Code in Slack runs every session under an individual user's own Claude account, against that person's repos, that person's plan limits, and that person's session history. Anthropic is retiring that version for Team and Enterprise workspaces in favor of Claude Tag, which runs @Claude as the organization's shared identity with admin-configured access. Pro and Max plans keep the old setup path, because Claude Tag is not available on individual plans.

This is not a cosmetic rename. Whose account the agent runs as determines which repositories it can reach, whose rate limits it consumes, and who is accountable when a PR lands. Anthropic lists Claude Tag as in beta for Enterprise and Team customers in Slack, with a Microsoft Teams version coming soon. Existing Slack apps and the @Claude handle survive the cutover; your account team supplies the date.

The limits worth knowing before you tag anything

Anthropic's docs for the earlier Slack integration list constraints that will shape your first week regardless of how nice the channel looks:

• GitHub only. Repositories have to be on GitHub. Self-hosted Forgejo, GitLab, and everything else are out.
• One PR per session. Each session can open a single pull request.
• Web access required. Users need Claude Code on the web enabled, or @Claude just answers as a chat assistant.
• Channels only. The integration does not work in DMs, which is consistent with the whole premise but worth knowing.

Those are limits of Anthropic's integration, not of Slack Code as a container. Devin, Copilot, and Vercel each bring their own. Nobody has published a matrix, so plan on discovering yours the interesting way.

Why this is a real shift and not a feature

The last two years of agent tooling optimized for one person going faster. The terminal session, the IDE sidebar, the background job that opens a PR while you sleep. All of it assumes a single human reviewer at the end of a private pipe.

Slack Code assumes the opposite: that the reviewer is a group, that the group includes people who cannot read a diff, and that the artifact worth keeping is the conversation around the change rather than the change itself. Whether that is how your team actually works is an open question. But it is now a default, shipped on the surface where most software teams already argue about what to build, with the four biggest coding agents already plugged in.

Key Takeaways

• Slack Code launched August 20, 2026. Mention a coding agent and it opens a temporary code channel with diffs, live previews, and planning docs, then archives itself as a searchable audit log.
• Four agents live, one pending. Claude, Devin, GitHub Copilot, and Vercel work today; OpenAI's ChatGPT is listed as coming soon.
• The container is free, the agents are not. Launch coverage reports availability on every Slack plan including free, with each agent subscription bought separately. Slack's own pages state no tier.
• Over 70% of code channels close within a day. That is Slack's own pre-launch figure with no sample size, and it tells you the product targets same-day tasks.
• A shared channel is a shared prompt surface. Anthropic's docs warn Claude may follow directions from other messages in the context. The approval gate covers merges, not reads.
• Check which @Claude you are running. Team and Enterprise workspaces are being moved from per-user sessions to Claude Tag's shared org identity; Pro and Max keep the old path.

Sources: Slack (announcement), Slack Code feature page, Anthropic: Claude Code in Slack docs, Anthropic: Claude Tag, SiliconANGLE, The Next Web, Unite.AI