Slack released Slack Code on August 20, giving teams and AI coding agents a single place to work on software together. A dedicated code channel holds the conversation, the code diffs, and a live preview side by side, so people who are not engineers can follow along and weigh in. It is available on every Slack plan.
Pulling agent work out of private tabs
Development with a coding agent has mostly been a solo activity. One person works with the agent in a terminal or a browser tab, and the team sees only the result. That leaves the reasoning invisible: nobody else knows what was considered, or why the implementation landed where it did. Reviews get pushed to the end, and every handoff means rebuilding context from scratch.
Slack Code moves the workspace itself onto the team's side. When development happens inside a channel, the exchange is visible by default, and context carries forward from person to agent and from person to person. Rob Seaman, EVP and GM of Slack, has framed the idea simply: AI creates value when it is part of how a team actually works, not when someone uses it alone in another tab.
One channel per project, archived automatically when the work ends
A code channel maps to a single project. Tag a coding agent from any conversation and it spins up a code channel for that task. Once the work is finished, the channel archives itself, while the record stays behind as an audit log. The sidebar does not fill up with disposable channels.
Inside the channel, the work is split into tabs: the conversation, the plan, the code diffs, and a live preview of the running HTML. The diff view strikes through the old line and places the new one beside it, the same review format engineers already use. What used to sit inside a separate tool is now in front of everyone in the channel. Agent behavior can be shaped through APIs, so account provisioning, user authentication, and parts of the workflow can be automated to match how an organization runs.
Building stops being an engineers-only job
Slack's own example makes the shift clear. A product manager spots a bug report in a channel and, instead of waiting for an engineer to free up, asks a coding agent to help work out a fix. The agent opens a code channel, reads the conversation, screenshots, and documents already shared there, and proposes code. The PM then brings in an engineer to check the diff, run the preview, and give the agent the go-ahead to open a pull request and merge.
There is no ticket and no meeting, but there are brakes. Anyone in the channel can pause, redirect, or stop the agent the moment it drifts. For high-stakes moves such as pushing to production, the agent packages its work and asks an expert to sign off, so nothing ships without a human decision. Permissions and security come from Slack's existing admin controls, which means IT does not have to stand up new infrastructure.
The major coding agents are already on board
At launch, Slack Code works with Anthropic's Claude, Cognition's Devin, GitHub's Copilot, ChatGPT, and Vercel's agents. All of them plug in through the code channel APIs, and access to each partner agent is contracted separately.
Slack also plans to open those APIs to the wider developer community. Once that happens, custom agents could join code channels for work well beyond software, from running a marketing campaign to reviewing legal documents. The surrounding agent experience changed too. Agent DMs now behave like any other conversation, with real thread titles that make it easy to pick things back up. A new Agents tab collects every agent and session in one place, showing status and allowing a run to be stopped mid-task. An Add to Slack flow lets agents built on Lovable, n8n, LangChain, and similar tools be dropped into a workspace in a few clicks.
Summary
Slack Code answers a practical question about AI-written code: where does the team actually review it. Rather than competing on agent capability, Slack is relocating the output to a place the whole team already watches, which is a grounded place to start. How far it spreads beyond engineering will depend on what happens after the APIs open up.
