Slack turns AI coding into a multiplayer sport

Coding with AI has largely been a solitary, private-tab sport. You fire up a terminal, prompt a model, and paste the results back into your team chat once you think it's done.

Salesforce wants to drag that entire workflow out into the open. Slack has rolled out Slack Code, introducing dedicated "code channels" where human teams and third-party AI agents build software together.

Multiplayer software development

Instead of squeezing multi-step coding sessions into messy comment threads or leaving developers isolated in web tabs, code channels offer a purpose-built workspace for building software.

The feature allows technical and non-technical teammates alike to work side by side with coding agents. A product manager who spots a reported UI bug can tag an agent directly in Slack. The agent creates a dedicated code channel, absorbs the conversation context and uploaded screenshots, and drafts the necessary fix.

From there, anyone in the room can watch the code diffs generate, inspect live previews, steer the agent if it wanders, and give feedback. Deployment to production remains strictly gated behind human sign-off, meaning an engineer still reviews and approves the final merge right in the channel. Once the task is finished, the code channel archives automatically, keeping sidebars clean while preserving the entire build history for global search.

Slack Code is rolling out across all Slack subscription tiers. At launch, teams can plug in agents from Anthropic (Claude), Cognition (Devin), GitHub Copilot, ChatGPT, and Vercel, provided users hold active access to those respective platforms. Slack is also opening up its code channel APIs to third-party developers, while a new Agents tab lets admins monitor running tasks and stop rogue agents mid-execution.

Why it matters

While AI labs and dev tool startups are spending billions to build the ultimate autonomous engineer, Slack is quietly working to own the conference room where those engineers actually work.

We think this positioning is brilliant. For end users, it removes the friction between identifying a problem and shipping a fix. Product managers and designers can prototype and shape code without touching a local terminal, while engineers stay in the loop as gatekeepers rather than blockers.

The bigger test will be code quality at scale. When non-engineers can comfortably prompt agents to alter codebases, engineering teams risk facing a tidal wave of PRs to review. Unless human gatekeeping remains airtight, moving fast in shared chat rooms could quietly balloon technical debt just as quickly as it squashes minor bugs.