Slack Code: AI Coding Agents Go Multiplayer

Slack Code pulls AI coding agents out of private terminals and into shared team channels. Announced on August 20, 2026, it lets you tag an agent — Claude, Devin, GitHub Copilot, or ChatGPT — from any conversation, spin up a dedicated code channel, and have your whole team watch the work, review the diffs, and approve the ship without leaving Slack. This guide covers how code channels work, which agents are supported, and the honest trade-offs.

What is Slack Code?

Slack Code is a coding environment built into Slack where teams and AI agents write, review, and ship software together in the open. Instead of a developer working alone with an agent in a browser tab or terminal, the work happens in a shared channel everyone can see.

The pitch is a shift in how AI coding happens. So far, most agent work has been single-player: one person, one agent, invisible to the rest of the team. Slack Code makes that work "multiplayer" — the plan, the code, and the review all live in a channel where teammates can steer and sign off.

It's available on any Slack plan, but you bring your own access to the partner agents you want to use.

How Slack Code channels work

The mechanic is deliberately simple. When someone spots a bug, has an idea, or needs a change, they tag a coding agent from any conversation. The agent creates a project-specific code channel to handle that task.

Inside the channel, the work is split across dedicated tabs so nothing gets buried in a thread:

• Conversation — the running back-and-forth between people and the agent.
• Plan — an auto-generated task breakdown of the steps the agent intends to take.
• Code diffs — a side-by-side view: old line struck through, new line beside it. The same review engineers already trust, now open to the whole channel.
• Live preview — the real HTML output running live, so the team can catch anything broken before it ships.

When the task is done, the channel archives itself. The record stays behind as a searchable audit log, so the sidebar stays tidy while the history remains as institutional context.

Crucially, a human stays in the loop for high-stakes moves. The agent can't push to production on its own — for actions like a production deploy, it packages its work for a designated reviewer to approve inside the channel.

Which AI agents does Slack Code support?

At launch, Slack Code connects to a set of founding-partner agents:

• Anthropic's Claude (including Claude Tag)
• Cognition's Devin
• GitHub Copilot
• ChatGPT
• Vercel's agent

Access to each partner agent is required separately — Slack provides the collaborative layer, not the models. Slack also says the code channel APIs will open to the broader developer community, so custom agents could eventually join channels for non-engineering work like marketing campaigns or legal document review.

There's also a lighter on-ramp: an Add to Slack flow that lets teams drop in agents built on tools like Lovable, n8n, LangChain, and Vercel in a few clicks, with the OAuth and manifest setup automated.

Why Slack built this

Slack Code didn't appear from nowhere. It's the full-featured evolution of earlier experiments — Anthropic's Claude integration and then Claude Tag, a persistent shared agent that lived in channels as a continuous collaborator rather than a one-off. Slack Code takes that idea and gives agent work its own purpose-built space.

The strategic bet is bigger than a coding tool. Salesforce is positioning Slack as the operating layer for AI agents in the enterprise — the single interface where diverse third-party agents show up as teammates, follow the same channel norms, and answer to the admin controls IT already trusts. Software engineering is described as the starting point, not the limit.

If your team already lives in Slack and a lot of your work starts as a "what do you think about…" that turns into code, Slack Code removes the tool-switching and the lost context between chat, IDE, and PR review.

The trade-offs to weigh

The demos look clean, but a few honest caveats are worth naming before you roll this out.

It won't replace real engineering. Slack Code is best understood as the scaffolding and review layer — the place where an idea becomes a reviewed diff. The nuts-and-bolts work in terminals and IDEs doesn't disappear. Expect it to shine for bug fixes, small features, and web changes more than deep systems work.

Everything runs in the cloud, inside Salesforce's platform. Multiplayer visibility is the point, but "work in the open by default" means your prompts, diffs, and context live in Slack's audit log. Teams with strict data-residency or sovereignty requirements should think through governance before wiring agents to production.

You still pay for the agents. The channel is free on any plan; Claude, Devin, Copilot, and the rest are not. Budget for the underlying agent access separately.

Governance is a real question. Letting non-engineers summon agents that touch code is powerful and risky. The human-approval gate for production helps, but who gets to tag agents, and in which channels, becomes a policy decision.

Do this with your own AI workforce — locally

Slack Code proves the direction of travel: coding is going from single-player to a team of humans and agents working together. But if you want that multi-agent workflow without sending every diff and prompt to a cloud audit log, you can run it on your own machine.

Eigent is an open-source "Cowork" desktop app — a multi-agent AI workforce that plans, writes, and reviews work locally, so sensitive context never leaves your environment. If your interest is agents that ship code, see our roundup of the best open source AI coding agents and the self-hosted AI coding agent guide. When you're ready, download Eigent and build your own multiplayer workflow — on your terms.