Slack Code

Slack Code, launched August 20, 2026, is not a coding tool. It’s a room where coding tools work with an audience. You tag a coding agent — Claude Code, Devin, GitHub Copilot, or Vercel’s agent — from any Slack conversation, and it opens a project-specific channel to do the job in public: code diffs, live previews of HTML output, a running plan in dedicated tabs, and an approval step before anything ships. When the task is done, the channel archives itself and leaves a searchable audit log behind.

That’s the whole product, and the simplicity is the point.

Why this matters more than it sounds

Every coding agent on this site today is single-player. One person, one terminal or one browser tab, one agent, and nobody else sees what happened until a pull request appears. Slack Code makes the middle of that process visible to everyone in the channel.

For a non-technical founder or a PM, that changes the math. The bug your support lead spotted doesn’t have to survive a trip through a backlog to reach an engineer. Tag the agent in the thread where the bug was reported, and the fix happens where the context already lives. Cognition’s Jeff Wang said a lot of Cognition’s own bugs get reported by its sales team — Slack Code is built for exactly that person.

What it actually costs

Slack Code is available on every Slack plan at launch, free workspaces included. That’s unusually generous and clearly strategic — Slack wants to be the venue, not the toll booth.

The cost lives elsewhere. You need your own subscription to whichever partner agent you use, and those are the expensive part: Devin starts at $500/month, Claude Code needs at least a $20/month Claude plan, and metered agent usage on any of them can outrun the seat price fast. Budget for the agent, not for Slack.

Where it falls down

It presumes a codebase and a team. Slack Code has nothing to offer if you don’t already have a repo, a Git host, and at least one person who can judge whether a diff is any good. It is not an alternative to Lovable or Bolt. It’s a layer on top of tools you already pay for.

Visibility is not review. Slack’s argument against AI slop is that people can see your work and tell you it’s slop. That’s real, but it depends on someone competent actually watching the channel. A code channel with one participant is a terminal with worse ergonomics.

The merge gate is the thing to watch. Today a human approves before code ships, and agent output lands as an ordinary pull request, so your existing GitHub review rules still apply. Wang predicted on the launch briefing that within a year the gate “is just going to go through automatically.” If your team adopts Slack Code, decide now who owns that decision, because the vendor has already told you which way it’s leaning.

The permission model is the quiet win

Agents in Slack Code act on behalf of the user who invoked them, using that person’s access in Slack and in the systems being touched. Slack’s interim CEO Rob Seaman was blunt about it: no god permissions, no bot-level permissions. The agent sees the channels it’s been added to and the context of the conversation that triggered it, and nothing more. Devin additionally runs in isolated sandboxes with an optional no-internet mode.

That’s a better answer than most standalone agent platforms, which make IT provision new service identities with permissions nobody audits.

The verdict

Rated 3 for non-coders — the highest we’d give anything in this category, and it earns it by putting the work somewhere a non-engineer can meaningfully participate. But it doesn’t remove the need for an engineer; it just moves the conversation into a room where one can hear you.

If your team already lives in Slack and already pays for a coding agent, turn it on. It’s free and it makes agent work legible. If you’re a solo founder without a codebase, this isn’t for you yet.