Slack Code turns Claude, Devin and Copilot into full channel participants

Coding agents join the conversation

According to a post on dev.to, Slack now lets coding agents operate as full channel participants rather than tools invoked from a terminal. A capability the post calls Slack Code supports agents including Claude, Devin, Copilot, ChatGPT and Vercel, which can be tagged into ordinary conversations and take on tasks end to end — from a request to fix a bug through to a merged pull request.

How code channels work

Tagging an agent creates what the post describes as a dedicated code channel: a temporary workspace with diffs, live previews and a running plan shown in tabs. Teammates can watch progress as it happens and steer the agent while it works. When the task is finished, the channel archives rather than disappearing, so the history of what the agent did remains searchable — in effect an audit trail for every change.

Reported speed and internal results

The post offers two attention-grabbing figures. It claims 70% of these code channels close within a single day, meaning a request raised in chat can go from question to merged pull request in under 24 hours, with no ticket, no sprint planning and no waiting in a backlog. It also relays numbers it attributes to Cognition, the company behind Devin: internal merged pull requests rose roughly tenfold over recent months while headcount grew about 40%. The author reads that combination as agents multiplying what each developer can deliver rather than replacing developers outright. These are secondhand figures from a single blog post and have not been independently verified.

Non-engineers enter the loop

The change the post frames as most consequential concerns who can start work. A product manager or other non-technical colleague can describe a problem in plain language, have an agent draft a fix, and pull in an engineer only to review it — all without leaving Slack. Review, usually a separate and often delayed stage, becomes something that happens in real time alongside the conversation that prompted the change.

Why it matters

The notable shift is not that agents can write code, which is now routine, but where they do it. Running agents inside the chat application a team already uses for coordination removes constant switching between terminal, editor and messages, and makes the agent's work visible to everyone at once instead of confined to one person's session.

That visibility cuts both ways. It invites faster feedback and broader participation, but it also raises questions the post leaves unanswered: how human review capacity keeps up when merge volume multiplies, and how quality is governed when people outside the engineering team can trigger code changes directly. If the Cognition figures are representative, the practical effect is a large increase in shipped work with only moderate hiring growth — and the bottleneck moves from writing code to supervising it.