Slack launched a new product called Slack Code that gives engineering teams a dedicated channel type built around AI-assisted software work. Mentioning a coding agent inside one of these code channels spins up a shared space where the agent, not a single engineer working alone, does the building while the rest of the team watches, comments, and redirects in real time.
That is a different move than what AI Insiders has covered this week. Vercel recently placed its own production agent inside Slack threads, and Anthropic’s Claude tag already lets teams summon an agent into a channel. Both are apps plugging into Slack. Slack Code is Slack building the room itself, the shared surface where those competing agents now have to show up and work.
According to Slack’s own announcement, code channels assemble themselves around whatever the task needs: a code diff, a planning document, or a live HTML preview. When the work wraps, the channel archives automatically but stays searchable, functioning as a standing record of what an agent did and who signed off on it.
Slack says the feature launched with five partners: Anthropic (Claude), Cognition (Devin), GitHub (Copilot), OpenAI (ChatGPT), and Vercel. Of those, Claude, Devin, GitHub, and Vercel integrations are generally available today. OpenAI’s ChatGPT integration is described as coming soon, not yet live. That distinction matters for any team planning to standardize on this workflow before the fifth partner ships.
Slack’s own account of the launch leans heavily on one internal metric: more than 70 percent of code channels open and close, from idea to merged pull request, inside a single day. Slack did not publish a comparison baseline for how long the same work took in a private terminal or browser tab before code channels existed, so the number describes an internal workflow shift rather than a measured productivity gain against a control group.
The pitch targets a specific failure mode. Under the old pattern, one engineer works an agent alone in a separate window, and everyone else on the team learns what got built only once it is finished, too late to flag a wrong assumption. Slack Code instead puts that exchange where colleagues are already watching, so a teammate can interrupt or redirect the agent mid-task. Whatever permission rules and administrative limits a workspace already enforces carry into the channel unchanged, which leaves IT nothing new to configure before the thing is safe to switch on.
The GitHub integration widens who is allowed to kick off the work in the first place. Someone without an engineering background can type a bug description into the channel, an agent picks it up and proposes a fix, and whoever reviews code on that team gets looped in automatically, all without a separate ticketing step. Vercel’s agent posts a live preview link the moment a change ships, so the team can check the actual result before it reaches users rather than trusting a description of it.
The strategic read is simpler than the feature list. By making Slack itself the place where Claude, Devin, Copilot, ChatGPT, and Vercel’s agent all have to operate, Slack is claiming the coordination layer above the agents, not competing to build the best one. That is a platform position: whichever coding agent a team prefers, the channel, the audit trail, and the admin controls stay Slack’s. Teams already running agent-heavy engineering workflows should treat code channels as the default review surface for agent output, and watch whether OpenAI’s pending integration ships with the same live-preview and diff visibility the four launch partners already have.
Slack detailed Slack Code and its code channels in a post on the company’s own blog announcing the launch.
