Slack launches Slack Code with 4 coding agents live in team channels
Slack launched Slack Code on August 20, 2026, a set of shared channels where AI coding agents do their work in front of the whole team instead of inside one developer’s terminal. Tag a coding agent in a conversation and it opens a dedicated code channel for that task. Slack’s announcement says the channel fills with whatever the job needs, such as a code diff, a planning doc, or a live HTML preview.
The mechanism matters more than the pitch. Because a code channel is just a channel, anyone in it can follow the agent’s output, pause it, or redirect it before anything ships. When the work lands, Slack says the channel archives itself and stays in Slack as an audit log. The Register reported that each channel maps to one project, which keeps the agent’s running commentary out of the main company chat.
Slack calls this multiplayer AI, and the argument behind it is that solo agent use caps what an organization gets back. Slack Chief Marketing Officer Ryan Gavin put the problem to SiliconANGLE in an interview.
All this individual work is a solo player sport. The knowledge and the speed at which you’re acting as an individual stays with you as an individual, and you’re not getting an exponential return for the organization.
Ryan Gavin, Chief Marketing Officer, Slack, via SiliconANGLE
Slack built the feature for its own engineers first, which is where the launch gets its only hard number. Over 70% of code channels spin up and close within a single day, from idea to merged pull request, the company says. That figure comes from Slack measuring its own internal usage, so it describes a team that already knew the tool rather than a customer base learning it. It’s the claim most worth watching once outside teams start filing their own bug reports into channels.
Five partners built integrations, but they aren’t all arriving at once. Slack’s own post sets out a narrower launch state than the trade coverage did.
| Partner | Agent | Status per Slack’s post |
|---|---|---|
| Anthropic | Claude | Live today |
| Cognition | Devin | Live today |
| GitHub | Copilot | Live today |
| Vercel | Vercel’s agent | Live today |
| OpenAI | ChatGPT | Available soon |
That gap is worth noting because the coverage rounded it off. SiliconANGLE reported the product as available today on any Slack plan, and listed ChatGPT among the agents it opens up to. The Verge also reported availability today on any Slack plan. Slack’s post says ChatGPT is still to come.
The Register flagged the same vagueness, writing that the announcement didn’t spell out which plans have access or which named integrations were live, and that it had asked Slack to clarify.
On governance, Slack’s answer is that there isn’t a new layer to govern. Agents inherit the permissions and admin controls a workspace already has, so IT doesn’t configure anything separate, and higher stakes changes route to a person for approval inside the channel. Whether that satisfies a security team depends on how much they trusted their Slack permissions in the first place. It’s the same question that decides which developer tools earn a place in a stack.
Slack Code enables AI to move beyond individual copilots to embedded, contextual AI orchestration.
Rebecca Wettemann, CEO, Valoir Research, via SiliconANGLE
Slack isn’t stopping at engineering. The company says the same model can stretch into marketing campaigns, legal contract reviews and IT onboarding, and The Register reported that the APIs are due to open to the broader developer community. SiliconANGLE reported that further connectors to third party agents and harnesses can be built through APIs, though you still need access to the agent itself before any of it works.
The thing to watch isn’t adoption, because it lives inside an app these teams already have open. It’s whether that 70% same-day close rate survives contact with teams who didn’t build the product, and whether reviewing a diff in a chat window holds up on work harder than a UI tweak. SiliconANGLE reported that Slack Code won’t replace the hard coding work engineers do in terminals and IDEs. That leaves the open question of how much real engineering fits in the gap, which is a different measurement from what coding agents score on a shared repository.
Get the daily rundown
One email each weekday with the AI news that matters, every claim linked to its primary source.
Free, one email each weekday, unsubscribe in one click. We never sell or share your address.
