On August 20, 2026, Salesforce and Slack released Slack Code, a way for coding agents and teams to work in the same place1. Tagging a coding agent inside a conversation spins up a code channel dedicated to that task1.
Slack describes this as AI coding becoming a team sport, and positions it as moving software development out of individual browser tabs1. The Verge reported the same day that it gives teams a dedicated, open space to collaborate with AI agents2.
A channel that starts from a conversation
According to Slack, when you have an idea, need to update a web page, or want to fix a bug, you tag in a coding agent and it spins up a code channel to tackle the task1. Each code channel maps to one project, stays live while you need it, and is archived when you don’t1.
Inside the channel, the conversation, the plan, the code diffs, and live previews each get their own tab, so everyone in the channel can follow the work as it happens1. Diffs show the old line struck through with the new line beside it — Slack describes this as the same review engineers have always relied on, now open to the whole channel instead of buried in a separate tool1. Previews run the real HTML output live so the team can catch anything broken before it ships1.
When the task is done the channel archives itself, but the record remains as an audit log1. Slack says the history stays fully searchable1.
Slack offers this scenario: a product manager spots a bug report in a channel and, instead of waiting on an engineer, asks a coding agent to help architect a fix. The agent spins up a code channel, reviews the conversational context, screenshots, and docs already shared, and proposes code. The PM brings in an engineer to check the diff, run the preview, and give the agent the go-ahead to open a PR and merge to production1.
Who can stop it, and where approval lives
Two questions decide whether handing work to an agent is workable in practice: can you stop it mid-flight, and where does a human sign off.
Slack says anyone in the channel can pause, redirect, or stop the agent if it drifts off track1. That permission is not reserved for engineers, which is a notable part of the design.
On permissions, agents in code channels inherit Slack’s built-in security model, permissions, and admin controls from day one1. The company points to there being no new infrastructure or identities to manage1. For high-stakes moves such as pushing code to production, the agent packages its work for an expert to sign off on, right in the channel1.
Updates that let agents pick up work on their own have been arriving through the same week. On August 19, Cursor added Subscriptions to its cloud agents, letting them subscribe to PRs and Slack threads and wake themselves up. If that release is about how an agent starts moving, Slack Code is about where people watch that work and where they stop it.
Five vendors’ agents in one channel
This release connects the agents of what Salesforce calls its founding partners: Anthropic’s Claude, Cognition’s Devin, GitHub’s Copilot, ChatGPT, and Vercel’s agents1. Slack Code is available on any Slack plan, but access to each partner agent is required separately1. The company also notes that pricing and packaging are subject to change and that availability may vary by region1.
The code channel APIs are currently used by partners, and Slack says they will open to the broader developer community “soon”1. Once they do, agents doing work other than software development — a marketing team building a campaign, legal running a document review — will be able to join a code channel, according to the company1. No timing was given.
The partner comments each reflect their own vantage point. Cat Wu, Head of Claude Product at Anthropic, notes that much engineering work already starts as a conversation in Slack — a bug report, a piece of feedback1. Mario Rodriguez, Chief Product Officer at GitHub, framed Slack as part of a broader GitHub promise in which humans set direction and agents close the loop1. Malte Ubl, Chief Technology Officer at Vercel, said a thread works for a quick back-and-forth, but work with an agent has become valuable and complex enough to deserve its own space1.
Salesforce also states that no other platform offers the same combination of an open ecosystem and a multiplayer coding environment for humans and agents working together at once1. That is the company’s own assessment, not the result of an independent comparison.
Agents get a home base across Slack
Beyond code channels, Slack is changing how agents appear throughout the product1.
Agent DMs work like any conversation with a teammate: agents are clearly labeled, and threads get real titles so a conversation can be picked up where it left off1. The new Agents tab gives every agent and agent session a home base — browsing agents, naming conversations, checking statuses, flagging a blocked thread, and stopping an agent mid-task all live there1.
Add to Slack lets anyone ship an agent built on NanoClaw, Lovable, Hyperagent, Superhuman, n8n, Vercel, ChatGPT, LangChain, Runlayer, or Skydive in a few clicks, with OAuth, manifest setup, and environment config automated1. Slack says agents show up with a clear identity, follow the same channel norms as any teammate, and answer to the same admin controls IT already trusts, whether in a code channel, a DM, or a project channel1.
Making chat the workspace for agents is something several companies have tried this year. In July, Jack Dorsey’s Block released Buzz, a team chat platform built around human and AI agent teams. Slack Code brings that idea from the direction of an existing large workspace.
Checking more output
As agents write more code, review and record-keeping have to keep pace. GitHub made a Trends tab for tracking organization-wide code quality generally available on August 19, a tool for seeing how open findings rise and fall across an organization.
Slack Code operates one level down, at the individual task. Who was watching the work, who could stop it, and who approved it all remain in the channel’s record1. Rob Seaman, EVP and GM of Slack at Salesforce, said AI only creates value when it is part of how a team actually works, rather than something people do alone in another tab1.
For anyone setting internal rules on agent use, the open questions are narrow: where to draw the line on operations that require approval, how to reconcile the audit log of archived channels with existing records management, and who pays for the partner agents — availability spans every plan, but access to each agent is billed separately1.
Sources
- Introducing Slack Code: Agentic Coding for Teams - Salesforce announcement (August 20, 2026)
- Slack is launching collaborative vibe-coding channels - The Verge (August 20, 2026)