One developer, one terminal, one agent quietly committing code - that has been the default shape of AI-assisted development for the past two years. Instead of one person quietly prompting an agent in a terminal and committing the output, Slack Code puts the entire conversation on the floor for the team to see. That is the problem Salesforce decided to fix on August 20, 2026.
What Slack Code actually does
Slack launched Slack Code on August 20, 2026 - a feature that gives teams dedicated channels for building software alongside AI coding agents. Rather than delegating a task to an assistant in a private chat, a team tags a coding agent such as Anthropic's Claude Code or Cognition's Devin into a conversation, and the agent opens a project-specific code channel where everyone can follow the work, review its output, and sign off before anything ships.
The channel is structured. There is one tab for the conversation, one for the plan, one for the code diffs, and one for a live preview of the output.
When an assignment completes, the channel archives itself automatically, and each channel maintains an audit log for record keeping.
Available across all Slack plans, Slack Code allows cross-functional teams to work directly with leading AI coding agents - including Anthropic's Claude Code, Cognition's Devin, GitHub Copilot, OpenAI's ChatGPT, and Vercel agents - in a shared, central space rather than isolated browser tabs.
That last detail matters more than it sounds. Most teams today manage agent work across a mix of browser tabs, terminal sessions, and DMs. The context lives in one person's local environment. When something breaks, reconstructing what the agent decided to do requires digging through that individual's conversation history - if it was saved at all.
The auditability problem this actually solves
The most underrated part of Slack Code is not the live preview or the diff view. It is the audit trail.
The problem most teams run into is auditability. When one developer owns the entire agent interaction, the rest of the team has no visibility into what got built, why, or what the agent decided to change. If something breaks in production, it is hard to reconstruct what happened.
The coding channel becomes the record of truth. Product managers can see what the agent is building and flag issues before code ships. Security and compliance teams can review the audit log.
That addresses a gap that has been widening quietly. AI coding agents are getting faster and more capable - Cognition reported a significant increase in productivity metrics, with internal merged pull request counts rising tenfold in recent months, despite only a 40% increase in headcount. But speed without visibility is how teams end up with shipped code nobody fully reviewed and incidents nobody can explain.
The plumbing that makes this possible was months in the making. On February 17, 2026, Slack released its Model Context Protocol server and Real-time Search API, giving third-party agents secure, governed access to workspace channels, messages, and files. On June 18, 2026, the company introduced the Slackbot MCP Client, letting developers connect MCP servers directly to Slackbot so it could discover and invoke outside tools in conversation. Slack Code is the user-facing layer on top of that infrastructure, not a standalone addition.
Beagle in action#eng-backend, 2:11pmWhat the team-coding model changes for non-engineers
Future uses could include marketing campaigns, legal contract reviews, and IT onboarding. That is not just product marketing hedging - it describes a structural shift. If the coding channel is transparent by default, a PM, a designer, or a security lead can weigh in on a feature before it merges rather than after a demo.
The before/after contrast is concrete:
The non-obvious consequence: this changes which conversations happen before code ships. Right now, a PM discovers a scope misunderstanding in a PR comment or a demo. With a shared coding channel, the plan tab is visible while the agent is still working. That is an earlier and cheaper correction point.
A teammate like Beagle can also bridge the gap for anyone who does not want to sit in the channel full time - surfacing a one-line status from the agent's latest plan tab into a separate project thread where stakeholders are already watching.
What Slack Code does not cover yet
The four launch agents are strong, but the list is short. At launch, four agents integrate with Slack Code - Claude Code, Devin, GitHub Copilot, and Vercel - though Slack intends to expand the list of options over time. Teams running Cursor, Windsurf, or other coding agents are watching from the outside for now.
The day before the Slack Code launch also brought a related move. On August 19, 2026, Salesforce made the Slackbot MCP Client generally available as part of a broader Headless 360 platform expansion, connecting Slackbot to the Salesforce platform and more than 20 partner applications including Atlassian, Box, Canva, Docusign, Notion, and Zoom. That means the Slack workspace is becoming a coordination layer for agents across tools - not just a place to read about what agents did elsewhere.
The open question is depth. Live HTML previews and code diffs are a good start, but a coding agent doing serious work - provisioning infrastructure, running migrations, calling external APIs - produces decisions and side effects that a diff view does not fully capture. The audit log records what was sent and approved; it does not yet record what the agent considered and rejected. That gap will matter more as agents get more autonomous.
Slack Code AI coding agents: common questions
What is Slack Code?
Slack Code is a feature launched August 20, 2026 that gives teams dedicated channels for building software alongside AI coding agents. A team tags a coding agent into a conversation, and the agent opens a project-specific channel where everyone can follow the work, review output, and sign off before anything ships.
Which AI coding agents work with Slack Code?
At launch, supported agents include Anthropic's Claude Code, Cognition's Devin, GitHub Copilot, OpenAI's ChatGPT, and Vercel agents. Slack has said the list will expand. Teams running other agents - Cursor, Windsurf, and similar tools - are not supported at launch.
Does Slack Code cost extra?
Slack Code is available at no extra cost on all Slack subscriptions. You do need your own subscription or access to whichever coding agent you plan to use inside the channel.
How does a Slack Code coding channel work?
Tag a coding agent from any conversation and it spins up a dedicated code channel. Everyone in that channel follows the work through its own tabs - one for the conversation, one for the plan, one for the code diffs, and one for a live preview of the output. Tasks end, and the channel archives itself.
Why does running AI coding agents in a shared channel matter?
The core problem with agent-in-a-terminal workflows is that the rest of the team has no record of the decisions the agent made - only the output. Slack Code shifts that dynamic. The coding channel becomes the record of truth, available for anyone who needs to understand or audit what was built and why.