Slack’s new AI-powered coding channels merge Slack threads with real-time dev tools like Claude and Devin.
When you mention an AI agent in a Slack channel, the system opens a dedicated tab for that task—such as refactor-auth-flow—where the agent runs its own isolated workspace. This tab includes a file explorer, diff view, and embedded live preview of HTML changes before the code even hits production. No more jumping between Cursor, GitHub, or separate IDEs.
The persistent tabs stay linked to the channel, letting teammates revisit changes, leave inline comments, or step in during live coding. Modifications appear as unified diffs, and approvals are handled with a single click—though a separate pull request tab remains optional. For web projects, the iframe preview lets you see CSS and JavaScript updates immediately, while backend changes display test outputs inline.
Teams can switch between agents seamlessly: Claude handles architecture, Devin takes over for coding, and humans jump in for final reviews—all within the same thread. The setup mimics GitHub Copilot’s ambition but leverages Slack’s existing context, from product specs to design docs, since it inherits all conversation history automatically.
Yet adoption faces hurdles. The biggest is authentication: agents need repo access, CI secrets, and staging keys. Slack mitigates this with a new Connected Apps layer, but admins must still configure token scopes per channel. Right now, the preview pane supports only static assets and simple Node or Python servers—Docker Compose and database setups still require manual preview environments.
If your team already uses Claude or Devin, this feature might just add a UI layer. Since pricing isn’t disclosed, assume it’ll be billed per-agent-minute, added to Slack seats. For organizations already paying for the AI APIs, it could feel like a wrapper. Check with your account rep before assuming it fits your workflow.
For teams already synced in Slack and VS Code, this cuts real context-switching costs. The idea isn’t perfect—calling it vibe-coding still feels clunky—but treating code as a first-class Slack feature is the right abstraction. I’ll test it on a small project soon; even the diff review alone might replace half our current pull request comments.
All Replies （4）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Curious about the setup. How are secrets and env vars handled in the new environment? For example, Slack has deployed a feature that feels obvious in retrospect: dedicated code channels where mentioning an AI agent spawns a complete development environment directly within the chat. This eliminates the need to paste snippets across Cursor, GitHub, and separate chat windows. The agent—currently Anthropic's Claude or Cognition's Devin—receives its own tab within the channel, complete with a file tree, diff view, and live HTML preview before any code ships. Daily impact ## How do persistent agent tabs work? - Persistent agent tabs—each development task receives a named tab, such as "refactor-auth-flow," that remains attached to the channel. Teammates can scroll back, review the diff, add comments, or take over mid-session. - Diff-first workflow—modifications appear as unified diffs inline. Approval, tweak requests, or merging occur with a single click. A separate pull request tab remains optional. - Live preview for web work—HTML, CSS, and JavaScript changes render inside an iframe pane. Backend changes display test output. It functions as a lightweight Codespace pinned to the discussion. - Multi-agent handoff—users can start with Claude for architecture, transition to Devin for implementation, and tag a human for review, all within the same thread. The integration resembles what GitHub Copilot Workspace aspired to be but got trapped inside a browser tab. Here, the context—product specifications, design screenshots, prior decisions—already resides
This is huge—especially with how Slack’s persistent agent tabs handle dependencies. Does it actually spin up a full node_modules environment without crashing during initialization, or does it rely on a pre-configured sandbox like the lightweight Codespace integration mentioned in their workflow? The diff-first approach sounds perfect for collaborative debugging, but I’d love to know if it gracefully handles missing or conflicting packages.
Finally! I've spent months fighting with webhooks to get this working. How's the latency? Also, I'm curious if this approach would eliminate the need for my current setup—where I'm constantly pasting snippets across Cursor, GitHub, and separate chat windows. With dedicated code channels, I could just mention an AI agent in the thread and get a full development environment spawned right there, complete with a file tree, diff view, and live preview before anything ships. That would save me hours of context-switching.
Love this, but webhooks are such a headache. Is the reliability actually any better? At least with the new Slack code channels you can keep a persistent agent tab for each task—like “refactor-auth-flow”—so teammates can review diffs inline without juggling multiple windows.