Slack Turns Coding Into a Team Sport

Slack Code pulls AI coding agents out of private terminals and into shared channels where your whole team can watch, steer, and approve the work.

Welcome Back,

For most teams, working with an AI coding agent still looks like one developer, one terminal, one private tab. The agent writes, the developer checks the diff, and everyone else finds out what shipped only after it's already live.

Slack wants to change that. On August 20, 2026, Salesforce's Slack unveiled Slack Code, a feature that brings software development out of private tabs and into the open, so teams and their agents can plan, write, and review together, according to Slack's own announcement.

Today we look at what a "code channel" actually is, how the workflow plays out from mention to ship, which AI agents are plugged in at launch, and why Salesforce is betting that building software shouldn't be a solo activity anymore.

💻 What Exactly Is Slack Code

Slack Code centers on a new kind of space called a code channel. Mention a coding agent on a complex project inside any conversation, and one spins up a dedicated channel around that specific task, pulling in anyone on the team, not just the person who kicked it off, according to Slack.

The channel then shapes itself around the work. Depending on the task, the agent builds out a code diff, a planning document, or a live HTML preview, so the format follows what the project actually needs rather than forcing everything into a single thread, per Slack's developer changelog.

Context carries forward automatically. A channel is created from an existing conversation, so agents and collaborators inherit whatever was already discussed instead of starting from a blank slate, as described in the same changelog.

When the task wraps up, the channel archives itself automatically but stays searchable, functioning like an audit log the team can revisit later, according to Slack's product page.

Code channels are scoped to a single task, so the work stays contained instead of sprawling across a team's main channels.

🔍 How the Workflow Plays Out

The trigger is simple: someone has a bug report, a feature request, or a one-off task, like "update the pricing page copy" or "fix the null pointer in the checkout flow," and mentions a supported agent directly in a channel, thread, or DM, according to RejoiceHub.

From there, the agent creates a project-specific channel scoped to that task and pulls in relevant teammates, inheriting context from the triggering conversation so nobody has to re-explain the problem, per the same report.

Inside the session, work is split across dedicated tabs so nothing gets buried: a Conversation tab for the running back-and-forth, a Plan tab with an auto-generated task breakdown, a Code diffs tab showing old lines struck through beside new ones, and a Live preview tab running the actual HTML output, according to Eigent AI.

That structure means the same review engineers already trust, line-by-line diffs, is now open to the whole channel. Teammates can join, review work, add context, and help make decisions, all while collaborating in the same session with the same artifacts, according to Slack.

🛠 The Founding Partner Agents

Slack Code launched connected to a set of founding-partner agents rather than a single in-house model. At launch it works with Anthropic's Claude, Cognition's Devin, GitHub Copilot, and Vercel's agent, with OpenAI's ChatGPT expected to follow soon, according to Slack's announcement.

Access to each partner agent has to be arranged separately. Slack provides the shared, collaborative layer where the work happens, not the underlying coding models themselves, per Dataconomy.

Slack supplies the shared workspace; teams still need their own access to each partner agent to put it to use.

⚖️ Approval Gates and Audit Trails

Giving an AI agent write access to a codebase inside a chat app raises an obvious question: who signs off before anything actually ships? Slack's answer is a human approval step built into the workflow itself, according to Enterprise DNA.

For high-stakes steps like pushing to production, the workflow requires human approval before proceeding, and audit logs keep a record of every action the agent took, per the same report. When a project wraps up, the code channel can be automatically archived while that record stays searchable.

That framing, taken from Salesforce's own announcement, is the real pitch: building software from idea to shipped product no longer has to be an engineering-only activity.

🧠 Why This Is a Bigger Bet Than It Looks

Most AI coding tools today live inside an IDE or a terminal, tools built for one person at a time. Slack Code instead treats the chat app your whole company already lives in as the coding environment, which is a genuinely different starting point.

Reworked summed up the shift plainly: Slack Code lets teams and AI coding agents write, review, and ship software together inside shared channels, without specialized tools or leaving the app, according to Reworked.

The bigger question is whether teams actually want their coding work reviewed by people outside engineering.

🔗 Sources and Further Reading

Thanks for reading AI Spotlight.

Our mission is simple: deliver clear, trustworthy, and actionable AI insights that help professionals stay ahead without the hype.