Slack Code Explained: How Teams Build Software With AI Agents in Shared Channels

How a code channel works

A typical workflow begins with an idea, bug report, website update, or feature request in Slack:

1. A team member mentions a supported coding agent.
2. The agent creates a dedicated code channel for the task.
3. Participants provide requirements, constraints, and project context.
4. The agent develops a plan and proposes changes.
5. The team reviews the work, including code diffs and available previews.
6. People can redirect the agent or approve the next step.

Slack Code is designed to give the channel more than a conventional chat transcript. Depending on the integration, participants can move between the conversation, the agent’s plan, code diffs, and a live preview of the output. The result is a shared view of both the reasoning around a task and the software produced by it.

Why nontechnical teammates are part of the workflow

Slack Code brings product managers and designers into the building loop without requiring them to work directly inside a developer terminal. They can explain the user problem, add product context, review a visible result, and ask for changes in the shared channel. Developers can then handle the technical assessment and implementation details.

For example, a product manager might report a bug in Slack and describe the expected behavior. An agent can investigate the request and propose a fix, while an engineer reviews the resulting diff and checks whether it is safe for the codebase. The team can then decide whether to proceed with a pull request or another review cycle.

This does not make an AI agent the final decision-maker. It changes where collaboration happens and makes the agent’s work easier for more stakeholders to inspect.

Human approval remains central

Slack Code is positioned as a visibility and collaboration layer, not as permission to deploy unreviewed code automatically. Teams can review proposed changes and require human approval for consequential actions, including changes that could reach production.

That approval model matters because an agent may be able to generate a plausible fix without fully understanding a system’s business requirements, security constraints, or operational risks. A shared channel gives engineers and other responsible team members a place to question the approach, request revisions, and record the decision before the work advances.

Searchable history, archiving, and Slack controls

Code channels are intended to preserve the context surrounding an agent’s work. When a task is complete, the channel can be archived while its conversation and work history remain searchable, creating a record of what was requested, what the agent produced, and how people reviewed it.

Because the workflow runs inside Slack, organizations can use their existing Slack identities, permissions, governance, security settings, and administrative controls instead of introducing a separate collaboration system for every coding-agent task.

The practical benefit is continuity: project requirements, decisions, reviews, and agent activity can remain connected to the conversations where the work was initiated.

Availability and launch integrations

Salesforce said Slack Code was available across Slack plans at launch. The founding integrations announced for the feature came from Anthropic, GitHub, Cognition, and Vercel, while Salesforce and Slack also presented ChatGPT as one of the agents that could participate.

The exact experience can vary by agent integration. Features such as plans, diffs, previews, and approval actions depend on what the supported agent makes available inside Slack, so “supported” does not necessarily mean that every agent exposes identical controls.

Salesforce’s broader strategy for Slack Code

At Dreamforce, Salesforce framed Slack Code as a way to make software development a team sport while positioning Slack as the coordination layer for multiple AI-agent providers. Rather than requiring teams to adopt one Salesforce coding model, the approach brings agents from competing providers into a shared workplace interface.

Salesforce has also described plans to open the underlying APIs more broadly. The longer-term vision is for organizations to create custom agents and shared channels for work beyond software development, such as coordinating marketing campaigns or reviewing legal documents. Those examples describe planned expansion, not proof that every noncoding workflow is generally available at launch.

The bottom line

Slack Code’s core idea is straightforward: mention a coding agent, give it a dedicated project channel, and let the broader team watch, guide, review, and approve the work. Its product distinction is the shared context around the code—not merely the ability to generate code.

For teams already working in Slack, that could make AI-assisted development easier for product and design stakeholders to follow while preserving engineering review. The feature still depends on the quality of each agent integration and on teams maintaining clear approval standards, but it offers a more visible alternative to isolated AI coding sessions.