Slack Code puts AI coding agents into shared team channels

a computer screen with a bunch of code on it

Salesforce just released Slack Code, a new service that drops coding agents directly into Slack channels so everyone involved in a software task, not just the developer, can follow along, provide input, and approve the result.

How It Works

Tag a supported agent from any Slack conversation and the service automatically spins up a project-specific channel, pulls in relevant team members, and gets to work. Supported agents include Anthropic’s Claude, Cognition’s Devin, GitHub Copilot, ChatGPT, and Vercel agents.

Inside the channel, participants get separate views for the agent’s conversation, its plan, and code diffs. They can provide feedback, steer the agent mid-task, or pause and stop it entirely. When the task is done, the agent packages the work for human sign-off and can open a pull request depending on the workflow. The channel then archives automatically but keeps a full searchable history.

The Case For It

The traditional path runs: idea in Slack, ticket in Jira, developer picks it up, works with an agent in an IDE, ships a diff nobody else watched happen. Salesforce is betting that collapsing those handoffs into one shared channel catches wrong assumptions earlier, before they require rework.

Advait Patel, senior site reliability engineer at Broadcom, noted that the transparency has a secondary benefit: watching how a prompt gets constructed and corrected teaches the whole team, not just the developer running the session. That searchable archive also helps when teams need to reconstruct why a code change was made weeks later.

The Honest Trade-offs

Patel was direct about the risks too. Multiple stakeholders piling into the same channel can produce competing instructions and slow the agent down as it adjusts to each one. There is also a real question about whether developers will actually expose messy sessions publicly, or just paste clean results into the channel after the fact, which defeats the transparency argument entirely.

Channel sprawl is another concern. Large enterprises running many parallel tasks could end up managing hundreds of additional Slack channels, with no clear answer yet on how notifications get handled.

Patel’s expectation is a hybrid outcome: shared channels for well-scoped, cross-functional work like copy changes, internal tools, and clearly described bugs, with private developer-agent sessions continuing to handle anything architecturally complex.

Security and Access

Salesforce says agents operate within Slack’s existing permissions, security model, and administrative controls. The service includes APIs for enterprises to customize agent behavior and automate account provisioning and authentication. Analysts flagged that channel membership should not automatically translate into repository access, since that path creates a privilege-escalation risk.

No pricing details were disclosed in the announcement.

Stay on top of AI & Automation with BizStack Newsletter
BizStack  —  Entrepreneur’s Business Stack
Logo