Slack today unveiled Slack Code, a new feature that embeds AI coding agents like Anthropic’s Claude Code and Cognition’s Devin directly into Slack channels. Teams can now watch, steer, and review AI-driven coding tasks in real time, turning what was once a solo terminal activity into a multiplayer process. Each agent spins up a dedicated channel for a project, logs its work with code diffs and live previews, and archives itself when finished, leaving an auditable trail.
The pitch hinges on shifting bottlenecks in software development. Slack argues that writing code has become cheap while human judgment—ideas, taste, and craft—remains scarce. Executives point to internal data showing a 10x jump in merged pull requests at Cognition after adopting agent workflows, even as headcount grew only 40%. The model extends beyond engineers: sales teams now report bugs in Slack, where agents fix them before they reach backlogs.
Security is central to the design. Agents inherit the user’s permissions and access only what the invoking teammate can see, eliminating new bot-level identities. Execution happens in isolated sandboxes, with optional offline modes to prevent unauthorized actions. Slack frames this as a guard against the industry’s ‘AI slop’ problem, arguing that public visibility and peer review raise quality rather than risk.
The launch arrives as Salesforce doubles down on Slack as its AI front door, following a recent executive reshuffle and the rapid adoption of its new Claude-powered Slackbot.
