Salesforce has released Slack Code, a service that embeds AI coding agents like Claude, Devin, GitHub Copilot, ChatGPT, and Vercel agents into dedicated Slack channels where stakeholders can watch an agent work, provide context, and intervene, aiming to reduce handoffs between product managers, developers, and reviewers. Analysts see potential productivity gains from catching mistaken assumptions early and making agent reasoning visible, but also flag risks: Slack's interruption-heavy nature clashing with deep coding work, developers hiding messy prompt iterations before pasting clean results, channel sprawl at scale, and privilege-escalation risks if channel membership grants code repository access. Analysts expect a hybrid model where shared channels handle small, well-scoped cross-functional tasks while private developer-agent sessions remain dominant for complex engineering work like refactors and debugging.
Questions this post answers
What is Slack Code and how does it work with AI coding agents?
Slack Code is a Salesforce service that creates dedicated, project-specific Slack channels when a user tags a supported coding agent such as Claude, Devin, GitHub Copilot, ChatGPT, or a Vercel agent. The channel lets stakeholders view the agent's conversation, plan, and code diffs, provide feedback, and pause or steer the agent, and it can package finished work for sign-off and open a pull request before archiving as a searchable record. See how daily.dev tracks emerging shared-workspace approaches to AI-assisted coding.
What security risks come with putting AI coding agents in shared Slack channels?
Adding someone to a Slack channel could effectively grant them access to modify an underlying code repository if channel membership is conflated with repository permissions, creating a privilege-escalation risk. Analysts caution that enterprises need to separately define what an agent can access within a channel and ensure Slack channel membership does not automatically translate into code repository access. Engineers weighing agent permission models can follow this discussion on daily.dev.
Why might developers resist using shared AI coding channels like Slack Code?
Developers may resist because coding requires deep, uninterrupted work while Slack is described as an interruption machine, and multiple stakeholders intervening midstream can create competing suggestions that slow the agent's progress. There's also concern developers will do messy prompt iterations privately and only paste clean results into the shared channel, undermining the intended transparency. daily.dev helps developers follow debates over how AI coding tools reshape daily workflows.