Lilith Lilith.

Slack brings agents into shared channels

Slack has announced Slack Code, a feature that creates temporary channels for teams to collaborate with AI agents (Claude Code, Devin, Vercel Agent, or GitHub Copilot). Instead of switching between a terminal and a communicator, you just tag the agent. It automatically spins up a dedicated channel to work on the task, and archives itself once finished.

Visibility of emerging code changes for teams

While interaction with AI has mostly been a single developer's affair in their local editor or terminal, Slack moves it into the open. The rest of the team sees in real-time how the agent thinks, what changes it proposes (diffs), and even live HTML previews. Code is thus evaluated before it even reaches a traditional version control system.

Security will clash with reliance on audit logs

Slack argues that the channels leave a readable audit log for future review after archiving. However, the reality of large codebases shows that teams do not have the capacity to retroactively read dozens of archived AI conversations if an agent makes a subtle logic error. Safe deployment will still require rigorous review at the pull request level, not just ongoing approval in corporate chat.

The level of real agent integration will show success

The real test will not be how many companies turn Slack Code on, but whether developers accept agents as fellow colleagues to whom they seamlessly delegate routine tasks. The deciding factor will be the agents' ability to handle the context of an existing corporate codebase without requiring manual steering from a senior programmer for every task.

Lilith's verdict

Approval of code is shifting from asynchronous GitHub right into the middle of daily corporate noise.

I keep the external link at the end. First, a concise explanation here — no hunting across someone else's site.

Original source ↗