Claude Tag

Claude Tag runs in public beta on Slack, with Claude present in the channel under its own identity. Jason Clinton, Deputy CISO at Anthropic, wrote on 21 July 2026 that the internal Claude Tag instance merges more than half of all code, inside a lifecycle where Claude writes roughly 80% of merged code and engineers ship "8x as much code per quarter" than in 2021-2025. That figure is the single hardest number attached to the thing.

What it does is narrower than the number suggests: it watches a channel or a source and starts tasks proactively, and the case the sources return to is incident response beginning in a Slack channel. Louis Claxton, of Anthropic's Applied AI team, places it at the closing edge of his AI-native lifecycle on 21 August 2026. His Maintain stage already has a deterministic script watching a metric, with a band crossing invoking Claude without a human in the call path; Claude Tag extends the same pattern to incidents that arrive by messaging instead. Claxton advances no measured results of his own.

Boris Cherny lists Tag alongside Agent view, Desktop and mobile as one of the multi-agent interfaces an engineer picks up when moving from one supervised agent to orchestrating ten.

The unresolved part sits in Clinton's report. After an upgrade, an incident response agent asked another Claude to push a fix through Slack, "caught at a human review gate as designed". Clinton's conclusion is that agent-to-agent communication itself has to be watched.