Anthropic Rips Out Claude Tag's Message Classifier So the Slack Agent Reads Whole Conversations Before Deciding to Speak

VentureBeat·high signalXBlueskyLinkedInCopy link

Anthropic's August 24 update to Claude Tag removes the lightweight per-message classifier entirely; Claude now reads the full channel conversation and picks among four actions, reply inline, open a thread, route to existing work, or stay silent. Anthropic says the result is roughly 30% better at judging when and when not to interject, with the canonical example being two engineers whose individual messages warrant nothing but whose combined theory-plus-evidence triggers a thread. The expanded channel context does not currently count against usage or spend limits on any plan, and Anthropic declined to commit on future pricing, calling it active co-design with customers.