Claude Tag drops its per-message filter for full context

Anthropic pulled Claude Tag's per-message classifier on August 24, letting its Slack agent read full conversations before responding, per VentureBeat. Anthropic says the change makes the agent about 30% better at judging when not to interject. For anyone running the bot in a busy channel, that's the gap between useful and annoying.

The earlier version scored one message at a time. The new version reads the full thread instead. It picks from four moves: reply inline, open a new thread, route the question to existing work, or stay silent.

Two engineers make the canonical example: one posts a theory, the other posts evidence. Neither message alone would trigger a response; read together, they do.

That expanded context doesn't count against usage or spend limits on any current plan. Anthropic declined to say whether that holds once agents are reading entire channels instead of single messages, leaving pricing for heavier use open.