Claude Tag now reads the entire Slack conversation and decides on his own when to intervene

Anthropic replaces the message-by-message classification with a reading of the entire thread: the agent chooses between replying, opening a work thread, linking the subject to an existing site or remaining silent.

Claude Tag's August 2026 update changes the way Anthropic's agent fits into Slack. Where a light classifier evaluated each message in isolation, Claude now accesses the entire context of the channel, its memory and the standing instructions given to it, allowing it to identify recurring reasons in an exchange rather than react to a sentence.

Four behaviours are available to him: to respond directly in the canal, to open a thread to engage in more in-depth work, to link the discussion to a project already followed, or to remain silent. Anthropic claims a gain of about 30% on the quality of this decision, stressing that progress is as much on cases where the agent must abstain as on those where he must intervene.

The restraint was deliberately included in the system, the company considering that an annoying agent is worse than an unnecessary agent: Claude puts himself on standby in the channels where his contributions prove to be repetitively interestless. For the time being, the wider context is not counted from the user quotas, without the publisher's commitment to the sustainability of this free of charge.