Bottom line: Claude Tag now evaluates proactive Slack responses based on the entire channel context instead of individual messages, which Anthropic says improves decision accuracy by around 30 percent and is available at no additional cost or impact on limits.
Anthropic has overhauled Claude Tag, Claude’s Slack integration: instead of evaluating individual messages in isolation, Claude now takes into account the entire channel history, its memory, and any stored default instructions in order to intervene proactively in a meaningful way. According to Anthropic, this improves the accuracy of deciding whether and when Claude should speak up by around 30 percent.
Claude Tag adds Claude as a participant in a Slack channel. Previously, a lightweight classifier decided for each message individually whether Claude should respond proactively – without regard to the surrounding conversation. Anthropic cites the example of two engineers tracking the same bug from different angles: one person provides a theory, the other supplies the matching evidence, but neither message contains an explicit request to Claude. Viewed individually, the old classifier correctly recognizes that no response is needed – but in the overall context, an obvious work task would have been apparent.
With the update, the isolated classifier is eliminated. Claude now chooses between four response patterns: giving a short, verifiable answer directly in the channel; starting its own thread for topics requiring more effort; assigning a message to an already ongoing workstream; or staying silent when no response is appropriate. In the example above, Claude opens a thread without an @-mention, links the theory and evidence from the two engineers, and includes both people. Follow-up updates on an ongoing investigation are assigned to the matching workstream, even if two threads later turn out to concern the same bug.
To decide when Claude should stay silent, Anthropic uses a rubric-based evaluation system per channel that takes into account, among other things, the potential usefulness of a contribution, the confidence of the answer, and whether another person would be better suited to respond. Claude also autonomously reduces its attention to channels where it repeatedly concludes it has nothing to contribute – an @-mention then immediately reactivates it. Users can additionally control this behavior via free-text instructions, such as “Only respond here when tagged” or “Feel free to chime in on anything related to the deploy pipeline.” The “Respond automatically” option also allows proactive responses to be fully disabled per channel, so that Claude only reacts to @-mentions.
Anthropic states that the expanded context processing is provided at no additional cost: the larger context volume that Claude Tag now retains does not count against usage or spending limits on any plan. In addition, the initial response is said to occur faster – according to Anthropic, Claude acknowledges a request within seconds, rather than starting invisibly in the background first.
Source: claude.com · Published August 12, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.