Summary
Anthropic’s official Claude channel releases a first-person showcase of Claude Tag, its new proactive AI agent designed to live inside team communication channels. Unlike conventional AI tools that respond only when explicitly invoked, Claude Tag monitors channels autonomously, initiates work on its own schedule, files pull requests, runs data analyses, and follows up over periods of days, weeks, or months — all without requiring users to re-prompt. Anthropic employees describe it as the third major leap in two years, progressing from line-by-line typeahead completion, through one person managing many concurrent Claude sessions, to Claude itself driving work across entire teams.
Three technical capabilities enable this shift. First, long-horizon autonomy: Anthropic’s latest METR evaluations show their models can now sustain autonomous work for 16 continuous hours, with the upper bound still unmeasured. Second, persistent memory: Claude Tag retains per-channel instructions from all users indefinitely, adjusting scope or behavior when any team member updates its mandate. Third, multiplayer collaboration: the entire team can observe and guide the same Claude Tag session in real time, allowing expert usage patterns to spread organically rather than through top-down training.
Internal use cases highlighted include a data Q&A channel where Claude Tag answers questions unprompted, a Claude Code feedback channel where it automatically opens fix PRs, and long-running scientific experiments where it checks results daily and pushes patches when bugs arise. The product positions Claude as a persistent team member rather than a tool users remember to open.
📺 Source: Claude · Published July 02, 2026
🏷️ Format: Keynote Launch
