Anthropic Ships 'Reflect With Claude' Beta to Audit AI Habits

Anthropic just added a bathroom scale for AI habits. On July 9 the company rolled out Reflect with Claude in beta, a dashboard that summarizes how Free, Pro and Max users (with memory turned on) actually use the assistant across 1, 3, 6 or 12-month windows, accessible from Settings on the web and desktop app.

The distinctive part isn't the time-series view, it's the scoring rubric. Reflect grades each user against a four-part AI Fluency framework: Delegation ("Setting goals and deciding whether and how to engage with AI"), Description ("Effectively describing goals to prompt useful AI behaviors"), Discernment ("Accurately assessing the usefulness of AI outputs") and Diligence ("Taking responsibility for what we do with AI"). It also fires periodic reflection prompts like "What's one thing you want to keep doing yourself, even if Claude could do it faster?" and offers quiet hours plus break reminders, a wellness UX vocabulary closer to Screen Time than to enterprise analytics.

The carveouts are the tell. Incognito chats, files from connected tools and "any conversation connected to a health integration tool" are excluded, and Anthropic says sensitive conversations only surface "at a high level." That is a lab publicly acknowledging the reputational risk of a company-hosted mirror that inventories what its users have been asking the model, and pre-committing to how much of it stays off the report.

The honest caveat is that a reflection layer sitting on top of the same conversations Anthropic already stores is a UX gesture rather than a governance change; the announcement does not spell out retention, whether Reflect data feeds training, or how the Fluency score is actually computed. Cowork-conversation reflection and time-spent tracking are both flagged as coming "soon," so the beta ships a step shy of the full pitch.

Where this points is habit-shaping as a product surface. If Anthropic can nudge power users toward more deliberate delegation, it lowers the odds that the next "AI made me dumber" essay names Claude, and it hands enterprise buyers something concrete to wave at procurement when they ask about responsible-use tooling.