Introducing a way to reflect on how you use Claude

How many emails have you let AI draft this week? How many brainstorming sessions did you hand over? When was the last time you tackled a genuinely hard problem entirely on your own? Anthropic’s new reflection feature, launched yesterday, isn’t here to celebrate how many hours you save. It’s quietly asking: are you the pilot, or are you slowly becoming a passenger?

The reflection dashboard lives in Settings, wrapped in a health-app-like interface that shows your Claude usage patterns over the past few months—the topics you gravitate toward, the times of day you lean on Claude most, the types of tasks you keep delegating. A time-spent view is on the way. More intriguingly, it will periodically surface prompts like: “What’s one thing you want to keep doing yourself, even if Claude could do it faster?” And then it invites you to talk through your answer with Claude itself.

Behind the feature sits something larger: the 4D AI Fluency Framework. It’s not about crafting better prompts; it’s about the thinking layer above the prompt. The four dimensions:

• Delegation: knowing when to call in AI and when to stay in the driver’s seat. • Description: translating your goals into descriptions that get genuinely useful AI behavior. • Discernment: accurately gauging the reliability, bias, and real-world fit of AI outputs. • Diligence: taking responsibility for what you create with AI—verifying, editing, standing behind it.

In your report, Claude labels your past interactions along these lines and serves up a summary. It might tell you: “You tend to have Claude draft emails, then rewrite them in your own voice.” Or: “You typically settle the strategy yourself before handing off execution steps.” These are signals of mental sovereignty—evidence that you’re still the one steering.

Most AI products are racing to deliver more speed, more power, higher version numbers. Anthropic is doing the opposite: installing a “slow thinking” buffer inside the AI experience. It’s consistent with the company’s safety identity, but it also points to a larger trend. As AI seeps into every cognitive corner of our lives, the bottleneck isn’t what models can do. It’s whether humans can keep their metacognition alive.

The privacy design matters too. Incognito chats, health-sync integrations, and the underlying files you upload are explicitly excluded. The reflection only summarizes the conversation between you and Claude. That makes the dashboard a kind of psychological fitting room—safe to look in the mirror.

Some will call it paternalistic. I see it differently. The core isn’t restriction; it’s a nudge—like a watch reminding you to stand up. Claude’s quiet hours and break prompts are dismissible, but they’re not invisible. And that light-touch design answers a genuine need among power users: they’re not wary of AI; they’re so in love with it they’ve lost track of their own rhythm.

If you’re a heavy Claude user, try opening the report once a month—not to optimize for speed, but to verify something: that the creativity, judgment, and just the simple satisfaction of finishing something with your own hands are still intact and in your hands.

Maybe Anthropic’s real message to the industry is this: when everyone else is teaching you how to offload more to AI, we need a dashboard that shows us when it’s time to find ourselves again.