claude reflect, explained: how to read your ai usage dashboard.
anthropic just shipped a dashboard that shows how you actually use claude. most people will glance at it once. read the right way, it tells you whether you are handing ai the right work.
claude reflect is a usage dashboard anthropic launched on july 9, 2026. it shows how you have been working with claude: your topics, your active days, and the kinds of tasks you hand it. it lives in settings on web and desktop, looks back across 1, 3, 6, or 12 months, and ties your usage to anthropic's four skills of ai fluency.
i pay attention to this one for a specific reason. the thing that decides whether ai sticks is not how often you open it. it is whether the work you give it is the right work. reflect is the first anthropic feature that puts that question in front of you with your own data behind it.
what claude reflect actually shows you
reflect opens with a summary of how you have been using claude. per anthropic, that summary covers three things: your key topics, your usage patterns, and the kinds of tasks you tend to work through. you can move the window between the last 1, 3, 6, or 12 months, and time spent inside claude is listed as coming soon.
| what reflect shows | what it is |
|---|---|
| your topics | the subjects you bring to claude most often, pulled from your history. |
| usage patterns | when you work with it: your active days and the stretches you lean on it hardest. |
| task types | the kinds of jobs you hand it, grouped so you can see the shape of your delegation. |
| lookback window | the same view across the past 1, 3, 6, or 12 months, so a busy week does not read as a habit. |
| quiet hours + break nudge | reminders you set for yourself. both are your own preferences and can be dismissed. |
who can see it, and where it lives
reflect is in beta for free, pro, and max users who have claude's memory turned on. you find it in the settings section of claude for web or the desktop app. anthropic says cowork conversations are coming next.
on privacy: reflect leaves out incognito chats and files pulled in from connected tools, and it omits health-integration conversations entirely. sensitive conversations can still show up, but only at a high level. anthropic states the insights stay in the dashboard and are not used for any other purpose. that matters, because a usage log is only useful if you are honest in front of it.
how to read it through the 4Ds
here is where the dashboard earns its place. reflect connects to anthropic's 4D ai fluency framework: delegation, description, discernment, and diligence. i wrote the plain-language version of all four in the 4 skills anthropic named. reflect is the first time those four skills point back at your own behavior. here is what each one asks you to look for.
| the skill (the 4Ds) | what reflect surfaces | the question to ask |
|---|---|---|
| delegation | the task types you hand claude | are these the repeat, low-risk jobs, or the ones you should still be doing yourself? |
| description | the topics you return to again and again | are you re-explaining the same context every time? that is a project, not a fresh prompt. |
| discernment | the volume of work you route through it | how much of that went out without a second read from you? |
| diligence | your patterns across months | is your ai use deliberate, or has it drifted into a place you did not choose? |
the delegation row is the one to sit with. most people who quit ai did not pick a bad model. they picked the wrong task: something high-stakes or one-off, got a confident wrong answer, and backed away. the fix is upstream, in what you choose to delegate. the delegation test is in what tasks to give ai, and catching the misses is in how to tell when ai is confidently wrong.
the question the dashboard is really asking
reflect will periodically put a prompt in front of you like: what is one thing you want to keep doing yourself, even if claude could do it faster? that is not a wellness gimmick. it is the delegation skill in one sentence. delegation is not "give ai everything." it is deciding, on purpose, what stays yours.
this is the part worth being careful about. anthropic built reflect partly to help you use claude well, and partly to help you use it more. those are not always the same thing. the dashboard is genuinely useful, and it also nudges you toward deeper integration, like moving repeated context into a project. read it as a mirror, not a scoreboard. the goal is not more claude. the goal is the right work going to claude and the rest staying with you.
how to read your reflect dashboard in five steps
- open it. settings on claude web or desktop, then reflect. memory has to be on for it to appear.
- pick a window. start at 3 months, so real patterns show instead of one loud week.
- read task types first. that list is your delegation map: the work you already trust it with.
- turn your top topic into a project. if you keep re-explaining the same thing, that is a project waiting to happen.
- set one nudge you will keep. quiet hours or a break reminder. one you will respect beats three you ignore.
the takeaway
most usage dashboards count activity and call it insight. reflect is more useful than that only if you read it against the right question. not "how much do i use claude," but "am i handing it the right work, and keeping the right work for myself." that is the whole of delegation, and delegation is the skill that decides whether ai stays a toy or becomes part of how you operate. open the dashboard, read the task types, and change one thing.
want the systems, not just the dashboard?
reflect tells you what you are doing with claude. the ai builder toolkit is the set of ready-to-use skills that make claude do the repeat work reliably, so the task types you delegate actually hold. built for solo operators and small teams.
see the ai builder toolkitor just follow along. new field notes most weeks on x, instagram, and tiktok.