Claude will now show you what you use AI for, and the report doubles as a free systems audit

Reflect is live in Settings for every Claude plan with memory on. Ten minutes in it finds the task you keep re-explaining, and that task is your next standing system.

TL;DR: Anthropic shipped Reflect on July 9, a beta dashboard inside Claude’s Settings that shows what you’ve used AI for over the last 1 to 12 months: top topics, peak hours, task categories, and recommendations built from your own patterns. It’s free on every plan, it takes ten minutes to read, and the most useful thing in it is the task you keep re-explaining in fresh chats, because that task is a system waiting to be built.

This newsletter got researched and drafted this morning by a scheduled Claude task, the same setup I covered in the Work From Anywhere issue, so I have a running record of where my own AI usage concentrates without needing a dashboard to tell me. Most people don’t have that record, and until last Thursday there was no easy way to get one. Anthropic shipped it: Reflect, a beta dashboard that reads your chat history and hands you the usage report.

By the end of this you’ll know what Reflect shows, the one pattern worth hunting for in your own report, and how to promote that pattern from a chat you keep re-starting into a system that runs without you re-explaining it.

The pattern has a name worth keeping: The Context Rebuild Loop. It’s the cycle where you open a fresh chat, re-explain your business, your format, your preferences, get the output, and throw the setup away, then pay that same setup cost again three days later on the same task. Anthropic’s own product team built detection for it, Reflect watches for people who, in Engadget’s words, “frequently re-establish the same or similar context” and nudges them toward Projects. The loop is invisible while you’re in it, which is what makes a mirror useful.

• What shipped: a Reflect tab in Claude’s Settings (web and desktop) showing a summary of your usage, most active day, peak hour, total chats, and a topic breakdown with percentages.

• Who gets it: Free, Pro, and Max accounts in beta, with memory turned on.

• What it recommends: fluency suggestions built from your own patterns, like moving a repeated task into a Project or a custom skill.

• The move below: find your most-repeated task and promote it from chat to system in one sitting.

Who this is for: anyone using Claude more than a few times a week, and especially operators who suspect they’re solving the same problem from scratch every Monday.

What Reflect shows you

Open Settings in Claude on web or desktop and pick the Reflect tab, then generate your report. The default view collates your last month, and you can widen it to 3, 6, or 12 months. At the top sits a paragraph summary of what you’ve been working through, followed by your most active day, peak hour, and total chats, then a breakdown of your topics with a percentage on the ones you hit most. The last stretch is the interesting part: a set of recommendations grouped around the 4D AI Fluency framework Anthropic built with academic partners, generated from your own patterns rather than generic tips.

Ryn Linthicum, Anthropic’s head of wellbeing policy, told Engadget the intent behind it: “We were really intentional about building [the dashboard] with an eye toward how we can upskill people’s usage of Claude, not in a way that encourages them to spend more time with it, but instead enables them to get more efficient at meeting their goals.” There’s also a quiet-hours setting and a break nudge in there, and the dashboard periodically asks questions like “What’s one thing you want to keep doing yourself, even if Claude could do it faster?”, which is a better question than most productivity content asks.

If you generate a report and get nothing, memory is off, that’s the dependency. Reflect only reads memory-enabled history, skips incognito chats entirely, and leaves out the underlying files from connected tools, so an inbox summary might appear in your reflection while the source emails never do.

Walk away with: your report generated on the 3-month view, which is long enough to show patterns and short enough to reflect how you work now.

How to read it: hunt the Context Rebuild Loop

Read the report as an audit, a Spotify Wrapped skim wastes the ten minutes. The topic percentages and peak hours are trivia. The signal is repetition: the task that shows up week after week, phrased twenty different ways, each time in a fresh chat where you rebuilt the context by hand. Every one of those chats billed you a setup cost, and the report is the first tool that shows you the receipt.

Once the report is on screen, run this in the same session:

Based on my usage patterns, answer three questions:

1. What task have I brought to you most often in the last 3 months?
2. What context did I re-explain across those chats that never changed
   (business details, format rules, tone, constraints)?
3. If I could only systematize one recurring task this month, which one
   saves the most repeated setup, and why?

Reflect’s own recommendation engine is pointed at the same target. Igor Bonifacic at Engadget ran it against his real usage, repeated research chats tracking down executive statements for a story, and the dashboard recommended building a custom fact-checking skill, which then produced a source-and-confidence template for every claim. His verdict: “I’ll admit I found the template helpful, and it wasn’t something I would have thought to ask Claude to do on its own.” That’s the shape of the value, the report surfaces the system you didn’t know you were already running manually.

Walk away with: one named task, the unchanging context behind it, and a reason it’s the first candidate.

Promote one task from chat to system

A repeated task can land in three homes, and the report tells you which one fits:

Where your repeated task goes:

Same task + same background context every time
  -> a Project (context lives there once, every chat inherits it)

Same task on a calendar rhythm (weekly report, Monday review)
  -> a scheduled task (it runs without you opening a chat)

Same steps applied to different inputs each time
  -> a custom skill or saved workflow (the steps live once,
     the inputs change)

The one-sitting version: take the task from your report, paste the context you’ve been re-typing into whichever home fits, and run it once from the new setup while the old chats are still fresh enough to compare against. The output quality usually jumps, because context written once and deliberately beats context re-typed from memory at 4pm on a Tuesday. My whole publishing operation runs on the third category, and the setup cost was paid one time.

Walk away with: one task moved out of the Context Rebuild Loop, and the pattern for moving the next one.

Where this wobbles

• It’s beta, web and desktop only for now, mobile is coming later, and the time-spent metric doesn’t exist yet. Linthicum says that’s because Anthropic wasn’t tracking it internally, a metric the product team “didn’t want to maximize.”

• Memory has to be on, so the analytics come bundled with a privacy tradeoff some people have deliberately declined. Sensitive conversations surface only at a high level, health-integration chats stay out entirely, and Anthropic says reflection data isn’t used for other purposes, worth reading their privacy note yourself if that matters to your business.

• The mirror is sold by the company that profits from what you see in it. TechCrunch's Sarah Perez makes the case that “Reflect’s larger purpose is about shaping how users think about AI itself,” and that deeper workflow integration “helps retain users and discourage them from switching to competitors’ AI tools.” She points back to Gmail Meter in 2012, an analytics tool that doubled as a demonstration of how central Gmail had become. Fair on both counts. The retention motive and the usefulness both live in the same feature, so take the audit and leave the sentiment.

• The insights are only as honest as the history behind them. If your real repeated work happens in tools Reflect can’t see, the report understates your loop instead of revealing it.

Recap

Reflect is free, live now, and takes ten minutes: generate the 3-month report in Settings, ignore the trivia, and hunt the task you’ve re-explained twenty different ways, that’s the Context Rebuild Loop and it’s been billing you setup time all year. Promote that one task into a Project, a scheduled task, or a skill, run it once from its new home, and you’ve converted the report from a curiosity into a system.

If what the report shows you is a longer list than one task, and you want a second set of eyes on which systems to build first, that’s the work I do with operators on an AI Clarity Call at muddventures.com/book.

P.S. A few next steps if today’s issue was useful:

• The setup this newsletter runs on: Work From Anywhere Just Got Tested by Two AI Labs

• The first Operator’s Read: 3 quiet AI shifts, and what I'd do about each one

• Last week’s GoHighLevel piece on scoped AI workflow edits: blog.muddventures.com/p/gohighlevels-ai-builder-just-learned

• Operators comparing builds like these every day: whop.com/abra-ai

• New here? The full archive lives at muddventures.substack.com

Back in your inbox tomorrow.

Andrew