Claude Reflect Review: Pretty Analytics That Miss the Point

Today I tested Claude's new Reflect feature from Anthropic. In theory, it's supposed to help users understand their AI usage patterns. After 30 minutes with it? Just confusion.

The charts are visually polished, sure. I can see what percentage of queries were for coding versus text analysis. But why do I care? As an ML engineer, I need latency metrics and response quality data—not a breakdown showing I spent 37% of my time on "work tasks."

Compare this to GPT-5.6 Sol's unlimited mode that I tested last week. At least there's clear value—you can run heavy prompts without token anxiety. This? Decorative analytics at best.

The Metrics That Matter (And Aren't Here)
Reflect shows:
Category distribution (code, text, ideas)
Time-of-day activity
"Emotional tone" graphs for queries
But where are:
Average response times by query type?
Refusal rates (when Claude says "I can't")?
Answer quality comparison between early and late context?
It feels like Anthropic copied productivity tracker features but forgot that AI tools need actionable insights—not just pretty visualizations.
Maybe I'm just tired of feature-driven marketing. New tools should solve real problems, not create the illusion of progress. Does anyone actually find this analytics useful?