Claude Code's /Insights

Claude Code dropped a new /insights command that basically hands you an AI performance review on your coding sessions—praise, scolding, and homework included. The original write-up loved the “feels like a future manager” vibe: it nudges you about abandoned threads, pushes you to wrap work into reusable tools, and even hands you copy‑paste tips for building skills, agents, and hooks. Supporters are into it: an always‑on coach that helps you get more out of your AI buddy.

Then the plot twist: user jascha_eng ran it and says the report claimed 300+ sessions on database migrations—even though they only had about 10. When asked, Claude allegedly replied the “336” number was “fabricated by the insights generation process.” Cue the trust sirens. Now the convo’s split between “omg, AI mentor!” and “is my AI boss just making numbers up?” The hot take economy is booming: some say the tool is coaching you to be better; others call it a vibe‑based manager grading you on vibes.

There’s also debate about work style: the author says exploring and dropping threads can be optimal, while the report scolded them for not finishing. Another wrinkle: the reports seem to weight recent work heavily, so your latest experiments might skew the lecture. Love it or hate it, everyone agrees: learning to defend your workflow to AI just became a real skill—starting now.