Writing
Behavior over kit: what your Claude Code setup says about you
Installed tools tell you almost nothing. Two people with identical configs can be four rungs apart — because the level lives in behavior: does Claude have to prove its work? How much runs in parallel? Who reviews what? I built a prompt that measures exactly that, from your own machine.
Why I built it
I needed to calibrate a talk for ~50 engineers, all on Claude Code, ranging from "daily helper" to "harness engineer". A survey would have been ignored — devs hate homework. So instead: a prompt they paste into their own Claude Code. Claude profiles its own usage, scores it against a six-rung ladder, and hands back a card. It's meta, it's two minutes, and it gives the dev something about themselves.
The ladder, briefly: L0 Explorer (occasional chat) · L1 Daily Helper (regular use, default setup) · L2 Power User (tuned settings, some verification) · L3 Agent Builder (delegates real work, authored some machinery) · L4 Harness Engineer (built and operates their own system) · L5 Force Multiplier (the team runs on what they built). Most engaged daily users centre on L2–L3 — and the card is calibrated to say so, neither flattering nor crushing.
Each card also maps you onto Boris Cherny's public five steps of AI adoption and names your one next unlock — the single bottleneck-breaking move for the step you're actually at. That's the part people keep.
Your rung, your path
A level without a path is a horoscope. Here's the accompagnement — what I'd have you read, build, and (if you want company) book, per rung:
| You're at | Read this | Build this next | With me |
|---|---|---|---|
| L0–L1 | What every team should know about LLMs | A real CLAUDE.md; plan mode as a habit | Foundations |
| L2 | plan → build → review | A self-verification loop — Claude proves its work | Power Hour |
| L3 | The harness, not the prompt | Your first lifecycle hook; living memory for one agent | 1:1 / 4-Pack |
| L4 | Getting yourself out of the loop | Evals, cost lanes, exception-only monitoring | Half-Day Deep Dive |
| L5 | everything above — then teach it | Make your harness the team's harness | CTO & longer missions |
If you run it, I'd genuinely love to see your card — sharing the anonymous JSON line back is optional, and only ever shows up as a distribution. The aggregate view (where a whole team stands, and its highest-leverage unlock) is the same instrument I use inside client engagements.