Brigade Economics

Every lesson in this course has been slipping you the same note in different handwriting. Lesson 1: one cook by default. Lesson 5: parallel is faster, never cheaper. Lesson 6: the expediter bills its own runs. Lesson 7: checking costs cooks. Time to put the notes together and read the bill in full — because the multi-agent conversation happening everywhere right now has a loud half ("orchestrate everything!") and a quiet half (what it costs), and this kitchen serves both halves or neither.

Here's the plain economics: your subscription is capacity, and every cook draws from the same pot. A subagent isn't free labor — it's your same Claude, working more, reading its instructions fresh each time. Three parallel cooks ≈ three cooks' worth of capacity plus coordination overhead. Usage limits are real; hit them mid-afternoon because a five-cook pipeline ran three times, and the rest of your day is one very expensive lesson. Better to learn it here, on purpose, for free.

What you'll plate today

Your own usage, seen clearly; a cost intuition calibrated on your real call sheets; and the decision rule that makes crew-versus-cook a thirty-second judgment instead of a vibe.

Ingredients

• Your kitchen-brigade project — every bill your call sheets have stamped
• About 25 minutes, mostly thinking

Cook

1. Read your own bills

You've been collecting the data all course — the expediter's honest bill, the night shift's cook counts:

Audit this kitchen's spending patterns. From the call-sheet bills,
run logs, and heartbeats: which of our routines uses the most
cooks per run? Roughly how does one survey run compare to a plain
one-cook session in work performed? Rank our routines from
heaviest to lightest, and flag anything that runs on a SCHEDULE —
scheduled cost repeats whether or not anyone's hungry.

The ranking always surprises somewhere — usually the checking. A FULL QC pass (ticket-check + adversarial) can cost more than the work it checked. That's not an argument against checking; it's why the QC dial has settings, and why LIGHT exists.

2. Learn to see the meter

Now teach me to watch this myself, honestly:

1. Where do I see my actual usage and limits for my subscription
   plan? Walk me to the real numbers, whatever plan I'm on.
2. What are the SIGNS of running hot mid-session — the things I'd
   notice before hitting a wall?
3. Add the meter habit to CLAUDE.md: every call-sheet bill now
   includes a rough "this was a light / medium / heavy run" tag,
   so the ledger reads at a glance.

Go look at the real numbers — actually look, this is the lesson's field trip. Capacity stops being an anxiety and becomes a dashboard the moment you've seen where it lives. (Plans and limits change; the habit of knowing where your meter is outlives any specific number.)

3. The decision rule — this course's actual graduation exam

Write THE CREW RULE into CLAUDE.md, at the top of the crew section:

Before any multi-cook run, three questions:
1. INDEPENDENT? Do the parts pass the independence test? (No →
   one cook, sequential. Crews can't fix a job that doesn't carve.)
2. WORTH IT? Is the job big enough to feel the wait, or the
   stakes high enough to need cold eyes? (No → one cook. The
   wait IS the price signal.)
3. AUDITABLE? Will I actually read the receipts this run
   produces? (No → don't run it. Unread automation isn't
   productivity, it's noise with a subscription bill.)

Three yeses = crew. Anything less = one excellent cook.

Question 3 is the one nobody else will tell you, so hold it hard: the limiting resource was never tokens — it's your attention to the output. A crew that produces more than you can read hasn't made you faster; it's made you a manager who doesn't read reports. Every unread receipt is spend with no dish.

4. Prune like an owner

Apply the rule backward over everything we've built: for each
routine on the chart — survey, night shift, QC settings, the
specialists — does it still pass all three questions? Anything
that fails gets demoted honestly: lighter QC, less frequent
schedule, or retired to the handbook as by-hand-when-needed.
Update STATIONS.md. Save point: "brigade economics".

If something gets retired today, that's the system working — you built it, tasted it, and chose. A kitchen that only ever adds stations ends up cooking for its own machinery. Chefs prune; that's what makes them chefs.

When it burns

• You can't find real usage numbers for your setup — plans differ and interfaces move; the honest fallback is the behavioral meter: the run-tags in your ledger plus how often you feel the ceiling. Track two weeks of tags; that's a real dashboard too.
• Everything passed the three questions — suspicious! Question 3 deserves a re-ask with teeth: when did you last actually read the night shift's full receipt? Be the auditor of your own attention; that's the honest version of this lesson.
• You feel cheated — "so I should mostly use one cook?!" — yes, and that's not the course taking the brigade back. It's the difference between owning tools and being owned by them: you now run crews when they pay — fresh-eyes checks on things that ship, surveys too wide for one context, night prep you actually read. That's exactly the rare skill: most people either never touch crews or can't stop.
• Your subscription tier feels too small for your appetite — maybe! Now you're deciding from a ledger instead of a vibe, which is the only respectable way to upgrade. Run the heavier tier's math against your run-tags before paying, not after.
• A future model/plan changes all the specifics — it will, repeatedly. Notice what survives every such change: independence, worth-the-wait, will-you-read-it. The rule is portable; that's why it's the graduation exam.

Order up

□ Your routines ranked by real cost — scheduled spend flagged
□ You've seen your actual meter, wherever your plan keeps it
□ THE CREW RULE is law — including the unread-receipts question
□ One honest pruning pass; something got lighter or retired
□ Save point "brigade economics"

Next up — Lesson 10: Your Brigade. Graduation: design the crew your actual work deserves — stations, sheets, laws, and all — run it on a real week, and write the handbook entry that makes it yours for good.

Stuck on a step? Question box below.