Your Brigade

Nine lessons ago, "multi-agent" was a conference word, and every course teaching it wanted you to learn Python first. Since then you've hired fresh eyes, written the chart, equipped the pantry with receipts, named a specialist, fired parallel plates, written call sheets, wired in checkers, run a night shift behind a fence, and read the honest bill. Piece by piece — always on real work, always with the counterweight in the other hand.

Graduation is the same move every course in this kitchen ends on: no recipe. You design the brigade your actual work deserves — not the maximal one, the correct one — staff it, run it on a live week, and prune it like an owner. The dish is a kitchen.

What you'll plate today

A brigade of your own design: charted, staffed, law-bound, economically pruned — proven on one real week of your work, and documented on a single page that makes it survivable past your enthusiasm.

Ingredients

• Everything — nine lessons of it
• Your real workload for the coming week (the brigade's shakedown cruise)
• 2–3 hours across the week, mostly at the start

Cook

1. Design from the work backward (the only direction that works)

Not "what agents would be cool" — what does your week actually contain? You have the evidence: your journal, your CHORES scoreboard if you cooked Prep Station, nine lessons of run logs.

Design my brigade from evidence. Read the kitchen's history —
journal, logs, call-sheet bills, chores — and propose BRIGADE.md:

1. THE WORK: the 3-5 recurring shapes my real work takes
   (survey-ish, draft-ish, check-ish, prep-ish...).
2. THE CREW: for each shape — one cook, a specialist, or a call
   sheet? Apply THE CREW RULE ruthlessly; I want to see "one
   cook, sequential" beside anything that doesn't earn more.
3. THE SCHEDULE: what (if anything) genuinely deserves the night
   shift, under the law.
4. THE BILL: rough weight of this design in light/medium/heavy
   runs per week — and what I should expect to actually READ.

Interview me first — five questions — where the evidence is thin.

Push back on the proposal like the owner you are. Every station must point at a work-shape you recognize; anything that exists because it's clever gets cut now, cheaply, on paper.

2. Staff it and make it law

Approved with my edits. Build it: the specialists with job
descriptions I'll read, the call sheets with checkpoints where
I still want hands, QC dials set per the law, the night shift
(if any) fenced and receipted. Everything on STATIONS.md.

Then the graduation test — the cold-cook drill from Lesson 2,
on the WHOLE kitchen: a fresh subagent reads only the handbook
and explains back to me how this brigade works and when each
station fires. Where it stumbles, our handbook lies. Fix until
a stranger can't get it wrong.

That drill is your quality bar made final: a brigade that lives in your head disbands the week you get busy. A brigade that lives in the handbook survives — and can someday be shown to someone, which is how every kitchen in this story grows.

3. The shakedown week

Run your actual week through it. Real orders, real plates, no demo data. Hold three disciplines while it runs:

• Read your receipts — question 3 of THE CREW RULE is now a lifestyle. Unread automation gets retired at week's end, no sentiment.
• Calibrate in the files — every miss becomes a job-description edit or a call-sheet line, not a longer prompt shouted in the moment. That's the manager's move; you've had it since Lesson 4.
• Log the saves — each time the brigade catches something or hands you an hour, one journal line. Friday needs evidence, not vibes.

4. Friday: the owner's page

Cook the shakedown report from this week's logs: what ran, what
it caught, what it cost, what I actually read, what got pruned.
Then write the owner's page at the top of BRIGADE.md — "what the
chef learned": three things I know about running crews that
Lesson 1 me didn't. Ask me; don't invent them.

Save point: "my brigade" — and push it to the cloud, like
everything that matters.

Answer the three-things question honestly. If one of your answers is some version of "most jobs still deserve one excellent cook, and now I know exactly which ones don't" — that's not the course failing to sell you its own topic. That's the judgment that was the actual product. It's rarer than the orchestration, and you built it on evidence.

When it burns

• Design paralysis — the blank BRIGADE.md stare — you're designing from imagination instead of evidence; the interview exists for this. Worst case, start with the two stations every knowledge-work kitchen earns: The Inspector on things that ship, one survey sheet for things that pile. Grow from real stalls, like the pantry taught you.
• The shakedown week got busy and the brigade sat unused — the most honest finding possible: your real week may need fewer standing crews and better one-cook habits. Prune to what got used; what got used is your true brigade. (This outcome is a pass, not a fail. The certificate doesn't require headcount.)
• Something broke mid-week and you routed around it instead of fixing — Friday's report will show it; put it back through the calibration loop or retire it explicitly. A kitchen with one quietly broken station teaches the whole crew that broken is normal.
• You want to show someone — that instinct is the whole ecosystem this site runs on. The handbook the cold cook validated is exactly showable: walk a colleague through BRIGADE.md the way Prep Station's Lesson 10 hands over a butler. Their first question will improve your handbook; it always does.
• "What's next after this?" — nothing is next; things are adjacent: point the brigade at an app with users, at your office's chores, at whatever the recipes are cooking this month. The courses end; the kitchen doesn't.

Order up — the final pass

□ BRIGADE.md designed from evidence — with "one cook" written where it's true
□ The whole kitchen passed the cold-stranger drill
□ One real week: receipts read, files calibrated, saves logged
□ Friday's report and the owner's page, in your own words
□ Save point "my brigade", mirrored where laptops can't kill it

After the brigade

Three courses of this kitchen end the same way on purpose: something real, running, with your name on the handbook. What's different this time is scale — you're no longer a person who uses an AI, or even one who builds with an AI. You run a kitchen. Stations you designed, laws you wrote, a bill you can read, and the judgment to know when all of it should stand down for one excellent cook.

• The recipes keep cooking at the edge — new appliances, new techniques, honest tastings.
• Subscribe and they find you.
• And show us your BRIGADE.md at @AgentKitchenHQ — charts, bills, pruned stations and all. The kitchens graduates build decide what this one cooks next.

Service — full brigade. 👨‍🍳