The Orchestrator

This part of the book is meant to work like an idea book, so here is a recipe book: nine experiments an organization can run without a transformation program, a reorg, or permission from anyone more senior than a director. Each is small, each is measurable, and each teaches the institution something about itself.

1. Run the meeting autopsy. Take last month’s recurring meetings and classify each one against the seven jobs from Section 10.3. Any meeting doing only describing, guessing, or assigning is a candidate for deletion. Shopify ran the blunt version: in January 2023 it deleted every recurring meeting with three or more attendees — about twelve thousand series — reinstated no-meeting Wednesdays, and told teams to recreate only what they could justify. Meeting time per person fell 14 percent in five months.⁵ “The best thing founders can do is subtraction,” CEO Tobi Lütke said at the time. “It’s much easier to add things than to remove things.”

2. Put a price tag on the room. Shopify’s meeting-cost calculator shows the estimated cost of a meeting inside the calendar as you schedule it, and the number climbs with every attendee. “Your calendar is a strategy,” COO Kaz Nejatian told staff. You do not need the tool to run the experiment: take your worst weekly meeting, multiply the headcount by the length by the loaded cost of the people in it, and put the number next to the overnight preparation run that could have replaced it. The dollar figure is a guess; the habit change is not.

3. Flip one meeting. Pick the most painful investigation-assignment meeting you own. Have the system prepare the evidence first — read-only access, prototypes in a test environment, nothing touching production. Run the same meeting against the pack. Measure the length, the decision, and how the room felt. This is the whole method in one experiment, and if the room does not get better, find out why before doing anything else in this book.

4. Cancel the status meeting; publish the board. If work state lives in a task system the systems update as they go — Chapter 16 again — a status meeting is a human reading a database aloud. Cancel it, publish the board, and see who complains. Whoever complains is telling you what the status meeting was actually doing for them, and that is the requirements list for its replacement.

5. Stop estimating delegated work. For work a system will execute, story points measure nothing you care about. Replace the estimate with the goal in three sentences, the exclusions, and the evaluation case. The estimate was a habit; the goal is the thing the system actually needed.

6. Staff the review queue. Name the humans who review the system’s output, give them protected hours in the plan, and measure the queue like any production queue: arrival rate, latency, oldest item. The common failure of early orchestration is not “the agents produced nothing.” It is “the review queue quietly grew until everyone started rubber-stamping.”

7. Write the one-page approval map. Levels 0 through 3, owners by name, boundary conditions in plain language — new data purpose, new region, irreversible money movement, public promises. Write it before the first boundary-crossing proposal arrives, because the first proposal is when you discover whether the map exists.

8. Keep a witnessed-decision list. Write down which decisions will always be made in a room, by named people, on the record — refunds above a threshold, customer promises, anything with a regulator’s phone number attached. Publishing the list is the institution telling itself the truth about where its judgment lives.

9. Give the facilitator the new job. If you have a scrum master or project manager whose calendar thinned out, hand them the delegation loop: the approval map, the review queue, the fence-hit reports, the evidence-review meetings. It is a real senior job, and it is the cheapest way to keep an experienced human engaged with the new work instead of resenting it.

None of these requires the institution to believe anything in this book. Each one produces evidence — shorter meetings, faster decisions, a review queue with a visible owner — and evidence is what converts a skeptical manager, because it is the same thing that converted the skeptics of every previous shift: the numbers on their own calendar.

The room was the first institution to change. The org chart is next: what organizations keep, what they stop pretending, and how ninety days of transition actually plays out — the subject of the next chapter.

HQ 6 — Assembled. The human and AI each wrote portions of this chapter. I assembled, reviewed, and take responsibility for the whole; the voice and arguments are mine, and I know which parts are which.