WEEK 3 · MON 16 NOV 2026 · US AND THEM
Count your agents (you’ll be surprised)
How to run a first agent census in an afternoon, and what usually turns up.
Ask a room of leaders how many agents their organisation runs and you’ll usually get a shrug, a guess, or a number that turns out to be wrong. That’s not a criticism; agents are invisible by default. They live in tenants, chat sidebars, flows and scheduled jobs.
The vendors have noticed: Microsoft now ships an agent registry and calls Agent 365 a “control plane” for agents [S9]. When the platform builds an inventory, you know the inventory problem is real.
An afternoon census
- Pull what the platform knows. Export any agent or app inventory your admins can see.
- Ask the people. A two-question form: “What AI do you use that acts for you?” and “Did you build any?”
- Look for schedules. Flows, scripts and jobs that call AI services overnight are your autopilots.
- Write a card for each. Name, purpose, owner, data, and how much it does without asking.
- Put the cards on a wall, real or virtual, where everyone can see them.
What usually turns up
Duplicates (several teams building the same helper), orphans (the builder left), and a handful of agents doing far more than anyone thought. In our experience that’s normal. The point of a census isn’t blame; it’s getting everything onto the same floor. OUR VIEW
IBM found 63% of breached organisations had no AI governance policy or were still writing one [S2]. A census is the first page of that policy.
DRAFT SEED POST · PART OF THE 52-WEEK PLAN IN DOCS/04-BLOG-PLAN.MD