Manufacturing has a name for unsold inventory sitting between stations: work-in-process. It's treated as a cost, not an asset — every unit sitting on a shelf is capital that hasn't turned into anything yet, and the longer it sits, the worse it looks on the books.
Knowledge work doesn't have an equivalent term, but it has the equivalent problem. AI can generate drafts, analyses, dashboards, and communications faster than any team I've worked with could generate them manually. That looks like enormous productivity, right up until you ask what happened to all of it.
Some of it gets reviewed and used. A meaningful amount doesn't. Drafts nobody read twice. Dashboards nobody opened after the second week. Analyses sitting on someone's desk waiting for a decision that keeps getting pushed to next quarter.
I started counting it the way a plant manager would count pallets — not as a productivity number, but as a liability sitting on the floor. Once I started looking for it, I found it everywhere I hadn't been looking before.
The uncomfortable version of this note is that generating more of something faster isn't automatically progress. It's only progress once it moves — once it turns into a decision, an action, or a delivered outcome. Until then it's just a bigger pile, produced faster.
I don't think the fix is to generate less. I think the fix is to start treating unreviewed AI output the way a plant manager treats unsold inventory: as a number worth watching, not a number worth celebrating.
Current Hypothesis
Unreviewed AI output is a liability on the same balance sheet as unsold inventory, even though no accounting system currently records it that way.
Not a conclusion — a working idea, revisited as evidence comes in.