The Absorbers

Every AI deployment produces some amount of wrong output, and something has to catch it. Better models reduce the amount. They do not remove the need for somewhere to put what is left.

The absorber is whatever is already positioned to catch it. It is why deployments that look nothing alike survive for the same reason, and why deployments with better models sometimes stall.

The three

A reviewer. Detection is assigned to somebody. A clinical note gets signed. Generated code goes through review and a test suite. The variance arrives in front of a person whose job is to catch it, and that person was staffed and budgeted before AI showed up.

A bounded answer space. Variability is constrained before detection is needed. Support on a documented product. Routing against a fixed taxonomy. Extraction against a known schema. A narrow domain does not need a reviewer because it does not generate much to review.

Cheap wrongness. Detection can be external or delayed, and both the consequence and the correction cost are low. Cheap does not mean harmless. An answer about a refund, a contract term, or a medication is not cheap wrongness, and a wrong posting to the general ledger is neither noticed nor cheap.

The fourth

Reversibility. It is not on the original list because the three above were built from the language side, where nothing has happened yet and the worst case is a wasted answer.

Once output becomes action, an undo has to exist. It is the absorber an organization is most likely to have to build rather than find, which is why scoping for action is construction where scoping for language is search.

Two things that do not work

Removing an absorber without noticing it was there. And failing to build one that will clearly be needed.

This is not an argument for preserving manual review, which is sometimes exactly the waste worth removing. It is an instruction to find what currently absorbs the error before deciding what to do about it.

They are not equals

Manufacturing settled this long ago: you cannot inspect quality in, you have to build it in. A reviewer inspects. A reconciliation sweep inspects later. Only the bounded answer space builds it in, by constraining the process so there is less to inspect. Take the bounded domain where it is available, because it is the only absorber that does not charge you every time it runs.