navigateenter openesc close
Signal · Method

Adoption is the wrong thing to measure

Use is near universal and tells you nothing. The question that matters is what stops working when you switch it off.

JA
Joseph AbrahamFounder and Principal Analyst
5 September 20265 min read

There is a difference between an employee pasting text into a chatbot and a model approving credit at three in the morning with nobody present. Almost no published research distinguishes between them.

That is why the same field can report near-universal uptake and near-universal failure in the same week without anyone noticing the contradiction. Both numbers are correct. They are measuring different things and calling both of them adoption.

One question, asked of every system

If this were switched off tomorrow morning, what actually happens to the work?

The answer sorts every deployment onto a ladder. At the bottom, nothing happens, because nobody downstream was depending on it. At the top, an external commitment stops being met.

The interesting line is not at the top. It sits between level three, where a human reviews every output before it takes effect, and level four, where the process degrades without the system and no manual fallback is maintained.

Organisations cross that line without noticing

Nobody signs off on becoming dependent. A tool that was optional becomes habitual, the manual alternative stops being practised, the people who knew the old process leave, and one day the fallback exists only on paper.

That is why the governance question always arrives late. It is not that anyone ignored it. It is that the moment it became relevant was never marked.

Cite this asGlobal AI Forum, Adoption is the wrong thing to measure, 5 September 2026.