DevOps / SRE / Platform · 08.08.2026, 15:10 UTC
AI adoption isn’t the same as AI usage
| Schweregrad | info |
|---|---|
| Kategorie | DevOps / SRE / Platform |
| Quelle | The New Stack ↗ |
| Veröffentlicht | 08.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
Every engineering org I’ve talked to this year has some version of the same chart. Seat activations climbing. Token spend climbing. Somebody’s slide says 80% weekly active. Then you sit in that team’s planning meeting, and nothing about how they build software has changed.
That gap is the actual problem, and most of the way we measure AI adoption is designed, accidentally, to hide it.
Everything you’re measuring is a usage metric
Token spend tells you someone typed into a box. Pull request count tells you commits landed. “Percentage of code written by AI” tells you an autocomplete got accepted. None of it tells you the work got better.
Goodhart’s Law, right on schedule. The moment a number becomes the target, it stops describing reality and starts describing the incentive.
“The moment a number becomes the target, it stops describing reality and starts describing the incentive.”
The clearest version of this failure isn’t even AI-specific. Every engineer has watched a team clear a coverage gate with tests that assert nothing. expect(true).toBe(true). Line covered, number green, nothing tested. The metric was satisfied, and the goal was abandoned, and the dashboard has no way to tell the difference.
AI usage metrics have the same shape. If I tell my team that usage is a performance signal, usage goes up inside a week, and I’ve learned nothing except that they can read incentives. Which I already knew.
Better question: if you removed these tools tomorrow, what would break? Not what people would complain about. What would break. If the honest answer is “nothing, people …