AI adoption is a myth if usage is the metric
This is a useful enterprise AI framing: adoption dashboards can look healthy while real work stays manual because AI skill is a craft and usage is a weak proxy.
Links: Original source · Shared link · Related link
Logged at IST: 2026-08-08 16:05 IST
What it is: Mario Zechner recommends Vas's X Article, "AI Adoption is a Myth", saying he has seen similar patterns at smaller scale.
Gist: Vas argues that enterprise AI adoption metrics hide a barbell. In his telling, a rollout can produce 5-10% power users, roughly 20% weak users, and a large majority who barely use the tool at all. The dashboard still says adoption happened, but the organization does not get faster.
The stronger point is that using AI well is a craft, not a login event. The useful user knows when to clear context, when to turn repeated work into skill files, which parts of an automation need model judgment, which parts should be deterministic code, and how to read a diff before accepting it. A weaker user can paste a ticket into Claude and get a passing PR that quietly changes the wrong thing.
The piece also makes a spend-management point. If 10% of the organization burns 90% of the tokens, then broadening real power-user behavior may turn an eight-figure rollout into a much larger bill. That does not mean the rollout failed; it means binary usage metrics are the wrong control surface.
The proposed alternative is to split strategy by user type. Train everyone partly as diagnosis, give power users somewhere to publish and rank reusable skills, and for everyone else put automation into the background of systems of record. The board-level metric should shift from "adoption" to what share of work is manual, hybrid, or fully automated.
Newsletter angle: Good companion to the agent/tooling notes. It says the hard enterprise problem is not merely access to a model, but changing work design and measurement so AI does useful work without requiring every employee to become an AI-native power user.