AI mania is degrading decision-making at large companies

An executive at an organization with more than $2 billion in revenue presented a technical strategy entirely centered on artificial intelligence. Right afterwards, he confessed he had never used ChatGPT or any other AI tool in his life. The anecdote does not come from satire: it is told by Nik Suresh, a consultant to large companies, in his essay “AI Mania Is Eviscerating Global Decision-Making”, which Simon Willison recommended on July 19 on his blog.

Willison, one of the most widely read voices in the AI tooling ecosystem, describes the piece as “an entertaining perspective” on the mania overwhelming the large companies Suresh consults with, and warns that it comes “crammed with spicy anecdotes from anonymous sources”. The original essay deserves a full read; here we go over what it reveals about how technology decisions are being made in 2026.

The token leaderboard as a productivity metric

Among the collected testimonies, one stands out for what it says about internal incentives. An engineer explains that his company keeps a token consumption leaderboard, a table ranking employees by how much AI they use. His rational response to that system: “checking out a parallel copy of our Go repository and telling the AI to rewrite the whole thing in Zig while I work on something else, just so I can keep my job”.

It is a textbook case of Goodhart's law: when a measure becomes a target, it stops being a good measure. If AI usage is rewarded by volume, teams will produce volume, even if that means burning tokens on rewrites nobody will ever review or deploy. The cost is double: money spent on worthless compute and a polluted data signal that management will read as enthusiastic adoption.

Decisions without first-hand experience

The case of the executive who never opened ChatGPT points to a problem deeper than waste: entire corporate strategies are being drafted out of competitive panic rather than evidence. Suresh also recounts a conversation with a skeptical executive inside a company overexcited about AI, a sign that even in the most enthusiastic organizations there are people who see the problem and cannot find a way to stop it.

None of this is exclusive to any single company. Suresh writes from his experience consulting for multiple large organizations, and the pattern repeats: committees approving multimillion budgets for AI initiatives without anyone in the room having spent an afternoon trying the tools they are about to buy.

One important nuance is worth underlining. Neither does Suresh write from a rejection of the technology, nor does Willison, who has spent years documenting productive uses of LLMs, recommend the essay out of cheap skepticism. The criticism is not aimed at the models: it is aimed at management. Budgets approved on the back of headlines, adoption mandates without training, and activity metrics that mistake spending for results.

Who this is useful for

The piece is working material for three profiles. For CTOs and engineering leaders under board pressure to “do something with AI”, it works as a catalog of antipatterns. For consultants, as an uncomfortable mirror of what they see every week. And for technical teams trapped in absurd mandates, as confirmation that the problem is not them and that there is precise vocabulary to name it.

The alternative metrics exist and they are boring: code in production, cycle time, defect rates, satisfaction of the teams using the tools. None of them fits into a weekly token leaderboard, and that is precisely why they are harder to game.

At ElephantPink we use Claude and other AI tools daily, and that is exactly why this essay feels necessary: adoption works when it is driven by people who actually use the tools and measured by delivered results. A token leaderboard measures API bills, not productivity.