Nik Suresh, a consultant to large enterprises, publishes an article full of anonymous anecdotes about how AI mania is freezing decision-making in the organizations he works with. It’s worth reading because it tells something true that marketing pieces don’t say.

The clearest finding is an executive who admits they’ve never opened ChatGPT in their life right after signing off on a technical strategy entirely based on AI for a company with over 2 billion dollars in revenue. An engineer tells of asking AI to rewrite an entire repository from Go to Zig while he works on something else, just to “keep his seat” at a company obsessed with token leaderboards.

Why it matters to you. If you work at a company that’s “doing AI”, the problem isn’t the tool; it’s the social dynamic built around it. Suresh describes the mechanism with precision: executive clients announce 100x productivity gains, and vendors can’t challenge them because doing so would be seen as an attack, risking the enterprise contract and getting them fired. Hype becomes a silent pact where nobody has incentive to say how things actually are.

It’s the corporate version of “stop asking me to ask an LLM” that we covered on July 12: delegating judgment to something that can’t answer, but collectively and with a six-figure budget. The exact opposite of the method we propose here: test on your case, measure, decide with your own numbers. As the ROI metrics OpenAI published on July 17 reminded us, the way out of the fog is replacing claims with verifiable data. If you can’t measure it, you can’t even defend it to an executive who’s already decided what to believe.

In detail

Suresh’s article is entertaining and sharp, but its value doesn’t lie in juicy anecdotes: it describes a social mechanism, not a technology problem. Large enterprises that hire consultants fail because pressure to “do something with AI” has outpaced the ability to judge whether that something is needed, not because AI doesn’t work.

The case of the executive who’s never used ChatGPT but signs an AI strategy is the extreme end of a widespread pattern. Suresh tells it as an anecdote, but it’s the summary of something anyone working in large companies recognizes: the distance between budget holders and tool users has become so wide that budgets get allocated on narrative, not experience.

The second anecdote, the engineer rewriting an entire repository in another language, is the same disease seen from below. When the internal metric becomes “how much do you use AI” instead of “what did you produce”, people rationally learn to show AI activity instead of results. It’s not a matter of bad faith: it’s a matter of incentives.

The most interesting part is the vendor-client dynamic. A salesman tells Suresh that his enterprise clients’ executives publicly declare 100x productivity, and if he or a colleague tried to say those numbers aren’t plausible, the consequence would be contract cancellation. In practice, hype self-protects: those who’ve staked their credibility on an excessive promise can’t allow someone to debunk it, and those who depend on that client can’t afford to be the debunker.

This is the context for reading pieces that publish measurement frameworks, like OpenAI’s scorecard we wrote about on July 17. The question “how do I measure AI ROI” only makes sense if there’s someone willing to look at the numbers. If organizational dynamics reward those who promise and punish those who verify, no framework is enough.

The piece’s limits should be stated clearly. Sources are anonymous and filtered through a consultant with a viewpoint (and an audience to entertain). The anecdotes are plausible but not independently verifiable. Suresh doesn’t propose systemic solutions, and rightly so: the problem he describes is structural and won’t be solved by a tool. What you can do, in your own sphere, is keep verification channels open: use AI on your concrete cases, compare results, and when asked “does it work” bring your own numbers instead of your opinion.