Two years ago, a piece called “I Will Fucking Piledrive You If You Mention AI Again” circulated widely. The author — a consultant who blogs at Ludicity — was frustrated at hype, and the piece landed because it was specific where most criticism was vague.
Yesterday they published a follow-up: “AI Mania Is Eviscerating Global Decision-Making.” It’s worth reading in full, but the core observation deserves to be stated plainly: across 18 months of consulting engagements, the author’s team has observed zero successful AI implementations. Not a low rate. Zero.
This isn’t a capabilities argument. The essay doesn’t claim LLMs are useless. The problem the author is identifying is structural: organizations have built incentive systems that make it impossible to observe, measure, or report failure honestly.
The specific mechanics are worth noting:
One client’s engineering team, under pressure to show AI value, set up LLMs that were prompting each other in loops — producing token consumption that looked productive in dashboards while generating no usable output. The metrics were performing; the product was not.
A Fortune 500 executive was directing a $2B+ revenue strategy organized entirely around AI. When the author asked a few direct questions, it became clear the executive had never used ChatGPT. They didn’t know what the thing they were betting the strategy on actually did. They knew the strategy required AI. The knowledge stopped there.
The most interesting structural diagnosis is about coordination. The author argues that many executives know, privately, that their AI projects are failing. But publicly admitting it is individually career-ending, because you’re undermining peers who have publicly committed to the same strategy. There’s no mechanism to coordinate admission of failure. “If they could all admit the truth at once there might be some hope, but there is no way to coordinate that event.”
This is a recognizable social dynamic — it’s the same structure as most speculative manias — but the organizational version is particularly difficult to resolve. A market corrects when prices move; an organization corrects when someone in authority changes the internal narrative. In an environment where AI commitment has become a loyalty test, the person most qualified to start that correction is also the person who stands to lose the most by doing it.
The essay is, understandably, angrier than this summary makes it sound. The author has been doing real consulting work for 18 months, watching real projects fail, while clients around them escalate investment. Watching a thing not work while the people around you insist it’s working is its own particular kind of frustration.
What I find most useful about this piece isn’t the zero-success-rate number — single-firm consulting data is inherently anecdotal — but the typology of failure modes. Games metrics, demos that buy without tracking outcomes, projects that get labeled AI after the fact to satisfy organizational ideology. Each of these has appeared in previous boom cycles. The specific version of each is different, but the structure is familiar.
The essay doesn’t offer a solution, which is probably honest. You can’t individually exit a coordination trap; you need either an external shock or a trusted channel for everyone to admit the same thing at once. Neither is available on demand.
What you can do is measure something real. Token consumption is not a business outcome. If your AI implementation doesn’t have a metric that changes if the model stops working — something downstream and concrete — then you’re not measuring whether it’s working.