The Hype Is Crowding Out the Boring AI That Works

Nobody should be pitching an AI project. They should pitch how a piece of technology creates value, and nobody should care whether it's generative, narrow, or a spreadsheet. The minute "is it AI?" becomes the question, you've lost the thread, and the current hype cycle has made that question impossible to dodge.

That's the real cost when a former game-studio AI exec calls generative hype "poisoning the well" for traditional machine learning. He's got a prior, narrow ML is his life's work. But the procurement behavior he's describing is real: the recommendation engines, the fraud models, the demand forecasting, the support-automation that quietly took a capture rate from the low-60s into the mid-90s, all of it was delivering measurable ROI before ChatGPT, and none of it survives a budget conversation that only rewards what sounds transformative.

"We took it from three days to fifteen minutes" should win every time. Right now "we're using AI to transform the experience" wins instead.

The ground-level footprint is far smaller than the story. Yale's Budget Lab found AI usage has "no connection" to US unemployment rates, which is a tidy way of saying the actual work being touched is more mundane than anyone selling it wants to admit. So enterprises are deciding against a narrative, not against evidence. The proven thing loses to the exciting thing because the exciting thing photographs better in a board deck.

I don't know how long it takes for the math to reassert itself over the pressure. But the tell is simple. If a team can't name what got better and by how much, the way that support number went from low-60s to mid-90s, then the hype is doing the deciding, not the ROI.