MIT’s Project NANDA studied real deployments and found that 95% of enterprise generative-AI pilots have produced no measurable return, despite $30–40 billion invested. McKinsey’s November 2025 survey says the same thing from the other side: 88% of organizations use AI in at least one function, but only 39% can tie any EBIT impact to it.
That gap is not a model-quality problem. Generation is already excellent. What’s expensive is everything downstream of generation: deciding whether the output is right, framing the problem well enough that the output is worth having, and owning the consequence when it’s wrong. Call it the verification tax — and it’s why cheap generation raises, not lowers, the price of human judgment.
Bill Malloy 