The McKinsey AI Report Was Probably Outdated When They Hit Print
Ethan Mollick is mostly right: the advice is fine, the models it's calibrated to are gone.
Consulting firm AI reports remain calibrated to worlds that stopped existing during the writing process. The genre persists. The recommendations are still benchmarked against last year's models.
There is a genre of document — the Major Consulting Firm AI Report — that arrives polished, heavily footnoted, and calibrated to a world that stopped existing while the document was being written.
McKinsey dropped one. Mollick pushed back. The note I scribbled while reading both was just: they did all this work benchmarked against last year's models.
That's the whole thing. The recommendations aren't wrong exactly — build internal workflows, reduce friction here, automate that — but by the time you finish the executive summary, o3 can already do the department-level task the report was teaching you to train humans to do better. The gap closed. The advice didn't.
This is the structural problem with consulting on a technology that compounds quarterly. The research takes eight months. The models take three. You ship a report on how to get value out of GPT-4-class tools the same week the next tier drops and makes the framing obsolete. Your roadmap assumes a ceiling that isn't there anymore.
Mollick called it. There are good bits in the McKinsey report — there always are, the people writing these aren't dumb — but the frame is wrong, and a wrong frame makes good details useless.
The uncomfortable version of this: most organizations are still building for last year's baseline too. The McKinsey report isn't behind the times because McKinsey is slow. It's behind the times because everybody is slow and the reports just make it legible.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.