Satya Nadella put something out on X a couple of weeks back that got past seventy-five million views, and most of the coverage boiled it down to “feed your company data into AI.” I think that misses the more interesting half of what he’s getting at.
His line was that “a frontier without an ecosystem is not stable.” In plainer terms: the big AI models are becoming a commodity, the same kit your competitor can buy off the same shelf. So your advantage can’t live in the model. It lives in the loop your people build between their own judgement and the AI they use.
And this is the bit with teeth. As the AI gets better, your people don’t get less valuable. They get more valuable. Somebody still has to decide what’s worth doing and point the thing at the right problem. Take that away and, in Nadella’s words, you’ve just got compute running in circles.
Then there’s a study out of China that backs it up. Twenty-six thousand students: the ones who let AI do their homework saw scores jump, then their exam results fell about twenty percent within six months. The ones who used AI to help them think barely took a hit. Same tool, opposite outcomes.
That’s the exact trade a team makes when it rolls AI out to do the work instead of to build capability. So stop fighting the model war. Build AI capability in your people. And measure whether you’re getting better, not just faster.
Sources & citations
- Satya Nadella, “A frontier without an ecosystem is not stable”, X, 14 June 2026.
- “The Generative AI Learning Penalty: Evidence from Chinese Secondary Education”, CEPR Discussion Paper DP21577.
