Is AI making our people faster, or changing how they learn?
One answer means the shape of the organisation distorts and reverts. The other means it never goes back, and this year's graduate intake is a bet on which one you believe.
Why this question
Every technology I have backed arrived on time and mattered late. The gap between the two was my job for twenty years, and for the first time I am not confident how long it is. That matters because the thing being automated first is entry-level work, and entry-level work was never only work. It was how a junior became someone whose judgement you could bill for.
So the question is not whether AI makes people faster. It does. The question is whether, in making them faster, it has removed the way they learn. A productivity shock changes the org chart for a while. A learning shock changes it for good. The evidence so far, and my own experience of measuring this inside a firm, says learning.
- The lag. Capability arrives on schedule; the effect on a P&L arrives years later; the gap is closing. Why the question is urgent now.
- Two shocks. Productivity or learning. The distinction that decides whether anything reverts. The hinge of the question.
- The curriculum. Sort every entry-level task by what it is worth and what it taught. The decision.
- The shape. Two middles changing places, and a ceiling on the new one. The consequence for the organisation.
- The prediction. Leverage peaks and turns down between 2029 and 2031, or I was wrong. The date.
- The Lag Curve. Margin rises now; capability falls later; nobody sees both in one report.
- The Curriculum Grid. Value of the output against judgement taught. The right-hand half is where the learning shock lives.
- And the paper's diamond figure, in The shape: the middle everyone assumes, and the narrower one the evidence allows.
This question decides whether a firm can keep producing the judgement the other two questions depend on. A firm that keeps people deciding while agents draft, keeps a deliberate share of the entry-level work for people, and gives the people who supervise agents a real role is building that judgement. A firm that automates the base and staffs the middle to a hundred per cent is spending it.
Most firms aren't using AI at all (The lag; conceded in part). AI is an apprenticeship accelerator, not a destroyer, and Jevons: cheaper analysis means more analysis (Two shocks; both conceded in part). The base is being up-levelled, not removed (The curriculum; conceded in part). The diamond framing isn't yours (The shape; conceded). Leverage is at a seventeen-year high, so the thesis is already wrong (The prediction; held). Each is answered in full inside the concept it attacks.