AI will not fix education. Thinking about education differently might.

I am building with AI for education, so I want to be careful here. What I am about to argue is not that AI is useless in education. It is that AI applied to a broken mental model of education produces faster, more scalable failure.

The mental model problem

The dominant mental model of education, embedded in institutions, assessments, and now in most EdTech products, is roughly this: students are containers to be filled with content, and the measure of success is how accurately they can reproduce that content on demand.

AI makes this model more efficient. You can personalise the content delivery. You can adapt the pace. You can generate more practice questions. You can provide instant feedback on whether the answer matches the expected answer.

None of this changes the underlying model. And the underlying model is the problem.

What a different model would require

A model centred on developing understanding would demand entirely different things from technology. It would need to make reasoning visible, not just outcomes. It would need to help teachers understand why a student is confused, not just that they are. It would need to track the evolution of a student’s thinking over time, not just their performance on discrete assessments.

This is what we are trying to build. I am not certain we will succeed. But I am certain that the alternative — faster content delivery to unreflective students — is not education. It is a simulation of it.