Much of what has been on my mind this week comes back to a single question: what are we really teaching, and building, when we talk about AI? I want to look at how AI is loosening the old constraint of scarce human time, why the skills we emphasise for students may be aimed at the wrong target, and why we keep confusing learning to live with AI with learning to build it.
From resource scarcity to personalised abundance
Almost everything we have built in business, education and healthcare has been shaped by one constraint: there is only so much of a person's time to go round. We ration attention, triage need, and design systems around the assumption that expertise is expensive and finite. AI is starting to remove that constraint, and the implications reach well past efficiency.
I explored this in a recent piece on what happens when personalisation at scale stops being a theoretical promise and starts reshaping how value gets created. In a classroom, it changes what a single teacher can attend to. In a business, it changes where the bottlenecks sit. What interests me most is the leadership question: when you are orchestrating abundance rather than rationing scarcity, the instinct to control the flow becomes a liability. The harder skill is deciding what all that freed-up capacity should be pointed at.
What skills actually prepare students for AI
EdSurge published a thoughtful piece on what skills actually prepare students for the future. It draws on a two-year research project which found that most teachers, including those in computer science and engineering, still cannot identify a clear instructional use case for widespread AI integration. Its recommendation is to focus on computational thinking rather than tool-specific skills like prompt engineering.
I agree with much of this. Decomposition, pattern recognition and algorithmic reasoning are genuine foundations, and they age better than any particular tool. But they sit within a "how AI is built" frame, and on their own they are not sufficient. The missing part is the other 80 per cent: how to evaluate what AI gives you, how to hold on to your own judgement when the tool is confident and wrong, and what AI actually means for the work students will go on to do. A curriculum that only teaches how AI is built is like a driving lesson that only covers engine mechanics. Useful, but not the thing that keeps you safe on the road.
AI literacy is not the same as AI science
Most freely available AI curricula for schools were designed to teach students how AI is made, not how to live and work with it. The major K-12 frameworks are consistent on this. The AI4K12 initiative organises its curriculum around five Big Ideas, four of which are fundamentally about how AI systems work internally, with only the fifth addressing societal impact. These are well-intentioned programs, but many were built before ChatGPT existed, when almost nobody interacted with AI directly. When you teach AI primarily through neural networks and algorithm design, you advantage students already strong in maths and science, and you quietly reframe AI education as a feeder for the computer science pipeline.
Shuchi Grover's distinction between "AI education" and "AI literacy education" is useful here. A student who understands that AI learns patterns from data, that those patterns can carry bias, and that outputs are probabilistic rather than authoritative has enough foundation to use these tools well, without needing backpropagation. Cynthia Breazeal's counter-argument matters too: students need to grasp that AI is something people make and control, or it hardens into an unquestionable black box.
The resolution, I think, is a clean separation:
- Every student needs AI literacy — critical use, evaluation, prompt craft and ethical awareness.
- Students who want to build AI additionally need AI science — the internals, the maths, the model design.
Too many frameworks bundle these together, forcing students to learn AI science in order to gain AI literacy. That order is backwards, and it locks out the very students who most need to be confident users rather than reluctant bystanders.