What to Study When AI Can Do the Work

My son is choosing what to study, and where. And I find myself wanting to say something to him, and to every parent and every leader having a version of this conversation right now, that pushes back on a narrative I keep encountering.

The narrative is: AI can do so much of the work now, why grind through four years and a significant debt load? Why specialize in something that might be automated? Why not stay flexible, learn to use the tools, and see what the world looks like in five years?

I understand this reasoning. But the neuroscience does not support it, and I think it answers the wrong question. The real question is not whether university is worth it. It is whether the hard, effortful work that builds a brain gets done at all, because that is what these years are for, and AI now makes that work easy to skip. University is the most common place that work happens. It is not the only one. Whatever path a young person takes, these are the years the architecture gets laid down, and the work is what lays it.

The developmental window

The adolescent and early adult brain is not a smaller version of a mature brain. It is a brain in active construction.

Dr. Jay Giedd, whose longitudinal MRI research tracked brain development across hundreds of young people over two decades at the NIH, documented something that changed how neuroscientists think about adolescence. Cortical gray matter peaks before adolescence and then thins via synaptic pruning through the teens and into the 20s. This pruning is not loss. It is refinement. The connections that are used are strengthened. The ones that are not used are eliminated. Use it or lose it, at the cellular level.

The prefrontal cortex, the region responsible for executive function, judgment, long-range planning, and impulse control, is among the last to mature. Myelination of prefrontal pathways, the process by which neural connections become faster and more efficient, continues into the mid-20s.

Different cognitive capacities peak at different times. Hartshorne and Germine's 2015 study in Psychological Science, drawing on 48,537 participants, found that processing speed peaks around age 20. Working memory peaks around 30. Vocabulary continues developing until around 50. The "brain matures at 25" shorthand is too simple. Different systems develop on different timelines, and the early 20s are an unusually rich window for some of the most consequential ones.

What deep specialization builds

Here is what gets lost in the flexibility argument: what you engage with deeply in these years does not just give you knowledge. It builds the architecture your brain uses to process complexity for the rest of your life.

My bachelor's degree was in electrical engineering. Mathematics, physics, vectors, matrices, calculus. I do not use those tools in their original form every day. But the way I approach problems, the instinct to find the underlying signal before responding to the noise, the habit of looking for structure in systems before looking for solutions: that came from years of training in the frameworks electrical engineering requires. Those frameworks are still operating, decades later.

A student who goes deep in economics builds different frameworks: how incentive and constraint shape behavior, how to think at scale. A philosophy student builds others: how to hold a claim, examine its foundations, trace its implications. The subject differs. The principle is the same. Deep specialization during the years when the prefrontal cortex is being refined is how the brain builds the architecture it will use for everything that follows.

Some young people will build that architecture without a formal degree. They are genuinely self-directed: they seek hard problems, find their own mentors, and do the rigorous work on their own initiative. That is real and worth acknowledging.

Most are not wired that way, and expecting them to be is wishful. What university provides, at its best, is structure: a curriculum that pushes you into subjects you would not have chosen on your own, peers who challenge your thinking in real time, and experienced people who can show you what mastery looks like and help you get there faster than you could alone. For most 18-year-olds, that scaffolding is what makes the difference between actually building the architecture and assuming they will build it later.

AI can answer questions. It cannot build that architecture for you. The architecture comes from the struggle.

The social brain

There is a second developmental process happening in these same years that almost never comes up in the "is university worth it" conversation.

The social brain, the networks that govern how you read other people, navigate conflict, build trust, and form lasting bonds, is also in active development through adolescence and into the early 20s. Dr. Sarah-Jayne Blakemore, whose research published in Nature Reviews Neuroscience documented this, found that relational capacity is being actively shaped during this period: how you repair after conflict, how you bond across difference, how you build the kind of friendships that last.

The conversations at 2 a.m. in a dormitory room. The conflict with a roommate and the working-through. The cross-cultural friendships formed during a window of unusually high relational plasticity. These are not extracurricular to the education. They are part of what the brain is building.

When we chose the high school for our son, we chose it partly for this reason. The program is rigorous, but we also wanted him somewhere the bonding was real, the conflict was real, and the resolution had to be real. The capacity for humanness is being built in these years. AI fluency is not a substitute for it.

What the AI research says about young people specifically

The research on AI and cognition has something specific to say here.

Gerlich's 2025 study, examining 666 participants across age groups, found that younger people are the most vulnerable to cognitive offloading from AI use. Lee and colleagues at Microsoft Research and Carnegie Mellon found that the strongest protective factor against AI-induced decline in critical thinking is self-confidence in your own cognitive ability.

Young people who skip the deep work are not just missing knowledge. They are missing the experience that builds cognitive self-confidence. And without that confidence, AI fluency becomes a crutch rather than a tool. The students who benefit most from AI are the ones who have already done enough rigorous work to know when the output is wrong.

The both/and path

I am not arguing for a rigorous degree and nothing else. I am arguing for a specific combination.

Pick something that trains your brain hard: physics, economics, engineering, philosophy, whatever requires building frameworks from first principles and working through complexity that does not resolve easily. This is what builds the prefrontal architecture during the window when that architecture is being laid down.

Pair it with something practical and human-centered: the kind of work where presence is the point, and AI can help behind the scenes but cannot stand in for you. Often that work is local, hands-on, and relational. The skilled trades, where judgment lives in the hands. Care and coaching: looking after the elderly, working with young people, running a team or a camp. Building and running things in the real world, events, hospitality, the outdoors. The common thread is that the value is in the doing and the trust, not in an output an algorithm can generate. Use AI actively throughout as a thinking partner, not to do the work, but to extend what you are learning to do yourself.

We cannot see five years out reliably anymore. The path that assumed a predictable trajectory from degree to corporate career is less available than it was. A brain trained in deep frameworks and a portfolio of practical, human-centered work is the form of resilience that makes sense in an environment where the path itself keeps shifting.

What parents, mentors, and leaders can do

This is also a leadership question.

For those of us raising children, or mentoring and managing young people at work: our role is not to protect them from difficulty. It is to sponsor them through it. The rigorous work young people might otherwise avoid, because AI makes it easier to skip, because the return is not immediately visible, is exactly the work that matters most in these years.

Bring them along. Assign the project that stretches rather than the one they can complete quickly. Ask the follow-up question that requires them to think rather than recall. Model what it looks like to work through a problem out loud rather than deferring to the first answer any tool produces.

If the work is what builds the brain, the habit that protects it is simple, and it is the one I most want to pass on. I am still practicing it myself. Before you take a problem to AI, form your own view of it first. Sit with it long enough to sketch a rough structure in your own head: what the problem really is, where you think it leads, what you would do if the tool did not exist. Then bring AI in to test and extend that thinking, rather than handing it the blank page. This is not about using AI less. It is about arriving with something of your own before you do. Whether you are mapping a project or planning a season of content, the instinct to outline it yourself first, before you open the tool, is most of the game. The difference between a thinking partner and a crutch is whether there was a thought there first.

I am asking this of my own team as much as of my son, and of myself most of all. None of us learned to think alongside a tool this capable. We level up together, and the ones doing the mentoring have to live it, not just prescribe it.

The window

The window I keep coming back to runs roughly from the late teens through the mid-20s. It is when the brain lays down the architecture it will think with for the rest of life, and it lays it through effortful work, the kind that does not resolve easily and cannot be outsourced.

That is the real stake in the age of AI. Not whether young people use the tools, they will, but whether the work that builds them still gets done, or quietly gets skipped because a capable tool made skipping it frictionless.

The brain is more plastic than we ever assumed. The window does not reopen in the same way. That is not a reason for anxiety, for them or for us. It is a reason for intention.