AI Lets Students Start Earlier. The Bottleneck Moves With It.

On August 10, Mark Zuckerberg published The Future Is for Everyone, an essay about how superintelligence should develop and who should benefit from it.

Much of the discussion around it has focused on open access, safety, and whether such powerful intelligence should be concentrated in a few institutions. I kept returning to a smaller, more immediate question: if powerful tools become available to almost everyone, what will a young person actually do with them?

Zuckerberg believes superintelligence may make its greatest contribution through invention. He illustrates the idea with his eight-year-old daughter:

“My 8 year old daughter can already code her ideas and produce videos in an evening…”

He imagines every student having a patient, personalized tutor with PhD-level knowledge across subjects. In his words, “everyone will soon have invention superpowers.”

You do not have to accept every part of that forecast to notice that the order of work is already changing.

Alongside our work in education, we run Predexon, a technology company building infrastructure for prediction markets. Not long ago, entering an unfamiliar codebase meant spending hours reading documentation and tracing how the pieces fit together before we could test an idea. Now a coding agent can map the system, suggest several approaches, and help produce a first version. We still have to check its work, but the initial exploration is much faster.

That sounds like a simple gain in speed. It is not. Faster execution also lets a weak assumption travel further before anyone notices. An analysis can look convincing while resting on the wrong data. A product can work exactly as designed and still solve a problem nobody has.

Students no longer have to wait as long to begin

Education has traditionally followed a sensible sequence: learn the foundations first, then apply them. Study mathematics before using it in a model. Learn to program before building software. Spend years in a field before attempting to contribute to it.

The foundations still matter. Genuine expertise cannot be prompted into existence. But AI makes it possible to move back and forth between learning and doing much earlier.

A student curious about genetics can begin with a real paper and ask for help where the language or statistics become difficult. Someone with little programming experience can test a small product idea, then learn each technical concept as the project calls for it. They are not skipping the knowledge. They finally have a reason to need it.

That reason changes the texture of learning. A child who often asks, “Will this be on the test?” may spend an evening working through material far beyond the syllabus because something they care about depends on it.

Parents are understandably trying to work out what this means:

No one can reliably describe the labor market a thirteen-year-old will enter after university. A more useful goal is to prepare that student to keep moving when today's forecasts turn out to be wrong.

So where did the bottleneck go?

A student can already ask AI to research a topic, write code, analyze data, or turn rough notes into a polished presentation. Learning to use those tools well is important. It is also the easy part to demonstrate.

The harder work sits behind the output:

This is why knowledge becomes more valuable, not less. When AI produces an answer that looks finished, subject knowledge is what helps a student notice that a source is weak, a comparison is unfair, or an important constraint has disappeared from the analysis.

Tool fluency gets a student into the room. Knowledge and judgment determine what they can do once they are there.

The order of learning has to change

Judgment is difficult to teach as a standalone subject. Students develop it when the answer key runs out: when they must narrow an idea, decide which source to trust, put unfinished work in front of another person, and respond when reality disagrees with them.

This does not mean replacing foundations with projects or letting AI explain everything. It means shortening the long delay between learning something and discovering why it matters. A student can study, try, fail, return to the material with a better question, and try again.

AI makes that cycle faster and lets it begin earlier. The point is not to produce impressive-looking work sooner. It is to give a student more chances to discover that an answer was wrong, understand why, and make the next decision themselves.

I do not want a student to spend three weeks fighting a technical barrier that AI can help clear in an afternoon merely to prove that they worked hard. I would rather see them use the best tools available, attempt a harder question, and still be the person responsible when the work fails.

Invention superpowers will matter only if students learn not to hand away the responsibility that comes with them.