Depth Is Cheap Now. Breadth Isn't.

For twenty years the safest advice you could give a young person was three words: go deep. Pick a specialty, learn to code, become the person who knows one thing better than almost anyone alive. Depth was the moat. It was scarce, it took a decade to build, and it paid.

Depth is now a tap. The edge is knowing which wells to connect.

Depth Used to Be the Moat

That advice is now backwards. The specialist's deep, narrow skill is exactly the thing AI reproduces first and best. The person who spent ten years mastering one domain is the most exposed, not the most protected. And the generalist, mocked for a century as a jack of all trades and master of none, is quietly becoming the master, because the one thing generalists always lacked, execution depth, is now available on tap.

Think of expertise as a well. For most of modern history, becoming valuable meant digging a single well for ten years until you hit water nobody else could reach. The depth of that one hole was your career. AI turned depth into plumbing. Now you can run a pipe to any field and expert-grade water comes out on demand, no digging required. When everyone can tap any well, the deepest single hole stops being the prize. The prize goes to whoever can connect wells that were never connected before.

The specialist earned the old moat fairly. When depth was slow and scarce, the deep engineer, the deep radiologist, the deep tax attorney captured the premium, because you literally could not get their output any other way. A generation was told to learn to code for exactly this reason, and for a while it was the best career advice going. The 10x engineer was real. The bootcamps were full. Depth was the highest-paid trait in the economy.

AI Made Depth Portable

Then the tap opened. A person who cannot write a line of code now ships working software with Cursor. A founder with no legal training gets a contract reviewed, no design training gets a brand built, no data training gets an analysis run, all by pointing a model at the problem. The specialist's core deliverable, the thing that took years to be able to produce, is now the first thing the model hands out for free. The well you dug for a decade is a faucet in everyone's kitchen.

Here is the part that should reframe how you think about your own career. Generalists and outsiders were often the better problem-solvers the whole time. They were just held back by not being able to execute. The research on this is unusually clean.

Outsiders Were Already Winning

Karim Lakhani and Lars Bo Jeppesen studied 166 tough R&D problems that had already stumped corporate labs, posted for anyone to attempt on the platform InnoCentive. Roughly a third got solved by outsiders, and the counterintuitive kicker was the pattern underneath: the further a problem sat from a solver's own field, the more likely they were to crack it. A molecular biology problem got solved by an aerospace physicist. The insider's deep knowledge of what usually works was the handicap, not the edge.

Philip Tetlock found the same shape in forecasting. Across 284 experts making 28,000 predictions, the average expert turned out to be no better than, in his phrase, a dart-throwing chimp. The ones who beat the field were not the deepest specialists. They were the foxes, people who knew many things and borrowed from all of them, as against the hedgehogs who ran every question through one big theory. In his later Good Judgment Project, teams of broad generalist forecasters beat professional intelligence analysts who had access to classified information the generalists never saw.

Brian Uzzi and colleagues put a number on the mechanism. They analyzed 17.9 million scientific papers and found the highest-impact work was not the most novel. It was work built on a solid base of conventional knowledge with one unusual, cross-field combination injected into it. The breakthrough lives at the seam between a familiar field and a foreign one, and those papers were roughly twice as likely to become top-cited hits.

You can see the same thing in single careers. Abraham Wald, a statistician with no aviation background, solved the problem of where to armor World War II bombers by noticing that the military was studying the planes that came back and ignoring the ones that never did. The aviation engineers were too far inside the problem to see it. Gunpei Yokoi built the Game Boy at Nintendo out of deliberately cheap, dated components, an approach he called lateral thinking with withered technology, and it crushed rivals with far better hardware. Frances Arnold won the Nobel Prize in Chemistry by refusing the specialists' method of rationally designing proteins and borrowing evolution instead, an idea imported wholesale from another field.

Breadth With Borrowed Depth

So if generalists were often the better solvers, why did specialists win the last era? Because breadth without execution is just a good idea you cannot ship. The outsider with the fresh analogy still needed a chemist to run the reaction, an engineer to build the machine, a lawyer to paper the deal. That dependency is exactly what AI dissolves. The generalist now supplies the analogy and taps the depth to execute it in the same afternoon. That is the whole shift, and it fits in four words: breadth with borrowed depth. Frame the problem, connect the distant wells, and pull expert output on demand to finish the job.

The honest objection is that breadth without any depth is just being wrong faster across more fields. A dabbler who cannot tell a correct answer from a confident hallucination is not a generalist, he is a hazard, and someone with real depth still has to build the models and catch the subtle mistakes they make. That is true, and the Uzzi finding is the tell. The winning combination was never pure novelty or the pure outsider. It was a conventional base, real depth in something, plus the atypical jump. The person who wins is not the dilettante. It is the one deep enough in a single field to judge whether the borrowed water is clean, and wide enough to borrow the rest. I hold one caveat loosely: it is possible that AI's quiet, subtle errors end up making deep verification more valuable rather than less, which would hand part of the moat back to the specialist. I don't think that is the main story. But I could be wrong.

Judgment Is the New Scarcity

Zoom out. The market always pays for whatever is scarce, and scarcity just moved. For twenty years the scarce input was depth, so depth won. Depth is now a tap, so the scarce input became the thing the tap cannot supply: knowing which wells to connect, and having the judgment to tell good water from bad. If you are deciding how to spend the next few years, here is the order I would put it in.

How to Build for It

Keep one real well. Be genuinely good at one thing, deep enough to judge quality, so the borrowed depth doesn't drown you.
Go wide on purpose. Collect distant fields, because analogies from far away are the raw material of every outsider edge, and you cannot borrow an analogy you have never seen.
Get fluent at directing and verifying AI in domains you do not own. That skill, not any single specialty, is the new core competence.
Hunt for problems at the boundaries between fields. That is where the outsider premium runs highest, and where the deepest specialist, stuck in one hole, cannot follow you.

For a generation we told people to dig one deep well, because digging was slow and the water was scarce. AI turned every well into a tap. The advantage moved to the people who can connect them. New era, new winner: the generalist who kept one well of their own and learned to tap the rest.