"A human being should be able to change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure, program a computer, cook a tasty meal, fight efficiently, die gallantly.
Specialization is for insects." — Robert A. Heinlein
I learned the basics of programming in high school by breaking my father's work computer: a Wang 286 he brought home in 1987. It was the first computer in our small town. It never occurred to me I could make a living out of it.
At fourteen, I started playing piano. Got good enough to play in a folklore projection group and won a trip to Europe. I ended up meeting the Pope. Then never played again.
At seventeen I left my hometown for Buenos Aires to study economics. I shared a flat with three guys I barely knew. I landed a job doing customer support and implementing ERP software. Got curious about how those ERPs were built, so switched careers to study information systems.
By twenty-five I had tried multiple roles and industries. I led tech projects across LatAm, and out of necessity I'd learned enough of each role around me (engineers, sysadmins, business analysts, project managers) to keep things running when someone took vacation, quit, or got hit by the proverbial bus.
I'd also learned to change the diapers of my newborn daughter.
Two kinds of careers
I love going down rabbit holes. Always have. Deeply learning something new and then using it to solve bigger problems. But in professional settings it came with a cost: the corporations I worked at needed to box me into a title. A role. A set of expectations. Managers weren't supposed to code. Tech leads weren't supposed to help sales. You couldn't act as if the org chart didn't exist. Yet, ignoring it was the only way I got to see the bigger picture.
So I left the corporate world (mostly) and started working at startups, where being a generalist was an advantage rather than a character flaw. I could jump from tech strategy to product definition, from hiring to architecture, from code review to customer calls, whatever the moment demanded.
But this kind of career is rarely up and to the right: mine was sometimes sideways, sometimes straight down. I took lower-paying jobs if they bought me exposure to things I didn't understand yet. Every jump expanded the surface of things I didn't know: FinTech, InsurTech, AgTech, BioTech.
Had I followed a specialist's career, things would have been quite different. Specialization pushes you deep into one area. And, in a stable world, that expertise pays: reliable returns, a clear ladder, a moat that gets deeper every year. But it's a concave bet. When your entire edge is knowing one domain better than anyone else, any disruption to that domain is an existential threat. Imagine dedicating your life to driving a horse-drawn carriage, or operating an elevator.
A generalist career is the opposite bet. It's convex. You're wrong a lot. Some jumps lead nowhere, you take pay cuts, you abandon skills and start over more times than you can count.
A few years back I walked away from a well-paying US contract to chase a domestic bet in Argentina. A political shift killed it before I capitalized it, and I ended up back where I started.
The bet didn't pay financially, but the years still expanded what I knew. That payoff can compound asymmetrically.
But this isn't an argument for knowing a little about everything. That was always a dead-end. What I'm describing is closer to what Kent Beck calls a "paint drip": you move the brush across the canvas following your curiosity, and depth forms wherever you stay long enough. The goal isn't breadth for its own sake. It's being genuinely good at several things, so the combinations become your edge.
So which strategy will this AI-shaped, power-law world reward more: specialists or expert generalists?
Daniel Rabinovich, COO and former CTO of Mercado Libre, says nothing is riskier today than being a super-specialist. He manages 140,000 people, and he's also a magician, a chess player, a speedcuber, and a musician.
There's even a pre-AI version of this debate between Malcolm Gladwell, who popularized the idea that mastery requires 10,000 hours of practice in one domain, and David Epstein, author of Range. Gladwell eventually conceded that his 10,000 hours rule works in what he calls "kind" environments: settings where the rules are clear, feedback is immediate, and patterns repeat. Chess. Surgery. Tennis.
If you're building a compiler, designing an airplane, or operating on a brain, you don't want the renaissance man. You want the person who has done this one thing ten thousand times.
But most of life isn't kind. It's what Epstein calls a "wicked" environment: rules are unclear, feedback is delayed or misleading, and you can do the right thing and still get the wrong outcome. Business is wicked. Strategy is wicked. Innovation is wicked.
The moat got shallow
The specialist is optimized for a world that doesn't change much. The expert generalist is antifragile. To borrow Nassim Taleb's term: someone who gains from disorder, because every disruption is just another opportunity to learn a new domain and spark new ideas.
This Harvard study posted 166 R&D problems that specialist teams couldn't crack. About a third were solved by outsiders, and the further the solver's field was from the problem's domain, the more likely they got it. Chemists solving biology problems. Physicists solving chemistry problems.
My years implementing ERPs for Argentine SMBs taught me how businesses actually run beneath the org chart. A decade later, that same knowledge helped me diagnose problems everyone assumed were technical when they were actually organizational. And my economics background also helped me frame those decisions in a way non-technical stakeholders could buy into.
The obvious counter is that AI is eating generalist work first: writing, summarizing, basic coding, basic analysis. But narrow specialist execution is being commoditized just as fast.
Five months ago I threw out every take-home test I used to send candidates: Opus would one-shot the domain logic plus obscure Node and React in five minutes.
Narrow depth got cheaper. Broad shallowness got cheaper. What didn't get cheaper is picking the right problem, smelling when an answer is wrong, and connecting patterns across fields. Call it judgment across range.
Maybe it's time to change a diaper, plan an invasion, write a sonnet, program a computer, and cook a tasty meal.
That's what humans are for.
Specialization is for insects.
