The Generalist's Revenge

For most of my career, I've introduced myself with a quiet apology. "I'm a generalist," I'd say, with the faintly sheepish tone of someone admitting they never quite finished anything. A bit of coding. A bit of design. A bit of product management, data analysis, business strategy. Enough to be dangerous in six domains and expert in none.

The word I actually used — and still use, in the tagline of this very site — is mediocre. The mediocre generalist. It started as self-deprecation, the way British people use humour as a prophylactic against criticism. If I called myself mediocre first, nobody else could hurt me with it. Verbal judo, essentially. Fall down before anyone pushes you.

But something has changed. And the world hasn't fully caught up.

In a world where AI provides instant, on-demand depth in nearly any domain, the person who can see across domains — who connects dots, asks the right questions, navigates between disciplines, and synthesises insights that no single specialist would reach — isn't mediocre at all.

They might be exactly what the moment needs.

The old world: depth was everything

The old paradigm earned its place honestly. I'll give it that.

For the better part of a century, the knowledge economy rewarded specialisation above everything. And it made sense. The world was complicated, information was scarce and expensive, and the only way to achieve real mastery was to go deep. Ten thousand hours. The T-shaped professional — broad awareness at the surface, one single deep spike of expertise that defined your value, and God help you if that spike pointed at something that became obsolete.

Medicine specialised. Law specialised. Software development fractured into front-end, back-end, DevOps, security, data engineering, machine learning, and seventeen other subdisciplines, each with its own conferences and certifications and tribal knowledge and inexplicable loyalty to particular text editors. Pick a lane, go deep, and the market will reward you.

And it did. Handsomely. The specialist commanded the premium. Isaiah Berlin's famous distinction: the fox knows many things, but the hedgehog knows one big thing. For decades, the hedgehog was king. The fox was a freelancer between contracts.

But here's the thing I've come to believe, and it's not a comfortable thought: specialisation was always a compromise, not an ideal. We specialised because human bandwidth is limited. There are only so many hours, so many years, so much a single mind can hold. AI loosens that constraint. The specialist era wasn't the natural state of human capability — it was a workaround. A necessary adaptation to scarcity. And we may be returning to something older.

The generalist doesn't know less than the specialist. They know less about more, which turns out to be its own kind of advantage.

Job descriptions said it all. "5+ years of experience in React." "Deep expertise in Kubernetes." "PhD in computational linguistics." Nobody was advertising for "person who reads widely, makes unexpected connections, and can hold a conversation with anyone from a neuroscientist to a plumber." That role didn't have a title. It barely had a salary.

The crack in the wall

The first hint that something was shifting came — for me, at least — around 2023, when I started using GPT-4 seriously for work.

I was trying to build a data pipeline. I am not, by any reasonable definition, a data engineer. I know what a data pipeline is. I know roughly what it should do. I can sketch the architecture on a whiteboard with enough confidence to avoid immediate embarrassment. But ask me to write production-quality Apache Airflow DAGs from scratch and I'll stare at you with the hollow eyes of a man who's about to spend four days on Stack Overflow, increasingly convinced that the entire field of data engineering is an elaborate prank being played on him specifically.

Except I didn't spend four days on Stack Overflow. I spent forty minutes in a conversation with GPT-4, describing what I needed in plain English — with enough technical context to be useful, enough domain knowledge to specify the constraints, and enough architectural awareness to evaluate whether the output was sensible or hallucinated nonsense dressed in plausible syntax.

And it worked. Not perfectly. I had to iterate, correct, adjust. But the thing got built, and it was good, and a specialist data engineer who reviewed it later said it was "solid, maybe a seven out of ten." A seven out of ten. For someone who is emphatically not a data engineer, that's not a seven out of ten. It's a different game. The gap between what I could do and what I could orchestrate had narrowed dramatically.

The AI didn't need me to be an expert. It needed me to know enough to ask the right questions, evaluate the answers, and see the bigger picture. It needed me to be exactly what I already was. A generalist. Except now, being a generalist actually worked.

The new world: breadth beats depth

Here's the inversion, stated plainly:

In the old world, the bottleneck was access to expertise. Expensive, slow to acquire, concentrated in the heads of specialists. The generalist couldn't compete because they couldn't go deep enough, fast enough.

In the new world, expertise is abundant — encoded in models trained on vast amounts of human knowledge. What's scarce is the ability to orchestrate that expertise. To see connections between domains. To frame problems correctly. To know which expert to summon, which question to ask, which output to trust and which to bin. Depth is on tap. Breadth is in your bones.

The generalist isn't competing with the specialist anymore. They're conducting the orchestra. And it turns out the conductor doesn't need to play every instrument. They need to hear them all.

What does a good generalist actually do? They translate between domains — explain a technical constraint to a business stakeholder and a business requirement to an engineer, because they speak both languages imperfectly but fluently enough for the meaning to survive the crossing. They spot patterns across fields — notice when a solution from logistics applies to healthcare, or when a design principle from architecture illuminates a software problem. They ask naive questions that turn out to be important, because they haven't been trained to accept the assumptions that specialists take for granted.

All of these skills become more useful when you have AI providing instant depth on demand. You don't need to spend four years learning Kubernetes if you can describe what you need and have a model generate, explain, and iterate on the configuration. But you do need to know that Kubernetes exists, roughly what it does, when it's the right tool and when it's overkill, and how it fits into the broader system you're building. The map matters more than any single territory.

Naval, specific knowledge, and the polymath's edge

Naval Ravikant has this concept of "specific knowledge" — knowledge that can't be trained for, that comes from your unique combination of experiences, interests, obsessions, and personality. It's the thing you know that nobody else knows in quite the same way, because nobody else has lived your particular life.

For specialists, specific knowledge tends to be deep and narrow — years of accumulated expertise in one domain, a feel for the material that transcends any textbook. The surgeon's hands. The pilot's instinct.

For generalists, specific knowledge is different. It's combinatorial. The unique intersection of everything you know a bit about, filtered through your particular perspective. And combinatorial knowledge has an interesting property: the number of possible combinations grows exponentially with the number of domains you've touched. Each new domain doesn't add to your value linearly. It multiplies it.

A person who knows a bit about software, a bit about education, a bit about child psychology, and a bit about game design might see opportunities for AI-powered learning tools that no specialist in any one of those fields would ever conceive. Not because they're smarter. Because they're wider. They can see the whole board while everyone else is staring at their own square.

AI amplifies combinatorial thinking. When you can rapidly prototype across domains — ask a model to draft the curriculum, sketch the UI, write the backend, analyse the psychology literature — the generalist's vision becomes executable in a way it wasn't before. You used to need a team of specialists to build on a generalist's insight. Now you need the generalist, a laptop, and a decent internet connection. The vision and the execution can live in the same person again.

Leonardo would have loved this

I think about Leonardo da Vinci sometimes. The original polymath. Painter, sculptor, architect, musician, mathematician, engineer, inventor, anatomist, geologist, botanist. The man designed flying machines and painted the Mona Lisa and wrote backwards in mirror script because apparently being a genius in twelve fields wasn't enough of a personality. He had to be eccentric about it too.

In Leonardo's time, the boundaries between disciplines were porous. A well-educated person was expected to know something about everything. The Renaissance ideal was the uomo universale — the universal man. Specialisation, in that context, would have seemed like a failure of imagination.

We lost that. The industrial revolution, the professionalisation of knowledge, the sheer explosion of information — it all pushed us toward narrower and narrower lanes. By the twentieth century, a physicist and a biologist could work in the same university and have virtually nothing to say to each other. The two cultures, as C.P. Snow called it. A real loss, even though it was a neccessary adaptation to the complexity of what we'd built.

AI reverses the pressure. Not completely — there will always be domains where deep specialisation is essential, where you want the person operating on your spine to have done nothing else for the past fifteen years. But the direction has shifted. The tools now reward the person who can think across boundaries, who sees connections, who moves fluidly between art and engineering, between the human and the technical.

Leonardo da Vinci, dropped into 2025 with a laptop and Claude, would have been absolutely insufferable. He would have shipped things at a pace that embarrassed the rest of us. Not because he was the best at any one thing, but because he was unusually good at seeing how everything connected.

The generalist's toolkit

Right. Practical bit. Philosophy is lovely, but you probably want to know what to actually do.

Talk to people who aren't like you. The most valuable insight I ever received about software design came from a conversation with a chef about mise en place — the principle of preparing and organising all your ingredients before you start cooking. The parallel to code architecture is surprisingly deep, and I've used it in every technical talk I've given since. A chef taught me more about software than most software books.

Use AI as a depth-on-demand tool. When you encounter an unfamiliar domain, don't spend three months becoming an expert. Spend three hours in conversation with a good model, asking it to explain the fundamentals, the key debates, the common pitfalls. You won't become a specialist. You'll become a well-informed generalist in that domain, which is — increasingly — the more valuable thing to be.

And get comfortable with "good enough." This is the hard one, psychologically. The specialist's standard is excellence in their domain. The generalist's standard is adequacy across many domains, orchestrated into something that exceeds what any single domain could produce alone. You will never be the best React developer in the room. You will never write the most optimised SQL query. That's fine. Your job isn't to play every instrument perfectly. Your job is to hear the whole orchestra and know when the oboe is flat.

The mediocre generalist's confession

I should tell you what it actually felt like. Before the inversion. Before any of this.

It felt like being permanently inadequate. Everyone around you had a superpower and you had... curiosity. Which is lovely, very endearing, tells well at dinner parties, but it doesn't pay the mortgage. I'd sit in meetings where the database expert would dive into query optimisation and I'd follow along at about 80%, then the security specialist would talk about zero-trust architecture and I'd track at maybe 70%, then the UX designer would present wireframes and I'd have opinions — good ones, I think — but I couldn't make them. I could see what was wrong but I couldn't fix it myself.

The imposter syndrome was industrial-grade. Not the cute kind people write LinkedIn posts about. The real kind. The kind where you lie awake at quarter past two on a Tuesday, convinced that tomorrow is the day everyone figures out you've been skating by on charisma and breadth while the real professionals — the ones with actual depth — do the actual work.

Here's the generalist's secret shame, though. It isn't that they don't know enough. It's that they know just enough about everything to see how much they don't know. The specialist has the bliss of bounded ignorance. The generalist lives in a permanent state of knowing what they're missing. Turns out that's called perspective. It just didn't feel like it at the time.

And then the world shifted. Not slowly. Not gradually. Like a switch being flipped. One day, breadth was a weakness. The next — I'm exaggerating, but not by as much as I'd like — it was the whole game.

The first time I used AI to build something that actually worked, across multiple domains, in a fraction of the time it would have taken a team of specialists, I felt something I hadn't felt in years. Not pride exactly. Relief. Like I'd spent twenty years training for a sport that hadn't been invented yet, and it just got invented, and I was already warmed up.

The revenge

I called this article "The Generalist's Revenge," which is a bit dramatic. I'm a generalist, not a monk — I'm allowed a touch of drama. Especially now that drama is finally paying.

But the revenge isn't personal. It's structural. The economy is reorganising around a new set of capabilities, and those capabilities — connecting, synthesising, translating, orchestrating — happen to be the ones that generalists have been quietly developing their entire careers. Not because it was strategic. Not because they saw this coming. Because it was who they were. Because they couldn't help themselves. Every time someone said "specialise," they'd nod thoughtfully and then go read a book about something completely different.

We weren't preparing for this moment. We were just being ourselves. And it turned out that being ourselves — curious, restless, perpetually half-informed about too many things — was solid preparation for a world where depth is cheap and breadth is valuable.

So. The generalists. The ones who never quite fit the job description. The ones who knew a bit of everything and felt faintly guilty about it, the way you feel guilty about reading a novel on a weekday afternoon. The ones who were told to specialise, to pick a lane, to go deep — and couldn't, because the world was too interesting and the connections too compelling and the lanes too narrow for a mind that wanted to wander.

Your time has come. Not because you changed. Because the world did.

The fox knows many things, but the hedgehog knows one big thing.

— Archilochus

In the age of AI, the fox has a new trick: it can borrow the hedgehog's depth whenever it needs to, and still keep its peripheral vision. The hedgehog, meanwhile, is discovering that knowing one big thing isn't quite enough when the ground keeps shifting.

The mediocre generalist sends their regards.