As foundation models continue to improve, the value of specialization erodes.
In the past, deep specializers were valuable because they were scarce. People worked their entire lives to corner the market on a particular domain or skill, and then reaped rewards because they were difficult to replace. Becoming an expert took time and effort (patent litigators, concert pianists, simultaneous interpreters) and often led to outsized financial gain (quantitative traders, neurosurgeons, actuaries).
But when foundation models possess expert-level knowledge about everything, depth isn’t scarce anymore. Of course, AI is transforming some fields more rapidly than others. Experts in jobs that involve physical environments or human relationships are less immediately affected by the increasing ubiquity of AI models, but LLM-enabled products will no doubt revolutionize those industries as well. If expert-level knowledge is no longer a moat, how do you distinguish yourself in a world of AI? What is the next manifestation of expertise?
AI progress enables humans to operate at higher levels of abstraction. Work has moved away from implementing low-level processes (writing lines of code, creating financial models, writing legal briefs) and towards managing high-level ones (goal definition, system design, workflow orchestration), even for formerly specialized roles. The skill set of the high-level coordinator is fundamentally different from that of the deep expert: the coordinator excels by knowing enough about many things to identify and create connections between them.
Depth is no longer where the advantage lives. What’s valuable now isn’t going deep in one domain, it’s operating across many—in other words, going wide. Successful people will be those who can learn quickly, synthesize information across domains, and direct AI-powered systems to handle the domain-specific work and accomplish human-mediated goals.
This is not to say that going deep is no longer useful. Building expertise teaches you how to think rigorously and thoroughly understand systems. It forces you to be disciplined and work through hard problems, something that an AI-aided generalist may never have to do. And perhaps most importantly, it allows you to understand what good looks like: depth gives you the pattern recognition skills to identify quality in other areas as well.
But in a world where work is mediated by AI systems that are better at coding than the best software engineers, more knowledgeable than the best researchers, and better at diagnosing than the best doctors, humans need a different way to corner the market. For now, that way is to sit a layer above: operating horizontally and delegating to systems that provide depth more efficiently than any human could.
The coordination layer isn’t immune to automation, but humans have more of a natural advantage there. AI systems are extraordinarily good at executing in service of a defined objective, but defining the objective is still a very human skill. I’ll spare you from another “taste is the moat” sermon, but I think it’s safe to assume that judgment, intuition, and values remain human-centered longer than knowledge does. Deciding what people actually want, and then translating that intent into goals worth pursuing, is still something humans do better than models.
And in a future where AI handles going deep, the humans who are best equipped to go wide—the generalists—will win.