The specialist’s case for job security used to rest on a simple argument: I know things most people don’t. The labour economist would call this human capital. The career coach would call it your competitive advantage. The idea was the same: narrow expertise, deeply held, is scarce. And scarcity pays.
That argument is not dead. But it has become dangerously easy to overestimate.
Large language models are trained on enormous volumes of specialist content: the accumulated documentation, tutorials, papers, books, doctoral theses, and practitioner writing of essentially every professional and academic discipline. They do not hold a domain the way a specialist does: with lived experience, with the judgment that comes from having made real mistakes in real conditions. But they generate competent output at a speed and cost that shifts the baseline of what employers and clients need to pay a person for. In other words: ai specialist skills that were once scarce are becoming table stakes.
A developer who writes code in one framework is more exposed than they were three years ago. Not because their code is bad. Because the framework is no longer the scarce part. An analyst who runs one class of data model is in the same position — not because the analysis is wrong, but because the model is no longer the bottleneck. A product manager whose entire value proposition is writing PRDs is more exposed. Because generating one is now a ten-minute task.
None of these roles disappear. But the moat around them — the thing that justified the salary, the title, the autonomy — gets shallower with each model generation.
There is a word for what happens when a moat gets shallow enough: commoditisation. It has happened before across every field where a scarce technical skill became abundant. The pattern is consistent. When a capability that was scarce becomes easy to access, the people who built their identity around that capability have two choices. Double down on depth, hoping to stay ahead of what the machines can do. Or build the horizontal bar: the breadth that makes depth useful rather than merely impressive.
The doubling-down strategy works for a while. Specialists who go deeper do maintain some advantage — the models approximate the median specialist far more quickly than the expert at the edge. But the edge keeps moving, and the human trying to stay at it is running a race against something that does not sleep and does not charge overtime.
The breadth strategy compounds differently. The person who combines genuine depth with genuine literacy across adjacent domains creates a profile that AI finds much harder to replicate: not because of any single thing they know, but because of the intersections they occupy.
I pity the specialist who mistakes narrow depth for security. Not because depth is worthless — it is the vertical bar of the T, and you cannot skip it. But because narrow depth, without the horizontal bar, is a moat that is getting flooded while the specialist is still inside it.
Erosion is not uniform, that is the detail people may miss when they talk about AI and specialist jobs.
AI models are weakest at the boundary between domains. They can search across domains well, but the interesting problems at those boundaries require a kind of contextual judgment that comes from having operated on both sides. A specialist who has spent their entire career inside one domain, however deep, has never had to develop that judgment. The T-shaped professional has.
The pure specialist is not primarily at risk in their domain’s core. They are at risk at its edges: the conversations that require domain expertise combined with commercial literacy, or technical depth combined with a real understanding of how engineering teams actually work. Those conversations are where the value has always been created. They are also the conversations the specialist was never trained for.
You Can’t Build a Centaur made the argument that the winning configuration in human-AI systems is not the assembly line — one agent for research, one for drafting, a human to supervise — but the centaur: a person deeply enough integrated with the tool to direct it, correct it, and take the judgment calls it cannot make. The pure specialist cannot be a centaur. They have the depth in one domain but none of the breadth that makes the direction meaningful.
The counter-argument may be depth is necessary. It is. Nothing in the T-shape model suggests otherwise — the vertical bar is not optional. But depth is now a prerequisite, not a differentiator. Necessary but not sufficient, as scientists say. What differentiates is the T-shaped combination.
The next question — which most people avoid asking directly — is: what does that combination look like in practice, and how do you build it if you don’t already have it?
That is what the next post in this series — Rebuilding Your T (9 June) — addresses. Most attempts to build the horizontal bar fail. Not because people are not trying, but because they are trying in the wrong direction.
Leave a Reply