AI didn't make the specialist obsolete. It changed what they're for — and how many of them you need.
Last month a designer on my team shipped a frontend change to production. Not a mockup, not a Figma comment — the actual feature, the code, the PR. A week later one of our PMs diagnosed, fixed and shipped a bug fix that was reported by a customer minutes before.
Two years ago, either of those sentences would have been impossible, partially due to clear skills gaps, but also due to organizational norms and established boundaries. A designer in the codebase? A PM touching the backend? You'd have pulled them aside. Now it's just Tuesday.
There's a familiar shape that explains what's happening here — and I think it's quietly bending out of its old form.
The T everyone knows
For years we've described good people as T-shaped: deep in one thing — the vertical stem — and broad enough to work across everything else — the thin horizontal crossbar. The stem is your craft, the thing you're great at. The crossbar is range: enough to sit in a design review or read a schema and hold a conversation, but not enough to actually do the work.
That crossbar was always shallow, and it was shallow for a structural reason. Pre-AI, you could only execute in the areas where you had depth. A backend engineer understood design well enough to argue with a designer; they could not produce the design. Breadth was conversational. It let you collaborate. It didn't let you build.
The classic T — and what AI turns it into.
Relabel the bars
Here's the move. Keep the T, but change what the two bars mean.
The stem is no longer depth of knowledge. It's judgment: taste, the ability to look at a piece of work and know whether it's any good. That's the thing AI doesn't hand you.
The crossbar is no longer breadth of knowledge. It's execution: what you can actually ship across the stack. And that's the bar AI is widening — fast.
The old crossbar was short only because execution was chained to depth. AI breaks that chain. A PM can now produce a working frontend; a designer can produce the API glue; an engineer can produce a credible first design. Execution has come unbound from where you happen to have judgment. The T is widening, and it's widening along the execution axis — the crossbar now reaches well past the stem.
The bar keeps moving
The unsettling part is how fast the crossbar grows. A year ago, AI could write code but couldn't really design — ask it for an interface and you got something competent and soulless. That's already untrue; it now produces design work that would have needed a designer not long ago. Nobody decided this. The crossbar just got longer. And the only safe bet is that it keeps going: whatever sits just past the end of the bar today — the discipline your tools are still bad at — is probably next.
So teams get more and more leveraged. Each person reaches further. Breadth stops being decorative and becomes load-bearing.
Where it gets interesting: execution outran judgment
This is where most takes stop, and where the real problem starts.
AI lengthened the crossbar. It did nothing to the stem. It raised the floor of production without touching the ceiling of judgment. Anyone can now generate a frontend, a schema, a design, a landing page. Far fewer people can look at what got generated and know whether it's any good — whether it's secure, whether it'll scale, whether it quietly encodes a decision you'll regret in eighteen months, whether it's the right answer or merely an answer.
That second skill is the stem. It comes from depth, and AI doesn't grant it.
The failure mode of the AI era: a long crossbar on a short stem.
So the specialist's job changes. It moves from producing the work to catching what's wrong with it — from being the person who makes the design to the person who can tell, in five seconds, that the AI-generated design is subtly broken and why. The good PM is the individual that truly knows how to articulate a customer's problems, knows the industry and knows how to ship incrementally. The value migrates from the crossbar to the stem.
And it hands us a clean picture of the new failure mode: someone with a long crossbar and a stubby stem. Ships everywhere, taste nowhere. Pre-AI that person couldn't exist — you couldn't produce across the stack without depth. Now they can, and they're dangerous, because a high volume of output looks like productivity right up until someone with a real stem reads it.
The uncomfortable part: fewer people
If the crossbar is now long and the scarce thing is the stem, then headcount comes unstuck from lane-coverage.
We used to staff to cover lanes. Five disciplines meant at least five people — usually more, because one person per discipline is a single point of failure, so you doubled up. The org chart was a map of lanes, and every lane needed a body. Larger orgs had "golden ratios" of PMs & Designers to SWEs.
Coverage used to mean headcount. Now it means a few deep stems plus AI-extended reach.
That math changes. You no longer need a body in every lane. You need enough depth — enough tall stems — in the right places to hold the quality bar, and you let AI-extended execution cover the span between them. The redundancy we used to buy with extra headcount, we now buy with the width of everyone's crossbar.
In practice that means smaller teams. Not "the same team, more productive," though that's true too, but structurally fewer people. And the people you keep are not necessarily the most prolific builders. They're the ones with the tallest stems — the taste to know when the work is wrong. When producing the artifact is cheap, the rare and valuable thing is the judgment to evaluate it. That's who you staff around.
I know how this reads. I'm not arguing for a skeleton crew, or that people are interchangeable. I'm arguing that the unit changed. We staffed to cover lanes. We should be staffing to cover judgment.
Where this breaks
None of it is clean, and the open questions are real.
If everyone touches everything, who owns coherence? A dozen people shipping across the whole stack with AI is a dozen people generating locally-reasonable, globally-incompatible decisions. Architecture, the data model, the design system — these need owners more than ever, not less. The Widening T makes individuals more capable and the system harder to hold together.
There's a human cost, too. If you turn your best specialists into full-time reviewers of everyone else's output, you'll burn them out and bore them. The job has to stay partly about building, or the people with the tall stems leave. Nobody got deep enough to have judgment by spending their days grading.
And the one I find genuinely unsettling: the crossbar isn't the only thing growing. AI's judgment is improving too. The stem — knowing what's good — is itself a frontier the models are moving toward. I don't have a clean answer for what's left when that ceiling rises. I just don't think it's arrived yet, and I think the teams that win the next few years are the ones that staff around judgment while it still belongs to people.
What I'm hiring for
For most of my career, the question that mattered about an engineer, a designer, a PM was what can you build? — a question about the crossbar.
The Widening T is retiring that question. The one replacing it is about the stem, and it's harder and a lot more interesting: what can you tell is good?
That's the skill that compounds now. It's what I'm hiring for, betting on, and trying to get better at myself.