Most of us who have hired for any length of time have used the T without ever calling it that. We wanted someone deep in one craft, the vertical bar, who could also talk sensibly to the rest of the business, the horizontal bar, and when we found that person we felt we’d hired well. The shape has a proper pedigree: McKinsey used it internally in the 1980s, David Guest gave it a name in The Independent in 1991, and Tim Brown made it famous at IDEO as the way to build teams where designers, engineers and anthropologists could work on the same problem without talking past each other. For thirty years it has been the default of good hiring, and for most of my years running an agency it was mine.

The T earned that place because it solved two problems that used to be expensive. Expertise was scarce, so you paid for depth and protected it once you had it. Coordination was costly, so you prized the person who could carry an idea across a departmental boundary without it dying in translation. Both bars were answers to a world where a human had to do the work and a human had to do the explaining, and both are now under pressure from the same direction.

That’s why the T is finished as a hiring model, and why we should stop drawing a shape that no longer describes the people we need. The horizontal bar has become cheap, because any of us with a decent assistant can now read across finance, marketing, law and code well enough to hold a useful conversation with the specialist in each. The vertical bar is weakening from underneath, because the first pass of expert work (the draft contract, the first financial model, the working prototype, the initial analysis of a data set) is precisely the layer AI does best. When breadth is free and the first pass of depth has been commoditised, what remains of the T is a stub, and nobody sets out to hire a stub.

I argued in February that the specialist era was shapeshifting and the T would survive in amplified form. Since then the agents have started taking on whole tasks, a campaign or a month-end reconciliation from brief to first draft, where in February they were handling fragments of one, and that changes what the letter describes. The T still tells you what a good person knows, and that’s all it tells you now, because what makes them valuable has moved somewhere the letter doesn’t reach.

Three shapes are replacing it

Picture someone who can read a profit and loss account and a data pipeline with equal ease, and who notices in a planning meeting that the strategy everyone likes depends on a number the data can’t support. That person is pi-shaped: two deep areas instead of one, with their value sitting where the two meet. Strategy and data is the obvious pairing, though product and finance, or marketing and engineering, work just as well. AI will hand you both bodies of knowledge in seconds, but it’s far worse at knowing which one the situation is really about, and that’s the judgement the pi-shaped person is paid for.

A second shape, spike-and-orchestration, keeps one deep craft and pairs it with a skill that has nothing to do with the craft: directing AI systems to do the work around it. A copywriter who can brief, run and check a dozen agents producing campaign variants is worth more to a business than a copywriter who can only write, however beautifully, because the spike gives them the taste to know when the output is wrong and the orchestration gives them reach they never had. The same is true of the developer who now spends the day reviewing code instead of typing it, and finds that the reviewing is where all the difficulty was hiding.

Hardest of the three to describe, and I suspect the most valuable, is the judgement-shaped person, whose skill is framing the problem before anyone starts solving it, deciding what the business won’t do, spotting the risk that hasn’t announced itself yet, and telling the story that gets a decision made. None of that produces an artefact you can point at in a portfolio, which is why judgement was undervalued for as long as we paid people for output. Hand the output to the machines and judgement is what’s left standing.

Leverage has overtaken output

Take two people given the same campaign brief on a Monday morning. The first is the better writer by some distance, and by Friday has produced one campaign, polished, with every line the way they want it. The second spent Monday working out what the campaign had to achieve, briefed a handful of agents against that, threw away most of what came back, and by Wednesday had three campaigns in front of the client, each about ninety per cent as good as the first person’s one, with two days left to improve whichever the client preferred. For the whole of my working life the first person was the better hire, and the T was how we found them. Now the second person wins, and wins by a margin that makes the specialist’s extra ten per cent look like a rounding error.

Leverage beats craft, and many of us built our careers on being the best in the room at one thing, so being told the room has changed isn’t a comfortable message. The businesses I see pulling ahead are the ones whose people have made peace with it: they still care about quality, they’ve just moved their own effort to the points where quality is decided (the brief, the review, the call on whether to ship) and let the machines handle the stretch in between.

What this does to hiring

If leverage is the currency, the interview has to change. “What is their deep craft?” is still a reasonable question, but it has stopped being the opening one, and a candidate who answers it brilliantly can still be the wrong hire. The questions that now predict whether someone will be valuable are these:

  1. Can they deconstruct an ambiguous problem? Give them something messy and under-specified and watch whether they start by asking what the problem actually is, or start producing.
  2. Can they design a workflow that includes AI? Ask how they’d get a piece of work done with a small budget and an assistant, and listen for where they place the human checks.
  3. Can they evaluate output critically? Show them something plausible and wrong and see whether they catch it. This is the one skill that separates the orchestrator from the person who forwards whatever the machine said.
  4. Will they own the outcome? Leverage without ownership is how a business ends up with a hundred half-finished things and nobody responsible for any of them.

The most useful hour I know of is to give a candidate a real brief, an AI assistant, and sixty minutes, then talk about what they did rather than what they made. The result matters less than the choices: what they questioned, what they refused to delegate, what they checked twice. You’ll learn more about their shape from that hour than from any list of skills, and the same four questions belong in the annual appraisal too, because a business that hires for leverage and then measures people on personal output will have the T back within a year.

Where the next judgement comes from

Where I’m less sure is what this does to the supply of judgement itself. Taste is usually learned by doing the first pass badly for a few years and having someone better correct you, and if the first pass now belongs to the machine, it isn’t obvious where the next generation of judgement-shaped people will come from. I wrote about that graduation cliff earlier this year and I still don’t have a clean answer. What I’d do in the meantime is keep some first-pass work human on purpose, as supervised training rather than production, and accept the cost of it, because the businesses that do will be the ones with anyone worth promoting in five years’ time.

We’ll still value expertise; we’ll just pay for it at a different point in the work. The old T assumed humans did the work. The new shape assumes humans design, direct and judge.