AI makes deep specialists scarce and precarious at once

You keep reading that AI is minting scarce new specialties, and the design world has the cleanest version of the story. The generalist designer is splitting into separate jobs, and the deepest of them, designing for AI agents and designing the AI itself, are the ones nobody can hire fast enough. If you’re a designer or a PM watching this land, the move looks obvious. Go deep, pick the path with the biggest shortage, and let the market pay you to walk it.

That framework comes from a sharp piece by the Nielsen Norman Group, the design research firm, called “The Four Design Jobs AI Created,” by Sarah Gibbons. It argues AI has fractured the generalist designer into four distinct jobs, each with its own hiring pipeline, career ladder, and pay band. The two deepest jobs carry the widest gap between demand and the people who can actually do them. The hiring market backs the scarcity. Specialists fluent with large language models, the models behind tools like ChatGPT, command pay premiums of 40 to 60% over peers who aren’t, according to hiring-market data compiled by Contrary Research. The skill is genuinely rare, and genuinely well paid.

Now hold that next to a second force running through the same AI economics. John Cutler, a product strategist who writes the widely-read TBM newsletter, laid it out in a late-July issue. When AI pushes the cost of producing work toward zero, the expensive thing left standing is coordination: the overhead of briefing one more specialist and folding their decisions into everyone else’s. On most tasks, Cutler argues, that coordination tax now runs higher than the value a specialist’s depth adds. So orgs keep fewer specialists on staff, not more.

Put the two forces together and they collide.

AI is manufacturing exactly the deep specialties its own economics make uneconomic to employ. The skill gets scarcer and better paid. The full-time job it should have become never gets created.

Watch which specialists Cutler’s model actually keeps in-house. Just one tier: the strategy and architecture people whose work is a durable competitive advantage for the business. Everyone else, the “narrow specialization,” gets rented. Contractors and staff augmentation, which means renting specialists by the project instead of hiring them, cover the rest.

Now run the four design paths through that filter. “Designing for AI agents” and “designing the AI” are not business strategy or system architecture. From a coordination-cost view they are precisely the narrow specialization, however deep it goes. They fall in the rented bucket. So the supply-gap paths the NNGroup framework tells orgs to hire into are the same paths Cutler’s logic tells orgs not to employ. The shortage is real. The staff job on the far side of it may not be.

The natural objection is the one the scarcity story rests on, and it’s fair. A 40 to 60% pay premium is not nothing. If demand outruns supply this badly, surely that buys you a good career. But a premium day-rate on a three-month contract isn’t a ladder, and it isn’t a seat that survives the next reorg. Scarcity can lift what the skill earns per hour while doing nothing for whether the skill comes with an employer.

Here’s where it gets sharper. The usual story about AI and jobs runs on age. The labor market is going K-shaped: senior workers pull up while junior workers drop out. Stanford’s economic-policy institute, SIEPR, found workers aged 22 to 25 in AI-exposed jobs saw employment fall 16%, while senior workers in the same roles gained 6 to 9%. Job postings that dress entry-level roles in senior requirements grew 35% since 2019, while traditional junior postings fell 10%. Those numbers are verified, and the shape is real.

But coordination cost doesn’t sort people by age. It sorts them by whether their work is a durable advantage to the business. The people who stay on payroll are the strategy-and-architecture roles. The people pushed out to contract are the narrow specialists, however senior and however deep. Age barely enters it. That’s a different axis running across the same market.

Which means the real bottom of the K isn’t only the 22-year-old who can’t land a first job. It’s also the mid-career specialist who spent three years going deep on designing for AI agents, sitting inside a documented talent shortage, and still struggling to get hired for it full-time, because that depth is exactly the coordination tax Cutler’s bundling logic exists to shed. Scarce and precarious, in the same person, at the same time. I want to be honest that this last step is an inference, not a measured fact. The pieces under it are solid. The intersection is a prediction, and nobody has run the numbers on it yet.

Each of the two stories misses this on its own. The scarcity read sees a shortage and calls it leverage. The coordination-cost read looks at the same specialists and sees a cost to shed. Where they cross sits the case both walk past: a talent shortage that is really a refusal to turn demand into headcount, wearing a shortage’s clothes.

So the number worth watching isn’t the size of the supply gap. Everyone’s already quoting that. Watch the share of it that ever becomes a full-time job. Track whether the deepest AI design roles get filled as staff positions with a ladder, or as contracts and staff augmentation that never convert. If it’s the latter, and if the shortage holds precisely because orgs would rather rent the skill than hire it, then “specialize deeply, the market needs you” was only ever half true. The market needs the skill. Whether it will employ the specialist is a separate question, and the harder one.

If you’re weighing a bet on going deep, ask that sharper question before the pay premium seduces you. The premium tells you the skill is scarce. It tells you nothing about whether anyone will keep you on the payroll to use it.