UX job postings fell 71 percent since 2022, and polish is the skill getting priced to zero

The UX job you hired for last year is not the one your team does now. Postings are down, titles are shifting, and the loudest voices can't agree on whether AI is the villain or the tool. Let me catch you up on what actually changed and what to do about it before your next review.

The deep cut

• Discovery earns the right to speed. Starting in an LLM gave one team a convincing admin persona that would have shipped faster mistakes.
• Output grew, confidence did not. Atlassian found PMs gaining two hours a day while 84% still fear their products fail.
• The leader without hands loses the room. Patrick Neeman's player-coach reads the P&L and ships work, not just meeting notes.

The field has done this to itself before

The downturn is real. Jeff Sauro and Jim Lewis found UX job postings down 71 percent from their 2022 peak, with one in three organizations cutting UX staff in a single year. That is the worst reading on record, worse than after 2008.

But AI is not the whole story. The field blew up and contracted twice before, once in the dot-com bust and again when bootcamps turned "UX designer" into a twelve-week certificate. Each boom pulled in people sold a fast track into the job, and each correction took those people first.

What makes this wave bite harder: the industry taught a generation that design is how something looks. Figma's own survey found 58 percent of designers defined craft as visual polish. AI is very good at making things look finished, so that is exactly the skill getting priced to zero.

The research that checks a box and learns nothing

The quickest way to look fast and go nowhere is to start in the model. Ask an LLM for user needs and personas, and you get polished results in minutes that feel like truth. They are averages, built from what has already been written about your problem, not what your users risk today.

One team asked to speed up an admin console prompted an AI for a persona and nearly shipped bulk actions and fewer confirmations. A week with five real admins flipped it: they did not fear clicks, they feared mistakes that lock out hundreds of people. They needed previews, audit logs, and undo. The persona was not wrong that admins are busy. It was average.

Synthetic participants fail the same way. When NN/g re-ran three of its own studies with AI-generated users, the answers were too shallow to help. In a User Interviews survey, 81 percent of researchers used AI regularly, but only 8 percent used it to simulate participants. Your users' workarounds from last month live in the field, not in training data.

The leader who stopped making things

The title stack that counted headcount is gone. Peter Merholz traced design inflation to fifteen years of near-zero rates, with Capital One growing from about 90 designers to 500 or 600 in a few years. Team size was the business case. Now Oxford Economics pins only about 55,000 of 2025's layoffs on AI, under 5 percent. A lot of what gets blamed on AI is the correction for over-hiring.

The leader who survives reads a profit and loss statement and can say what a resourcing change is worth before anyone asks. Patrick Neeman calls it the player-coach, borrowing from Bill Russell coaching the Celtics while playing center. The point is hands in the game, setting the quality bar by meeting it in public.

Craft has moved past polish to the problem solved: the fewest steps, the edge case nobody notices. For AI features, that means writing evaluations, the examples and rules that define good output. That work looks less like a mockup and more like an editor's markup.

The roles are bleeding into each other

The handoff chart, PM writes requirements, designer draws screens, engineer builds, is folding back together. Figma found designers doing development jumped from 21 to 41 percent in a year, and developers doing design rose from 44 to 60 percent. LinkedIn replaced its associate PM program with an Associate Product Builder track that teaches coding, design, and product together.

At small companies this is already the job. One 20-year IC put it plainly on Reddit: expect to be mini-PM, designer, and mini-engineer, self-sufficient enough that partners aren't the bottleneck. Vibe-coding a prototype is not enough.

Specialists do not vanish. Someone still has to know accessibility law or when an AI action needs a human confirmation step. What changes is who carries a feature from idea to launch, and your job descriptions should say so before someone outside design writes them for you.

Three questions for your team

• Before we build the next feature, can we point to an interview or support ticket behind the problem, or did the persona come out of a prompt?
• Which roles on our team own a feature end to end, and which only do one slice that AI now handles cheaply?
• When did I, as the leader, last ship work in the same tools and channel as the team, and does my promotion criteria reward problem-solving or polish?