Your Best Designers Just Stopped Being Your Fastest Ones

For years, a clean Figma file was proof you were good. Tight layout, no dead ends, a design system documented to the pixel. That was the signal. AI just made that signal cheap. Point a decent model at a clear prompt and you get a serviceable screen in seconds. So the thing you have been paying for, and hiring for, and defending in reviews, is sliding toward free. Let me catch you up on where the value went and what that means for your team on Monday.

The polish moved out, the judgment stayed

When making gets cheap, a lot of what you called craft turns out to be production wearing craft's clothes. Patrick Neeman puts it plainly: a Figma frame has never shipped revenue. It promises a product, it is not one, and teams spent a decade mistaking the two.

The scarce skill now is not making one good screen. It is knowing which of a thousand plausible screens deserves to exist. That is taste, and taste has been hiding inside the artifact this whole time. Your strongest designer may not be your fastest one anymore. Speed is the machine's job. Choosing the right problem and owning the call when evidence is mixed is the human's.

This is a relocation, not a loss. But it changes what you look for. A gorgeous portfolio tells you someone can produce. It does not tell you they can decide.

Research stops being a phase you skip

The tools are fluent enough to fake the work research was supposed to earn. Ask a model for a user persona and it writes one. Ask for a research readout and it hands you ten clean pages, every quote invented, every paragraph plausible. The document is done before anyone talked to a customer, and it looks exactly like the real thing.

That is why the fix is a habit, not more documents. Neeman points to continuous discovery, weekly contact with real customers by the team building the product. When the machine can fabricate a confident answer to anything, checking that answer against a live person is the thing standing between you and shipping fiction.

Breadth is close to free now. You can put ten directions on the table in the time one mockup used to take. So the job shifts to killing the weak ones fast, before your team gets attached, and the only reliable killer is evidence from someone not on the team.

The machine only knows the craft you can say out loud

Here is the part the panic skips. AI can carry your standards, catch your anti-patterns, and apply your definition of good across a thousand screens. It can only do that for the part of your craft you can put into words.

And most craft is tacit. You know a layout is wrong before you can explain why. Neeman leans on Michael Polanyi's line: we know more than we can tell. That gap is harmless when you do the work yourself. It becomes the whole problem the second you hand the work to a model, because it fills your silence with its own defaults, which are the average of everything it has seen. Average is exactly what craft is supposed to beat.

So the new literacy is description. Turn "I'll know it when I see it" into stated rules, named anti-patterns, an explicit account of what good means for this product. In April 2026, Google Labs open-sourced DESIGN.md, a single-file format that pairs machine-readable tokens with human-readable rationale and a validator for accessibility contrast. Standing context, written by a person, is the highest-value hour in your week.

Constraints are where the good work still comes from

Strip the tooling away and craft does not evaporate. It sharpens. STILFOLD builds load-bearing structures by folding flat sheet metal, no stamping dies, no fixed factory. Jonas Nyvang's line lands: the design is no longer hostage to the tool. The judgment about material and geometry becomes the whole game once the tooling bottleneck disappears.

Or look at Imago, the audio player from two Central Saint Martins grads. They used AI, but they set the constraints hard: no scraped data, everything processed offline, the artist owns the dataset. As Kieran Feechan says, every sound the model heard was recorded and owned by one musician who chose to be involved. The craft moved into the rules, not the output. Same lesson your team needs. The value is in what you decide to allow and disallow, not in the generation itself.

The deep cut

The move that changes your Monday is writing craft down. Not as a nice-to-have, as the way you both scale quality and defend headcount. The tacit knowledge in your best designers' heads is now your most valuable asset and your biggest single point of failure. If it stays in their heads, the machine ships the average and you cannot explain to finance why you need the team.

Start small. Take one live project. Write three things it must do and three it must never do, in plain sentences a stranger could follow. Hand that to the model before anything else. Then make it a standing rule: every rejected AI draft turns into one written sentence saying why it failed, added to your team's instructions. You are not just correcting output. You are building the thing that proves your team's judgment exists and belongs on the payroll.

Three questions for your team

1. When AI hands us three plausible directions, who owns the call on which one ships, and can they defend it in two written sentences? If the answer is "the loudest person in critique," you have a gap.

2. What is our written definition of good for our current project, and have we handed it to the model, or are we still expecting it to guess our taste?

3. When did someone on this team last talk to a real customer, and is that on the calendar weekly, or does it only happen when a launch goes sideways?