The mockups are free now

The developers went first. Vibe coding, describing what you want and accepting what the model builds, went from a joke on a timeline to a hiring requirement in about a year. Design got its version straight after: Figma Make, Google Stitch and a dozen prompt-to-UI tools that turn a paragraph into an interface while you watch. I use them. They are good. That is the point.

What they produce is the median of everything they were trained on, delivered instantly and with total confidence. Competent spacing, sensible hierarchy, a rounded button in the right corner. The screen that would have taken a mid-level designer a day now arrives in a batch of five, and every one of them looks like the app you already have installed.

Put the five drafts next to each other and notice the strangest thing about them: they agree. Same structure, same tone, same defaults. A generator has no opinion about your product; it has an average of everyone else’s. My feed is currently split between people declaring the design job dead and people declaring the tools toys, and both camps are reviewing the wrong thing. The draft was never the job.

Before product design I spent years in performance creative, where generation was already cheap. A small team can produce fifty banner variants in a day; we often did. I designed and tested over ten thousand ads that way, and the honest lesson from all of them is that making variants was never the bottleneck. Knowing which one to run, why it won, and what fear the winning line answered: that was the entire job. Volume without judgment is noise with a media budget.

The same pattern held when I moved to product. On Monefit we built an acquisition engine around five hundred ads for one credit product. The value was never the five hundred; it was the reads between the waves, the calls about what to kill, and the one insight about hesitation that fed back into the signup flow itself.

Economics has a boring name for this: when something becomes free, its complements get expensive. The complements of a free mockup are the brief, the decision, and the accountability for being wrong. Nobody is selling those by the token.

The five drafts agree with each other because none of them knows your customer’s fear. The line that actually converts a payment screen is rarely a layout decision. It is “you won’t be charged today”, or “your money arrives Tuesday”, or “cancel any time, two taps”. That sentence lives in your support tickets, your test results and your funnel data, none of which were in the training set. A generator can draw the screen around it, but it cannot know the sentence is missing.

Taste, meanwhile, is mostly removal, and generators only add. A model will happily give you the feature, the chart, the badge and the carousel; it will never argue that the screen should not exist. When I built a health platform from a blank page at Emerald, the work that mattered was deciding that 115 biomarkers should land as one plain answer, and everything else should get out of the way. Every tool I have tried would have shipped the dashboard with all 115 on it, beautifully.

And the drafts are relentlessly sunny. Generated flows demo the happy path because the happy path is what the training data photographs. The screens that decide whether people stay, the declined application, the empty dashboard, the failed payment, are exactly the ones nobody prompts for. I wrote a whole note on that: design the bad day first.

I co-founded a startup once, and the deck everyone praised was full of screens that later lost, politely and thoroughly, to what real users actually needed. Cheap drafts make that trap easier to fall into, not harder: the more finished something looks, the less anyone questions whether it is true.