The Bottleneck Was Never the Drawing

Most of the discussion around AI and design has settled on throughput. Generate more variations. Fill the artboard faster. Produce twenty options where you used to produce two. It is an easy story to tell because the output is visible, and visible output demos well.

It also misidentifies the problem. In almost every project I have worked on, the slow part was not rendering the interface. It was working out which interface deserved to be rendered. Generating screens faster does not touch that, and in some teams it actively makes things worse, because a wall of plausible options creates the feeling of progress without the substance of a decision.

Where the Leverage Actually Is

The teams getting genuine value are using these tools somewhere less photogenic — upstream of the canvas, and in the parts of the work nobody puts in a case study.

Interrogating the brief. Before any pixels, pressure-testing what the user is actually trying to accomplish, and what breaks if the core assumption is wrong. This is cheap to do now and it used to be expensive.

Surfacing the edge cases. The empty state. The name that runs to four hundred characters. The payment that fails halfway. The user who is offline on a plane. These are where products actually break, and they are precisely the states that get skipped when the deadline tightens.

Arguing with the work. Having something that will push back on a direction at two in the morning, with no investment in whose idea it was.

The Part That Does Not Get Cheaper

Generative tooling makes the cheap part of design cheaper. It leaves the expensive part almost entirely untouched: the judgment about what should exist, for whom, and at what cost to everything else on the screen.

That judgment is still the job. Speed at producing the wrong thing has never been a win, and a faster route to it is not an improvement.