AI can produce a polished landing page before the meeting about the landing page has ended. The harder question is whether anyone should use that page.

That is not the same thing as dramatically improving interfaces.

This distinction matters because the industry often measures AI design progress through output volume.

A prompt produces five landing pages instead of one. A tool creates an application shell in minutes. A design agent can populate frames and generate variants.

The production-speed gain is visible in many tools; the size of any gain depends on the team and the task.

The experience gain is harder to prove.

In my experience, much generated UI converges on familiar patterns: rounded cards, gradients, dashboard grids and a prompt field where a product decision should be. That is an observation, not a measured industry-wide rate.

This is not necessarily a failure of the models.

It is partly a failure of context.

AI performs better when it has access to real design systems, production components, content, business rules and user goals. Without those constraints, it predicts plausible design.

Plausible is not the same as specific.

Figma’s recent product direction reflects this. Its MCP and agent tooling increasingly focus on giving AI structured design context and allowing agents to work with existing systems rather than generating isolated visual outputs.

That is where the genuine UX improvement begins.

AI can help teams explore more alternatives. It can accelerate prototyping. It can identify consistency problems. It can move implementation closer to design intent. It can reduce repetitive production work.

But the final experience still depends on decisions.

What deserves emphasis? What should remain hidden? How much automation should be exposed? What does the user need to trust? What happens when the system is uncertain? Which interaction is appropriate for this audience rather than the average user represented in a training set?

These are not questions that disappear because generation becomes faster.

In fact, faster generation makes judgment more important because teams can now produce bad options at extraordinary speed.

Figma reports that 91% of respondents in its State of the Designer research said AI tools helped them “uplevel” their work. That is a self-reported assessment of work, not a measurement of the usability of shipped products.

But “uplevel” should not be confused with automatic design quality.

AI has improved the design process substantially.

Whether it improves the product depends on what humans do with the extra capability.