I use AI to reduce the time between an idea and something the team can inspect. It is most useful for exploration, repeated production work, and coded prototypes.
The output still needs the same design judgement and review as anything produced by hand.
Give it controlled context
I begin with the problem, user goal, constraints, and expected output. For interface work, I also provide the relevant component API, tokens, content rules, and accessibility requirements.
I keep sensitive customer and business information out of tools that are not approved for it. When realistic data is needed, I create representative test content instead.
Use it for the first working version
AI helps me turn a defined interaction into a coded prototype, create component stories, generate test scenarios, or compare implementation options.
This is especially useful for behaviour that is difficult to show in Figma. Responsive changes, keyboard interactions, loading states, and validation can be tested in the browser.
I keep the task small enough to review. Large prompts can produce a lot of code, but they also make mistakes harder to find.
Check the result
I review the output against the original design intent and the existing system. I test keyboard use, focus order, zoom, content wrapping, errors, and different viewport sizes.
An engineer reviews code that may move towards production. We check whether it follows project conventions, uses shared components, and avoids unnecessary dependencies.
AI gets us to a testable version faster. It does not approve the design, verify the code, or own the outcome. That responsibility stays with the team.