A good design process alternates between widening the field of options and narrowing it to a confident decision. AI is especially good at the widening part. Used well, it lets you consider more genuinely different directions in an afternoon, so your judgment has better material to work with.
Ask for different structures, not different colors
Variation in styling is cheap and rarely changes the experience. Ask instead for alternatives that change the underlying model. For a budgeting feature, you might explore a goal-first flow, a calendar-first view, and a conversational check-in. Ask the assistant to describe who each approach serves best and where each is likely to fail.
Design with real content early
Lorem ipsum hides problems. Use AI to generate realistic content for your scenario: long names, unusual addresses, short and long product descriptions, error messages, and empty states. Realistic content reveals layout problems and clarity issues long before a design review.
Then review that content for accuracy and tone. Generated microcopy often sounds polished while being vague. A button that says “Continue” is less useful than one that says exactly what happens next.
Stress-test every state
Ask the assistant to list the states a screen can be in: loading, empty, partial, error, offline, permission denied, and success. Ask what a first-time user, a returning expert, and a person using a screen reader would each need in those states. Treat the list as a checklist to verify, not as a design.
Prototype to learn faster
Coding assistants can turn a rough idea into an interactive prototype quickly. That is valuable when interaction is what you need to test, such as drag-and-drop, filtering, or a multi-step form. Keep prototypes small, use your design system’s components where you can, and click through every path yourself before showing it to anyone.
Make the decision visible
More options create more responsibility to choose well. Compare directions against the user goal, evidence from research, accessibility, technical constraints, and the outcome you are aiming for. Write down why you chose one direction and what would make you revisit it. That written reasoning is the part of the work that is unmistakably yours.
Try it yourself
Take a screen you are working on. Ask an AI assistant for three structurally different approaches and the tradeoffs of each. Sketch your favorite on paper, then list every state it needs and check how many you had already designed.