Design with AI

A design earns its place through the work it helps a person do. AI can help explore alternatives, critique a proposed interaction and prepare a prototype or research plan. Evaluate those contributions against the person’s task and the evidence available.

Follow an appointment form through the task

The manifesto guide’s hypothetical appointment form promises that someone can choose a time and submit without a mouse. A screenshot shows the finished page. Automated tests check required fields and saved entries. Neither check exercises the date picker’s keyboard behavior.

In the imagined failure, focus enters the calendar and cannot leave. Keyboard focus is the control currently receiving keyboard input. A person who depends on it cannot finish the form merely because the page looks complete.

Work through the promised interaction:

Task stepWhat to inspect
Reach the date pickerA keyboard route reaches the control and makes its current focus understandable.
Choose or change a dateThe required selection can be made with the keyboard.
Leave the calendarA keyboard method returns the person to the rest of the task.
Correct an errorThe person can discover the problem, reach the relevant field and change it.
Submit and inspect the resultThe form completes and communicates the actual result.

The W3C keyboard guidance addresses operation through a keyboard interface. Its separate no-keyboard-trap guidance explains leaving a component with the keyboard. A modal can keep focus inside while it is open if the person has a usable way to dismiss it; focus containment alone does not establish a trap.

Report the interaction you actually examined

For this constructed case, the claimed keyboard path is incomplete. A useful failure report names the step, expected exit method, observed behavior and the browser or other conditions used in an actual test. After a repair, repeat the path through correction and submission rather than checking only that focus moves once.

A successful keyboard demonstration covers the observed path. Wider accessibility also involves other requirements and conditions. Usability research asks further questions: can the intended users understand the labels, recover from interruptions and complete the task in their own circumstances? The user-research guidance provides a starting point for that work.

Put AI assistance around the evidence

An assistant can list likely failure cases, inspect markup, prepare a test sequence or organize permitted observations. It can also suggest an interaction to prototype. Check those proposals against the actual design and sources. An invented user reaction cannot replace an observation from a person.

Start with one task whose outcome you can recognize. Preserve the original design, explain what a proposed change is meant to improve and compare the full interaction after the change. Keep a useful detail when removing it would make the task harder, even if the revised screen looks simpler.

Maintained references and contribution

What would show that the work is done? develops the evidence question. Five checks for engineering decisions includes a separate design case about two independent settings. The thinking framework, by Rajiv Pant, supplies the shared reasoning method; W3C maintains the accessibility references linked above.

Designers can contribute a worked example or accessibility correction. Include the task, the interaction examined, what changed and the conditions under which it was checked. Clearly distinguish a constructed exercise from research with participants.

Design articles

Subscribe by RSS

  1. Methods

    Thinking through a strategic decision with AI

    Five questions help leaders examine assumptions, consequences, uncertainty, analogies and user needs with AI before committing to a strategic decision.

    Rajiv Pant