Product engineer and accessibility professional Jemima Abu explains how to make accessibility part of an AI-assisted development workflow.

Start with the people using the product

  • A product must work for people using screen readers, keyboards and other assistive technology, not just for the developer who built it.
  • Users who cannot navigate a site may leave; their absence is not evidence that accessibility is unnecessary.
  • Retrofitting accessibility at the end is like pushing blueberries onto a baked muffin: Abu argues it needs to be baked in.

Give agents a repeatable standard

  • Define requirements for the platform and audit the existing codebase, whose patterns the agent may copy.
  • Use a reusable design system so fixing a shared component can improve many pages.
  • In agent instructions, require semantic HTML and keyboard-operable interactive elements; use ARIA where appropriate.
  • Prefer suitable native browser elements, such as details for an accordion, over needlessly custom implementations.

Treat the experiment as a starting point

  • Abu compared the same Figma implementation with and without an accessibility skill: in her run, the guided version took five minutes rather than fifteen and cost 58% less.
  • She explicitly cautions that this was a single, uncontrolled experiment, not proof that accessibility guidance always saves time or money.
  • Instructions and automated checks can help, but they do not establish that a product is fully accessible; testing with people and assistive technologies remains important.

“The web is for everyone, so everyone should be able to use it.”