# The Cost of Accessibility in AI — Jemima Abu, CAIS

## Executive summary

Accessibility (A11y) must be treated as a foundational requirement throughout the entire product lifecycle, from business requirements to QA, rather than a costly fix applied at the end. While AI agents can accelerate development, they are susceptible to inheriting existing code biases and tend to generate verbose, non-semantic code unless provided with explicit, mandatory accessibility skills and guidance. Implementing A11y early saves significant time and refactoring effort, making the product usable by all people, including those relying on assistive technologies.

## Key takeaways

- A11y is a prerequisite for usability: An inaccessible product is inherently useless, regardless of how quickly it is built by AI. A study by WebAIM shows that 96% of the top 1 million websites have some form of accessibility issue.
- AI agents require explicit guidance: AI agents are 'impressionable' and will absorb existing code biases (e.g., 'it works for me' mentality). To ensure accessibility, developers must provide specific skills and prompts, treating the agent like a child needing clear instructions.
- A11y must be integrated early: Accessibility must begin at the business requirements and design stages. Waiting until the product is complete leads to avoidable refactoring and regression work.

## Technical details

- Semantic HTML vs. Custom Code: AI agents often default to verbose code, reinventing the wheel when a native semantic HTML element could provide the required behavior. Recreating semantics and expected browser behavior adds unnecessary complexity and testing.
- Accessibility Skills for Agents: To guide AI agents, developers should enforce mandatory skills. The 'Modern Web Guidance skill' from Chrome is recommended as it ensures the agent uses the newest HTML and CSS features rather than custom JavaScript implementations. ARIA should only be used when semantic HTML cannot express the required behavior.
- Core A11y Principles: Every interactive element must be operable by a keyboard, properly labeled, and readable by screen readers. The process requires auditing existing interfaces and utilizing reusable component libraries (design systems) to ensure consistency.

## Practical implications

- Integrate accessibility auditing and testing into the earliest stages of the development pipeline (Design/Requirements).
- Establish and enforce mandatory accessibility skills/guidelines for all AI agent prompts.
- Prioritize using native semantic HTML elements over custom JavaScript implementations.
- Develop and maintain reusable component libraries (design systems) to standardize accessible patterns across the codebase.

## Topics

Accessibility, AI Agents, Semantic HTML, Web Development, Build Process, Usability, Jemima Abu

Source: https://www.youtube.com/watch?v=ZccJsdOE54M