Skip to content
Article Deque Systems Jul 2026

Deque: For your business to thrive in the agentic AI era, treat accessibility as foundational

Preety Kumar, CEO of Deque Systems, published this post in July 2026 to argue a point that many AI design discussions miss: the infrastructure that makes pages accessible to people with disabilities is the same infrastructure that AI agents rely on to do their work.

The connection is technical. AI agents interact with web pages primarily through the accessibility tree — a structured representation built from semantic HTML, ARIA attributes, and WCAG conformance. Google’s own guidance for agent developers says the accessibility tree is their primary data model. This means that every missing label, every non-semantic container, and every interaction that fails to expose state through ARIA is a place where an agent will struggle, fall back to screenshot-based interpretation, or simply abandon the task.

The scale of the problem is significant. WebAIM’s 2026 Million report found detectable accessibility failures on 95.9% of home pages. A separate survey of 200 engineering leaders found that 88% trusted their AI-generated code to be accessible. Independent evaluations of AI-generated code have not supported that confidence.

Kumar’s economic argument is straightforward: fixing an accessibility issue at the design stage costs around $21 and under 15 minutes of work. Fixing the same issue after production costs over $637. Because AI agents amplify the speed of interaction — repeating steps that would take a human several minutes in a fraction of a second — the effect of an accessibility failure scales with the agent’s throughput. A single broken form field that a user might encounter once becomes something an agent hits hundreds of times in a single session.

The practical recommendations are not new, but the framing gives them new urgency:

  • Write semantic HTML rather than wrapping everything in <div>.
  • Label interactive elements clearly — what they do, not just what they look like.
  • Expose state changes through ARIA so that both screen readers and agents can track what is happening on the page.
  • Get leadership buy-in to treat accessibility work as ongoing, not as a periodic audit.

The audience for this piece is broader than it might appear. It is as relevant for teams building AI-powered products as for those optimizing existing products for AI agent use. If an agent-driven workflow depends on a sequence of UI steps, each of those steps needs to be accessible to work reliably.

The piece refers readers to Deque’s report, From debt to dividend, for detailed guidance on moving from reactive patching to proactive accessibility practice.