Summary
The study tested 1,560 agents from six commercial models across 13 tasks and six websites. Agents used 43% more tokens on average on inaccessible sites, with the largest measured compute increase reaching sixfold.
Why it matters
Web accessibility is becoming an operational dependency for agentic commerce and automation, not only a compliance or human-usability concern.
Related coverage
- Gurucul launches runtime risk controls for enterprise AI activity
- Verana Health releases AI agents for life-sciences research
- Dun & Bradstreet exposes governed commercial intelligence through agents