Agents struggle with the same bad UX we do

Product designer Stephen Patterson found that AI agents operating websites fail on the same poor design that people quietly work around, and now uses those failures in design review.

Stephen Patterson is a product designer at Deck, a company that builds software agents which log into websites and operate them the way a person would. After watching thousands of agent runs on utility, payment and back-office portals, he found the agents stalled where people also struggle. People guess, retry or ask a colleague, so the problems never show up as failures; agents cannot do that.

• Failures cluster in five patterns. They are unclear labels, state shown only on hover or by colour, pop-up interruptions, sessions that expire mid-task, and success that is never confirmed.
• Most fixes that help agents are the same ones that help people using screen readers or keyboards.
• Some steps, such as a confirmation before moving money, should stay. He suggests letting a person grant an agent limited authority in advance instead of removing them.
• He now asks five questions in every design review, such as whether someone who cannot guess could finish the task and whether success announces itself.

A design team can use agent failures as a cheap first signal of usability problems, then confirm them with people, since agents also fail where people do not. This is a designer's account from a company that sells agent software.

Derived by Working Surface from the article; more in the Playbook. Source line: The places agents get stuck are almost never exotic.

4 November 2026: teams building with AI agents are turning each failure or correction into a permanent check, rule or shared record, rather than fixing one result at a time.