Your Software Has a Second User Now. And It Isn’t Human.

AI agents are starting to use websites, and this presents new challenges for user experience design. I observed an AI agent navigating a site, making mistakes like a beginner computer user. The agent could write code, read docs, and give convincing explanations for its errors.

Designing for humans, we focus on familiar interfaces and intuitive actions. However, AI agents may not perceive a page the same way. They might see a screenshot, an accessibility tree, or a structured control representation. Buttons and layouts that seem obvious to humans can be confusing for agents.

AI agents persevere where humans might give up. A successful API response may show no error, but the agent might not understand the outcome, leading to accidental multiple purchases. Monitoring dashboards could miss these inconsistencies, as a 200 OK status doesn't guarantee a successful user action.

With AI agents, traditional user sessions become ambiguous. A successful API response doesn't indicate human comprehension. This challenges analytics and raises questions about whether a user or agent performed an action. Without visibility into the session, investigating these scenarios becomes difficult.

Designing for AI agents may require rethinking what constitutes a session. Analytics tools will need to account for agent behavior and potential confusion. We must consider how AI agents perceive and interact with interfaces, ensuring clear outcomes and user understanding.