Designer says AI agents are intermediaries, like medical go-betweens, not users

This week a discussion unfolded on Habr about how to classify AI agents in interface design. Kiara Pellegrini of Nielsen Norman Group proposed a simple logic: an agent has goals, interacts with an interface and tries to achieve those goals. So it is a user. But a UX designer with eight years of experience as a doctor disagrees: this is not just an imprecise definition but a deep misunderstanding of the architecture.

Why an agent is not a user

An agent clicks buttons, fills in fields and gets results, just like a user. But there is a critical difference in motivation and responsibility. A user acts for themselves. An agent acts for you, representing your goals through interpretation. This transforms the entire dynamic of interacting with an interface.

When a designer designs for a user, they create a channel of direct access: a person sees something, understands it and acts. A mistake is their own mistake of understanding or attention. When an intermediary enters the chain, new points of failure appear. An agent may misread your intent, may forget context, and may run into an exception without knowing how to react.

Medicine has known this question for a century

A nurse in a hospital is a classic intermediary. A doctor writes a prescription, the nurse carries it out, the patient receives the medication. Each level can distort the information: the doctor wrote the dose unclearly, the nurse misread it, the patient forgot the instructions. That is why medicine developed an entire culture of error insurance:

• Double-checking of critical procedures
• Standardized forms and protocols instead of free-form text
• Alarms when readings deviate from the norm
• Mandatory documentation of every step
• Clear feedback channels between levels

The result: medicine designs not for an ideal intermediary but for a real one, who can make mistakes.

What this means for interface design

If we accept that an agent is an intermediary and not a user, the designer's task changes. Interfaces need to not merely "work" but to work reliably even when the intermediary makes mistakes or does not fully understand the context. That means explicit confirmations before critical actions, detailed logging of every step, the ability to roll back an operation, and making the system state readable to the agent at every stage.

Today most interfaces are designed for direct human interaction. When you throw an agent at them, they often break, because there is no safeguard against the intermediary misunderstanding the design's unspoken agreement.