3 AI Interaction Styles: Turn-Taking, Task Agents, Always-On

Jakob Nielsen sets out design guidelines for always-on AI agents, including scope checks on instructions, a single page of standing rules and a visible status.

Jakob Nielsen is a usability researcher who writes the UX Tigers site. He describes three ways people now use AI: asking questions, assigning tasks and appointing AI agents to standing roles, as with OpenAI's Dots and Meta's Muse. He argues that each needs different design, and that the third brings new problems.

• Products should infer the right style from the request and confirm time and cost before spending, rather than making users choose a mode first.
• A passing remark can become a permanent rule, so the agent should ask whether an instruction applies once or from now on.
• Every standing rule should sit on one page where the user can read and revoke it.
• A quiet agent may have stopped working, so it needs a visible status, and its activity log cannot show what it missed.

Design teams building agents that act on their own can treat scope checks, a rules page and a status indicator as standard parts of the interface. This is an inference and is not the author's claim; Nielsen himself notes that the research evidence is thin.

Derived by Working Surface from the article; more in the Playbook. Source line: Users shouldn’t have to pick a rung before they speak.

15 October 2026: when AI agents make building cheap, the decision worth reviewing is the one made before the build, written where other people can dispute it.