agent definitions — human-computer interaction

overview

hci-oriented definitions emphasize assistance, user modeling, and experience design. agents are cooperative partners that learn preferences, present information, and reduce cognitive load while remaining accountable to users.

signature traits

• user-centered autonomy: agents act on a user’s behalf but prioritize transparency and controllability.
• adaptive interfaces: learning personalization, context awareness, and interruption management define success.
• collaborative framing: agents complement rather than replace human decision-making, aligning with mixed-initiative interaction.

illustrative definitions

• 1994 — pattie maes, “agents that reduce work and information overload”: software agents learn to filter information and make suggestions.
• late 1990s — interface agent literature (lieberman, etc.): focus on mixed-initiative systems that negotiate task division with users.
• modern llm tools — cursor, claude, chatgpt agents: extend the tradition with tool use, approvals, and user-in-the-loop guardrails.

relation to other dimensions

• autonomy spectrum: typically mid-to-high autonomy with explicit return-to-user checkpoints.
• entity frames: primarily machine-centered but tightly coupled to human supervisors—practical hybrids.
• goal dynamics: strong on adaptation; agents interpret user intent, reprioritize tasks, and sometimes question goals.
• persistence & embodiment: can be session-bound or persistent, depending on memory design; embodiment remains digital.

open questions

• how should interface agents expose uncertainties from llm reasoning to maintain trust?
• what consent and approval patterns keep users in control when agents can execute real-world actions?
• how do we reconcile mixed-initiative interaction research with autonomous tool-calling loops emerging today?