AI Agent

A software system that can interpret goals, use tools, and take actions with some autonomy.

When can a catalogue agent stop searching?

A reader asks for an available large-print edition of a particular novel. A fictional library assistant can search the catalogue and inspect holdings.

  1. Set the goal

    The title, format and availability all matter. A standard-print copy satisfies only part of the request.

  2. Inspect the result

    A title search finds two editions. The agent reads the holdings for the large-print edition and sees one available copy at the west branch.

  3. Stop and report

    The answer names the edition and branch and records when availability was checked. Reserving a copy would be a separate action.

What makes the search agentic?

The first result informs the next choice. A missing holdings record would call for an answer that acknowledges uncertainty. A search limit gives the agent a stopping point even if the catalogue never supplies a match.

How AI Agents Work

A common agent loop is simple: understand the goal, plan, act, observe, and revise. For example, an agent helping with travel might search flight data, compare options, ask a clarifying question, and format an itinerary. The model supplies reasoning and language ability, while tools provide access to live information and actions.

This is why agents are closely tied to tool use, function calling, system prompts, and guardrails. The model may propose actions, but the surrounding application decides what actions are actually permitted.

Follow the idea

Tool Use · Follow one call from request to result.

Function Calling · See how a request can be represented as structured fields.

Source · Anthropic on agent design and stopping conditions

Return to AI Agent in the A–Z glossary