Agent

Agents Beginner Also called: AI agent
Definition

An AI that does the task instead of describing how. It decides the steps, uses tools such as web search or file access, checks the results, and keeps going until the job is done. A chatbot answers; an agent acts. That is what makes agents useful, and what makes their mistakes expensive.

In plain English

Ask a chatbot how to rename 200 files and it tells you. Ask an agent and the files get renamed. An agent is a model that’s been handed tools and told to keep going until the job is done.1

In the real world

  • Claude Code, Codex, and Cursor working through a project’s files.
  • A “deep research” mode that searches, reads, and writes a report while you wait.
  • A support bot that actually issues the refund instead of explaining the policy.
  • Always-on agents that keep working after you log off: OpenAI’s Dots and SpaceXSI’s Grok Bot.34

When to use it

  • Tasks with a clear finish line and lots of steps. “Clean up this folder.” “Find every invoice over $500.”
  • Work you’d hand to a careful intern with a checklist.

When not to use it

  • Vague goals. “Make our marketing better” is not a task.
  • Anything irreversible without a human checkpoint. Payments, deletions, sending.
  • When the steps never change. That’s an automation, not an agent. Anthropic draws the same line between workflows and agents.1

Good, bad, ugly

  • Good: hours of clicking become a sentence.
  • Bad: it needs tools, permissions, and a clear definition of done, which is work.
  • Ugly: an agent that confidently finishes the wrong task. It did everything you said and nothing you meant.

Key facts

Rough formula
A model, some tools, and a loop that keeps going until the job is done
Versus a chatbot
A chatbot answers. An agent does.
The catch
It acts, so its mistakes have consequences

Common questions

What's the difference between an agent and ChatGPT?

ChatGPT in a plain chat answers you. An agent can open files, search, run code, and send things, then check its own work and continue. Same kind of model, more permission. Try the deep research mode in ChatGPT or Claude to see one work.

Are agents reliable?

For bounded tasks with a clear definition of done, often yes. For vague open-ended goals, they wander. Anthropic's advice is to start simple and add autonomy only when it earns its keep.

What tools do agents use?

Whatever they're given: web search, file access, a terminal, your calendar, a database. See tool calling.

References

  1. Building effective AI agentsAnthropic Engineering · 2024-12-19View source
  2. What are AI agents?IBMView source
  3. OpenAI launches dots, its bubbly agentic avatarTechCrunch · 2026-09-29View source
  4. Introducing Grok BotSpaceXSI · 2026-08-11View source
Updated 2026-11-10 Reviewed by Elena Boyd Agent graph