What is an AI agent?
The difference between an AI that talks and an AI that gets things done.
A chatbot answers your questions. An AI agent goes a step further: it can take actions, like searching, booking, updating a record or writing code, to finish a task for you.
An AI system that works toward a goal by planning steps and using tools, instead of only replying with text.
A chatbot is someone who gives you directions. An agent is someone who drives you there, checking the map and adjusting the route along the way.
How an agent works
- 01Goal
- 02Plan
- 03Use a tool
- 04Check result
- 05Repeat or finish
The key idea is the loop. The agent doesn’t just answer once. It acts, looks at what happened, and decides what to do next until the job is done.
An example
- 1Looked up the order — using the orders system
- 2Checked tracking — using the courier's tracking tool
- 3Read the policy — found the rule for lost parcels
- 4Took action — requested the replacement
- 5Asked before the next step — kept you in control
Tools agents commonly use
Web search, calendars, email, databases, files, code, and other apps. A common way to connect these tools is MCP.
Why it’s powerful, and risky
- Powerful: it can complete multi-step tasks that used to need a person.
- Risky: a small mistake early on can carry through every later step.
- Good practice: limit what it can access, and require a human to approve important actions like payments or deletions.
A common misunderstanding
“Agents run completely on their own.” The most reliable ones keep a human in the loop for anything that matters.
Coding agents such as Claude Code read the files in a project, run the tests, read the errors and try again until the tests pass. Each pass through that loop is the agent deciding, acting and checking the result.
Related buzzwords
Harness · MCP · LLM · Prompt · Tool calling