How does an AI agent differ from a regular chatbot?

Question

Accepted Answer

A regular **chatbot** just replies with text — you ask, it answers, and the exchange ends there. An **agent** runs a loop: it can call **tools** to take actions, observe the results, and iterate toward a goal until a stopping condition is met. The difference is doing, not just talking.

## The agent loop

```text
Goal → [think → call a tool → observe result → think again] → repeat → stop when done
```

At each step the agent decides what to do next based on what it has learned, rather than producing one answer and stopping.

## Concrete agent actions

An agent's "tools" let it act on the world:

- **Read and write files** — open a source file, edit it, save changes.
- **Run commands** — execute a build, run the test suite, install a package.
- **Search** — grep the codebase, query a database, look something up.
- **Call APIs** — hit an external service and use the response.

## Example

```text
Goal: "Fix the failing test."
  1. Run the test suite        → observe: test_login fails
  2. Read the failing test     → observe: it expects a 200, got 500
  3. Search for the handler    → observe: missing null check
  4. Edit the handler file     → write the fix
  5. Run the test suite again  → observe: all pass
  → stopping condition met: tests green, done
```

A chatbot could only *describe* a possible fix; the agent actually applies it and verifies the result by running the tests itself.

## Why it matters

Understanding the agent loop explains why agents can complete real multi-step work — debugging, refactoring, research — instead of only describing it. The tool-call-observe-iterate cycle, with a clear stopping condition, is what turns a language model from a conversational text generator into something that actually gets tasks done.