# AI agent

> An LLM that can take actions, check the results and keep going until the job is done.

- Level: Intermediate
- Topics: AI
- URL: https://tidbit.legios.com.ar/ai-agent/
- In Spanish: [Agente de IA](https://tidbit.legios.com.ar/es/ai-agent/)

An **AI agent** is an LLM wrapped in a loop with tools. A plain chatbot answers once and stops. An agent receives a goal, decides what to do, uses a tool (search the web, read a file, run code, call an API, send an email), looks at the result, and decides the next step. It repeats until the goal is met or it gets stuck.

The tools are what make it an agent. Each tool is a function the developer exposes with a description; the model chooses when to call it and with what arguments. A coding agent like Claude Code, for example, can read your project, edit files, run the tests, see them fail and fix the code, without a person typing each command.

This is powerful and also where the risks live. An agent that can act can also act wrongly: delete the wrong file, buy the wrong ticket, loop forever. That is why serious agents ask for confirmation before irreversible steps, run in sandboxes with limited permissions, and keep a log of what they did.

Why it matters now: in 2023 the question was 'what can the model say?'; by 2026 it is 'what can the model do?'. Agents are how AI moves from answering questions to completing tasks, which is where most of the economic value, and most of the open problems, are.

## An analogy

A chatbot is a consultant who writes you a memo. An agent is an assistant with your calendar, your inbox and a company card, who gets the trip booked and tells you when it is done.

## Keep in mind

- Agent = model + tools + a loop: think, act, observe, repeat.
- Tools are ordinary functions or APIs the model can choose to call.
- Agents act on the world, so permissions, confirmations and logs matter.
- Open standards like MCP (Model Context Protocol) let agents connect to tools and data in a uniform way.

## Go deeper

- [Anthropic: building effective agents](https://www.anthropic.com/research/building-effective-agents)
- [Intelligent agent on Wikipedia](https://en.wikipedia.org/wiki/Intelligent_agent)

## Related terms

- [LLM](https://tidbit.legios.com.ar/llm/): A neural network trained on huge amounts of text to predict the next word.
- [Chatbot](https://tidbit.legios.com.ar/chatbot/): A program you talk to by typing or speaking, as if it were a person.
- [API](https://tidbit.legios.com.ar/api/): The menu of things one program lets another program ask for.
- [Claude](https://tidbit.legios.com.ar/claude/): Anthropic's family of AI models, and the assistant app built on them.
- [Sandboxing](https://tidbit.legios.com.ar/sandboxing/): Running a program in a closed box so it cannot harm anything outside it.
- [Artificial intelligence](https://tidbit.legios.com.ar/artificial-intelligence/): Getting computers to do things that, until recently, only people could do.
- [Open standard](https://tidbit.legios.com.ar/open-standard/): A public rulebook anyone can follow, so products from different makers work together.
- [Model Context Protocol](https://tidbit.legios.com.ar/model-context-protocol/): An open standard for plugging tools and data into AI assistants the same way every time.