What is an AI agent?

An AI agent is software that uses an AI model to carry out a task for you, step by step, using tools such as a browser, files and your apps. A chatbot tells you how to do something. An agent does it and reports back.

Updated

A tidy desk by a window at night with a small glowing orb of light above an open notebook, a cup of tea, a plant and a desk lamp.

The short version

Give a chatbot the request "book a table for four on Friday" and it writes you a list of restaurants. Give an agent the same request and it opens the booking site, picks a time, fills in the form, asks you before confirming, and tells you when it is done.

Three things make that possible:

  1. A model that plans. The AI model decides what to do next, looks at the result, and decides again. This loop repeats until the task is done.
  2. Tools. The agent can do things, not only write: open web pages, run commands, read and write files, send messages, use the apps you connect.
  3. A place to work. It needs somewhere to run those tools: your own computer, a browser, or a computer of its own in the cloud.
A diagram of how an AI agent works. A goal enters a loop of three stages: plan, which picks one step; act, which uses tools on the agent's computer; and check, which reads the result. If the task is not done, the loop goes round again from plan. The agent leaves the loop in one of two ways: done, when the task is finished, or ask you, when it needs a decision or approval.
An agent plans a step, acts with its tools, checks the result and repeats until it is done or needs you.

What an agent can do

What to look for in one

QuestionWhy it matters
Where does it run?On your machine it can touch your files. On a separate computer, mistakes stay there.
Does it ask first?Sending, paying and deleting should wait for your OK.
Does it remember?Without memory you explain your preferences every time.
Can it start by itself?Schedules and triggers turn a helper into a teammate.
What does it cost?Each step uses a model, so long tasks cost more. Look for budgets.
Who can see your data?Check whether it trains on your data, and whether you can run it yourself.

Where agents go wrong

Agents make mistakes that chatbots can't, because they act. A model can misread a page, repeat a failing step, or follow instructions hidden in a web page it was only meant to read. Good agent software limits the damage: a sandboxed computer, approvals for risky steps, spending limits, and a way to undo.

How Casama fits

Casama is an open-source app for AI agents it calls teammates, or bots. Each bot has its own sandboxed Linux computer, a memory you can read and edit, and rules for what it must ask before doing. Bots run on schedules and triggers, and you reach them from apps on every platform. You can run it on your own server for free or use Casama Cloud.

Join the waitlist Run it on your own server