What is an AI agent?
An AI agent is a software system that pursues a goal by taking actions in its environment, not just producing text. Microsoft defines it as 'a system that achieves a set goal by taking action based on the inputs it perceives in its environment.' In practice it's a language model plus tools plus a loop.
Why — the first-principles explanation
Strip away the hype and an AI agent is defined by one thing: it acts toward a goal. Microsoft's official definition captures it, 'a system that achieves a set goal by taking action based on the inputs it perceives in its environment.' The two key parts are perceiving inputs and taking actions.
Modern AI agents are usually built on a large language model. The model provides reasoning: it reads the goal and figures out what to do next. But a model alone only outputs text. So agents add tools, functions the model can call, such as web search, code execution, a calendar, or a database. That turns 'talking about an action' into actually doing it.
The third piece is the loop. The agent takes a step, observes the real result from the environment, and uses that to decide the next step, repeating until the goal is met. Anthropic describes this as the model 'dynamically directing its own processes and tool usage' while getting 'ground truth' from each result. Memory carries context across steps so it doesn't repeat itself.
Agents come in flavors. Microsoft lists reactive agents (respond to inputs by rules), model-based agents (build an internal picture of their world), goal-based agents (weigh actions against an objective), and utility-based agents (pick the action with the best expected outcome). Simple agents can be built with low-code tools; complex ones take weeks of design and integration. All share the same core: perceive, decide, act, repeat.
An example that makes it click
Think of a thermostat versus a personal assistant. A thermostat is a tiny 'reactive agent', it senses the temperature (input) and turns the heat on or off (action) to hit a set goal. Very simple, but it fits the definition: perceive, then act toward a goal.
Now scale up to a digital assistant you ask to 'plan my Friday.' It perceives your calendar and emails (inputs), then acts, books a meeting room, replies to confirm a lunch, and adds a reminder (actions), checking each result as it goes. Same recipe as the thermostat, just a smarter brain and more tools. That's what makes something an AI agent: it doesn't only answer, it does.
Key facts
- Microsoft's definition: an AI agent is 'a system that achieves a set goal by taking action based on the inputs it perceives in its environment.'
- Technically, an agent = a language model (reasoning) + tools (actions) + a loop (perceive, act, observe, repeat).
- Microsoft lists four types: reactive, model-based, goal-based, and utility-based agents.
- Anthropic distinguishes agents (which direct their own tool usage) from workflows (which follow predefined code paths).
- Agents range from simple low-code bots to complex systems needing weeks of design, training, and integration.
▶ The 60-second explainer (script)
What is an AI agent? Simply put, it's software that pursues a goal by taking action, not just chatting. Microsoft's own definition: a system that achieves a set goal by taking action based on the inputs it perceives in its environment. So there are two parts, it perceives, and it acts. Most agents today are built on a language model, which handles the thinking. But a model alone only produces text, so agents get tools: web search, code, a calendar, a database. Those tools turn ideas into real actions. Then it runs a loop, take a step, see the actual result, decide the next step, until the goal is done. Even a thermostat fits the pattern: sense the temperature, adjust the heat. A digital assistant just has a smarter brain and more tools. The one-line takeaway: an AI agent doesn't only answer, it does.
What authoritative sources say
People also ask
Is an AI agent the same as ChatGPT?
Not exactly. A plain chatbot answers and stops. An agent uses the model plus tools in a loop to take real, multi-step actions. Chatbots can be turned into agents by adding tools.
What are the types of AI agents?
Microsoft lists four: reactive (rule-based), model-based (build an internal world picture), goal-based (weigh actions against an objective), and utility-based (choose the best expected outcome).
Does an AI agent need a language model?
Not always, simple agents like a thermostat use rules. But most modern AI agents use a language model for reasoning, plus tools to act.
Is an AI agent the same as AGI?
No. Agents are narrow AI focused on specific goals and tools. AGI would be human-level general intelligence across almost any task, which doesn't exist yet.