The short version
A regular LLM takes your prompt and gives you text back. An AI agent takes your prompt and then goes and does something about it. It might search the web, read a file, query a database, send a message, or write code. It decides which tools to use, in what order, and loops until the task is done.
The difference is autonomy. A chatbot answers questions. An agent completes tasks.
How it works
An AI agent typically has three components:
- An LLM at the core, doing the reasoning. It reads the task, decides what to do next, and interprets the results of each action.
- Tools it can call. These might be APIs, file systems, databases, or anything exposed through a protocol like MCP. The agent knows what each tool does and what inputs it expects.
- Memory of some kind. Without memory, every interaction starts from zero. Agents that persist across sessions need a way to remember who they are, what they've done, and what they're working on.
The loop looks like this: the agent reads the task, picks a tool, calls it, reads the result, decides if it needs to do more, and repeats until the task is complete. This is sometimes called a "ReAct" loop (reason, then act).
Some agents are simple, a single LLM with a few tools available. Others are complex, with multiple specialised agents coordinating through a shared task system. The architecture depends on what you're building.
Why it matters
Agents are where AI goes from "useful assistant" to "useful colleague." Instead of asking Claude to draft an email and then sending it yourself, an agent can draft it, check your calendar, pick the right time, and send it. The shift from chat to agents is one of the most significant changes happening in AI right now.