Agent

A model that doesn’t just answer but acts: it can run tools, search, write files, and take step after step toward a goal, deciding what to do next on its own.

An isometric cutaway: a boxy little robot alone in a kitchen, one hand on the open fridge, the other at the stove, an egg carton waiting on the floor.

An agent is an LLM wired up to tools and put in a loop. Instead of producing one answer and stopping, it can decide to search the web, read a file, run some code, or call another program, look at the result, and decide what to do next. It keeps going until it thinks the job is done, or something stops it.

The model itself hasn’t changed; what changed is the plumbing around it. Each tool result gets fed back into its context window , and the model writes its next move as if continuing a log of its own work. Coding assistants that edit a whole project, research tools that read dozens of pages, and browser bots that click through websites are all agents.

Agents are powerful for the same reason they’re risky. A chatbot that hallucinates gives you a wrong sentence; an agent that hallucinates can take a wrong action, and then build the next ten steps on top of it. That’s why so much alignment work has shifted from ‘what will it say’ to ‘what will it do.’