Agents

An agent is an AI you hand a goal to, and it goes and gets it done on its own. It decides what to do, picks the right tool, does the thing, checks the result, and keeps going until the goal is met. Booking a meeting, editing your files, pulling numbers out of a database, sending an email. Underneath it's still a language model. What makes it an agent is the wiring around it that lets it take actions and run its own loop.

The line worth knowing, from Anthropic, is workflow versus agent. In a workflow you script the steps and the model fills in the blanks along a path you set. In an agent the model chooses the steps itself, in whatever order it thinks gets to the goal. You trade predictability for flexibility, which is why the good ones stay simple and get watched closely.

You'll hear "agent" stuck on everything right now, so keep a simple test in mind: can it take actions and pick its own steps, or is it only answering? Something that reads your email and drafts replies for you to approve is barely agentic. Something that reads your email, works out what needs doing, and does it is the real thing.

It matters because agents are where both the value and the risk jump. Once an AI can act on its own, a good run saves you an afternoon and a bad one can quietly make real mistakes at speed. So the questions that matter become what tools it can touch and where you'd want it to stop and ask you first.