← All terms

define agent --plain-english

Illustration for "Agent" — Day 10 of the Non-Technical Technical Dictionary

Agent

TLDR:An AI that takes actions, not just answers.

A chatbot answers. An agent acts.

Ask ChatGPT a question and it replies. You ask, it answers, done. One exchange. Useful, but it's just talk. It can write you a beautiful email and then sit there, unable to send it. That's the ceiling on a plain chatbot: it hands you words and stops.

An agent doesn't stop. You give it a goal, and it works in a loop until the goal is met. Take a step, look at what happened, decide the next step, repeat. Same engine underneath, totally different posture. One talks. One does.

Here's what it actually looks like when you hand an agent a real job. Say you tell it "find last week's refunds and post the total in Slack." It doesn't answer in one breath. It runs a little cycle:

  1. Pick a step. "First I need the orders." It calls the Shopify window and pulls them.

  2. Look at what came back. It reads the orders, spots the refunds, adds them up.

  3. Decide the next step. "Now I post the number." It calls Slack and drops the message.

  4. Check if the goal is met. Posted? Done. Not done? Loop again.

A chatbot would have told you how to do that. The agent went and did it, one step at a time, adjusting as it learned what each step turned up.

What lets it act is tools. An agent on its own is still just words in a box. Hand it tools and it can reach into the real world:

  • call an API (order off another app's window)
  • run commands in the terminal
  • read and write files on your computer
  • send an email, update a price, fire off a message

Every one of those is a thing happening, not a thing described. That's the line between the two: a chatbot produces sentences, an agent produces outcomes.

The difference is asking someone for directions versus putting them in the driver's seat. One hands you a list of turns and wishes you luck. The other actually gets you there.

A word of grounding, because "it does the work for you" can sound like magic: an agent is only as trustworthy as the steps it takes, and it can be confidently wrong mid-loop the same way any model can. So the more an agent can touch, the more you want to watch what it's reaching for. Powerful and unsupervised is a combination you earn slowly.

That's why "agent" is the word of the moment. The leap isn't smarter answers. It's AI that can pick up the goal and go do the work. Without tools and a loop, you've got a really good talker. With them, you've got something closer to an employee.