🤖AI Agents
AI that plans and acts on its own · 28 terms
Agent Basics
7What an agent is, how it runs, and the kinds of setup it comes in.
A program that uses a model to work toward a goal. It picks each step.
The cycle an agent runs: think, act, check. It goes round until the job is done.
Fixed steps written in code, where some steps ask a model to help.
The code around a model. It runs the loop, the tools and the rules.
A team of AI agents. Each has one job, and they work on one task.
An AI agent that works on one task for hours, often in the cloud.
How much an AI agent may do alone. Past that, it must ask a person.
Agent Abilities
10What an agent can do besides write text. It can use tools, remember, plan and pick up new skills.
A way for a model to ask an app to run a tool. The app sends back what it got.
A note that tells a model how to call a tool. It lists the inputs.
A tool's output, sent back to the model. It says what happened when the tool ran.
A model asks for many tools in one reply. Your code can run them at once.
A tool that lets a model write code and run it in a safe sandbox.
A model that works a computer. It looks at the screen, then clicks or types.
How an agent keeps facts from one task for the next one.
An agent that breaks a big goal into steps and ticks them off as it works.
A folder of steps that an agent reads when a task needs it.
When an AI agent reviews its own work and fixes what it finds.
Agent Orchestration
6How several agents split up a job, pass it along and talk to each other.
An agent that splits a task, hands out the parts and joins the results.
A helper agent that starts fresh, does one job and sends back a short note.
One agent hands the task to a better one. All that was said goes too.
Rules that let agents from two firms talk and work as one team.
Many AI agents work at the same time on one job. Then their results are joined.
One model makes a draft and a second one judges it. They repeat until it passes.
Agent Control
5How people keep an agent safe: when it must ask, what it may touch, and where it runs.
A person checks an AI's work and says yes or no before some steps.
Rules for an agent. They say what it can do alone, what it must ask about, and what it can't do.
A closed box where an agent can run code. It can't reach the real computer.
A rule that tells an AI agent when to stop. A turn limit is one.
The record of each step an AI agent took in one run.