# AI Agents

When AI can plan and act across steps.

Track: AI Tools and Systems

## What is it?

An AI agent is an AI system that can work across more than one step. A normal chatbot mainly answers. An agent can plan, use tools, check information, take actions, and continue until a task is done or stopped. It might search a knowledge base, draft a reply, update a record, call another tool, and report back. The important word is not intelligence. The important word is action.

## Why it matters

AI agents matter because they move AI from advice into workflow. That creates value, but also risk. A draft you can review is one thing. An AI system that can send messages, change records, call tools, or trigger actions is another. The more an AI system can do, the more clearly you need boundaries, approvals, logs, and human control. Agentic AI is useful only when the task, tools, and limits are clear.

## How it works

An agent usually has a goal, instructions, access to tools, and some way to track progress. It breaks a task into steps, decides what to do next, uses a tool if needed, and checks whether the goal has been reached. For example, an agent handling a support request may read the customer message, search the help center, draft an answer, check the policy, and prepare the response for review.

## Analogy

Think of the difference between an advisor and an assistant. An advisor gives you suggestions. An assistant can book the room, send the note, update the list, and remind the team. That is powerful, but you would not give every assistant full authority on day one. You would define what they can do, what needs approval, and when they must escalate. AI agents need the same discipline.

## Example usage

An AI agent can help with customer support by finding the right article and drafting a reply. It can help sales by preparing account notes before a call. It can help operations by checking a checklist and flagging missing items. It can help a developer by reading an issue, looking at files, and suggesting a fix. The work becomes more useful when the agent has the right tools and clear boundaries.

## How to use this

Use agents for repeatable tasks with clear steps. Start small. Pick a task where the AI can assist, but a human can still review the outcome. Define what tools the agent may use. Define what it must never do. Define when it needs approval. Review the logs. If the agent can affect customers, money, systems, or data, keep human oversight in the path.

## Common mistake

The common mistake is giving an agent too much freedom too soon. A flashy demo can hide weak controls. If the agent can act, the question is not only whether it can complete the task. The question is what happens when it chooses the wrong step, uses the wrong source, or acts at the wrong time. Start with bounded tasks. Expand only after evidence.

## Question to ask

- **Task fit**: Is this a repeatable task with clear steps, or does it require judgment every time?
- **Tool access**: What tools or data should the agent be allowed to use?
- **Approval**: Which actions need human review before they happen?
- **Risk**: What is the worst reasonable mistake this agent could make?

## Quick quiz

What makes an AI agent different from a simple chatbot?

## Flashcard

**Question:** What is the key idea behind AI agents?

**Answer:** AI agents can plan and act across steps, often using tools. That makes boundaries and human oversight important.