AI Agents

An agent is an AI model that can act — not just respond. It can use tools, make decisions across multiple steps, and pursue a goal autonomously.

Models vs Agents

A model answers. An agent acts — it loops through observe, think, and act until the task is done.

📨

Observe

Receives a task and reads all available context

1/5

🧠

Think

Reasons about what needs to happen and in what order

2/5

📋

Plan

Decides which tools to use and sequences the steps

3/5

🛠️

Act

Calls tools, writes code, sends messages, reads files

4/5

✅

Respond

Returns the result — or loops back if the task is unfinished

5/5

0:00 / 0:08

The core difference

	Model (chat)	Agent
Input	Your message	Your goal
Output	A response	Completed work
Steps	One	Many
Tools	None	Web, code, APIs, files...
Memory	Session only	Persistent
Autonomy	None	High

A model answers questions. An agent accomplishes tasks.

The agent loop

Every agent runs the same basic cycle, over and over until the task is done:

┌─────────────────────────────────────────┐
│                                         │
│   OBSERVE → THINK → PLAN → ACT → OBSERVE │
│                                         │
└─────────────────────────────────────────┘

Observe — what's the current state of things? What information do I have?
Think — what does this mean? What do I need to do next?
Plan — break the goal into steps
Act — call a tool, write something, make a decision
Observe — what happened? Update the plan if needed

This loop runs until the goal is reached or the agent gets stuck and asks for help.

Types of agents

Reactive agents

Triggered by an event (a message, a file upload, a schedule) — do one thing and stop. Simple, predictable, great for automating specific tasks.

Example: An agent that monitors your inbox and drafts replies to customer questions.

Autonomous agents

Given a high-level goal and left to figure out the steps. More powerful, requires more oversight for high-stakes work.

Example: An agent that monitors your ad spend, identifies underperforming campaigns, and pauses them automatically.

Multi-agent systems

Multiple specialized agents working together, each handling a different part of a workflow. A "manager" agent routes tasks to specialist agents.

Example: A research agent that finds information + a writing agent that drafts the report + a review agent that checks accuracy.

What agents are good at (and not)

Great for:

Repetitive tasks with clear rules (invoice processing, data extraction)
Research and synthesis (gathering information from many sources)
First drafts (emails, reports, code)
Monitoring and alerting (watching for anomalies, sending notifications)
Multi-step workflows that currently require human hand-offs

Still needs human oversight:

High-stakes irreversible decisions (large financial transactions, legal commitments)
Tasks requiring genuine judgment in novel, complex situations
Anything where being wrong has serious consequences

⚠️

Think of current agents as a very capable, very fast intern. They do remarkable work but you still review anything important before it goes out.

Next: Agent harnesses →

Models Agent Harnesses