The story
In 2022 you typed a question into a chatbot and got an answer. That was it. You were the hands. If the answer said 'book the 7pm table', you opened the booking site yourself. By 2025 the same kind of AI could open the site, find the table and book it, then tell you it was done. The AI did not get smarter overnight. It got hands.
The one plain idea
An agent is an AI model that runs in a loop and can take actions to reach a goal. A chatbot answers. An agent does. The word 'agent' is used loosely, so this course uses one clear test: does the model decide its own next step, and can that step change something in the world?
The kitchen-table picture
A chatbot is a friend on the phone who tells you how to fix the leaky tap. An agent is the plumber who comes over, looks at the tap, tries a wrench, sees it still drips, tries another fix, and calls you when it is done.
Worked example
Goal: 'Find the cheapest flight to Chicago on Friday.' A chatbot writes: 'You could check Google Flights or airline sites.' An agent opens a flights tool, searches, sees 14 results, sorts by price, notices the cheapest has a 9-hour layover, searches again with a layover limit, and replies: 'Best option is $212, nonstop, 6:10am.' Count the decisions: search, sort, reject, search again. The model made each one. Nobody wrote those steps in advance.
No-code exercise
Pick five tools you use (a calculator, a search box, a spreadsheet formula, a Roomba, a GPS). For each, write: does it decide its own next step? Does it change something in the world? Only call it agent-like if both are yes. Most will be no. The GPS is closest, and you will see why when you reach chapter 2.
Self-check
- What is the single difference between a chatbot and an agent? Answer: A chatbot only answers. An agent also takes actions and decides its own next step in a loop.
- In the flights example, who chose to search a second time? Answer: The model did, after seeing the first results. No one scripted it.
- Is a fixed script that always does steps 1, 2, 3 an agent? Answer: Not by the test used here. The steps were decided in advance by a person, not by the model. Anthropic calls that a workflow (see below).
Sources
- Anthropic, Building Effective AI Agents: 'Workflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents are systems where LLMs dynamically direct their own processes and tool usage.' https://www.anthropic.com/engineering/building-effective-agents General knowledge / our framing: The chatbot-to-agent story and the plumber picture are our own teaching framing. The word 'agent' has no single official definition; Anthropic's post itself says people define it in several ways.