Vendor-neutral · Beginner · 5 lessons · about 34 minutes · updated 2026-09-12
What Is an AI Agent? From Chatbot to Agent, No Code Required
An agent is a chatbot in a loop with permission to act. Five lessons on what that loop does, why every turn of it costs more than the last, and how to keep the bill boring.
For people who use ChatGPT, Claude or Gemini in a browser, keep hearing the word agent, and want to understand what it means and what it costs before they turn one on. No code, no terminal. Every number is worked in the open.
Course outline
- 1What an agent is: a loop that reads, decides and actsA chatbot answers once. An agent reads the situation, decides on an action, acts, reads the result and goes round again until the job is done or something stops it. This lesson makes that loop concrete with one narrated run.6 min · 3 objectives
- 2What a tool call is and why it costs tokensA tool call is the model asking for something outside itself. The result comes back as text, gets added to the conversation, and is re-sent on every later step. That is why a 12-step run bills 166,800 input tokens, and this lesson prices it on four models.7 min · 3 objectives
- 3Try an agent-style task in an ordinary chatbotYou do not need an agent product to get the agent habit. Ask a chatbot to write its plan first, execute it step by step, and flag where it would have needed a tool. You see the loop before you pay for one.7 min · 3 objectives
- 4When a $20 seat beats the API for agent tasks, and when it does notA $20 seat with an agent mode is a fixed price for a capped amount of looping. The API is a per-step meter with no cap unless you set one. Work out which is cheaper at your volume using the reference trace.7 min · 3 objectives
- 5The five caps that stop a runaway loopAn agent fails expensively in one way: it loops. Five caps, in the order they save money, and a compare exercise on two agent designs so you can tell a safe loop from a bill waiting to happen.7 min · 3 objectives
Frequently asked
What is an AI agent?
A model given tools and a loop: it reads the situation, decides an action, calls a tool, reads the result and goes round again until the task is done or a cap stops it. The model is the same one behind a chatbot; the loop and the tools are what make it an agent.
How much does an AI agent cost to run?
On the site's 12-step reference run, about $0.0089 on DeepSeek V4 Flash, $0.139 on Gemini 3.8 Flash, $0.37 on Claude Sonnet 5 and $0.47 on GPT-5.4 without caching. Longer runs cost disproportionately more because every step re-sends the whole history.
Do I need to code to follow this course?
No. There is no terminal and no script. The Try it and fill-in boxes run inside this page, and the caps in lesson 5 are described as settings and habits you can apply in any agent product.
Is a $20 subscription or the API cheaper for agents?
For a few interactive agent tasks a day on a frontier model, the $20 seat is about the same price as the API and includes the tools, so it wins. For repeated, scheduled or high-volume loops, the API on a cheap model wins by a wide margin, provided you set caps first.