An AI agent is an LLM in a 20-line loop: where the complexity begins

"Build us an AI agent" now sounds as simple as "build us a button," yet in practice the word "agent" covers three distinct things: a prompt, a chatbot, and an actual agent. That's the argument from the author of a Habr article on the anatomy of AI agents.

How an agent differs from a prompt and a chatbot

The author argues that the confusion around agents comes from using one name for three completely different things: a simple prompt, a conversational chatbot, and a full-fledged agent. The difference is fundamental. An agent doesn't just answer a question; it operates in a loop and decides on its own what to do next.

What is the anatomy of an agent?

As the author describes it, a minimal AI agent is an LLM running in a loop, and you can implement it in literally twenty lines of code. That's where the real complexity begins: turning the minimal loop into a reliable system that doesn't waste budget and actually solves a real problem.

• The basic agent architecture is an LLM in a loop, roughly 20 lines of code
• The author identifies three distinct concepts that get lumped together as "agent": prompt, chatbot, agent
• The article breaks down five ways to accidentally burn budget on an agent
• It offers a simple filter for checking whether a task really needs an agent

What it means

Demand for "AI agents" is outpacing understanding of what they actually are. As long as teams confuse prompts, chatbots, and agents, development budgets keep going to waste. A clear distinction between these concepts, plus a simple "do we need an agent at all" filter, offers a practical way to avoid overpaying for a buzzword.