The article explores the concept of LLM (Large Language Model) agents, emphasizing their simplicity and potential. It compares understanding concepts like AWS S3 to learning to ride a bike, highlighting that some technological ideas require hands-on experience to truly grasp. The author argues that regardless of personal opinions on LLMs, engaging with them is essential for forming informed stances. Creating an LLM agent is presented as an accessible programming task. The author shares a straightforward code example using the OpenAI API to illustrate how to build a basic agent that simulates conversation. Key points include the management of context, where the agent recalls previous interactions, creating the illusion of a continuous dialogue. The piece also introduces the idea of integrating tools, such as a ping function, to enhance the agent's capabilities, demonstrating how users can expand its functionality with minimal effort. The author critiques the focus on complex agents like Claude Code and Cursor, arguing that they often obscure the fundamental concept of LLM agents. He encourages readers to build their own, emphasizing that doing so allows for greater control and understanding of the technology. The article stresses that MCP (Model-Controller-Plugin) isn't a fundamental technology but rather an interface for integrating tools into existing code. This suggests that aspiring programmers should focus on developing their own agents rather than relying on pre-built solutions.