Build a Coding Agent to Understand Coding Agents

Coding agents can search a repository, modify files, run commands, inspect failures, and repeat until a task is complete. But what is actually happening between your prompt and the resulting code?

In this talk, we will build a small coding agent from the ground up and use it to examine the core pieces behind tools such as Claude Code or your coding agent tool of choice. We will look at how an agent constructs context, exposes tools to a model, runs an agent loop, handles tool results, and decides what information should remain in the conversation.

The goal is not to replace mature coding agents. It is to develop a practical mental model for evaluating them. By understanding their underlying architecture, Android developers can better diagnose poor results, configure their workflows, and decide whether a features of their favorite coding agents is essential, useful, or unnecessary complexity.