Coding agents can read files, run commands, call tools, and modify entire codebases. But what is actually happening under the hood?

This talk opens the black box and breaks a coding agent down into its core mechanics. Rather than focusing on prompting techniques, we build mechanical sympathy for agentic development by looking at how the pieces actually fit together.

We will inspect how an agent communicates with the language model, how a text-only model gains access to source code through tools, how MCP extends those tools, and what is actually sent with each model request. Finally, we will put the pieces together and build a basic coding agent in C#.

We will cover:

- how coding agents communicate with language models
- how tools let a text-only model read, modify, and execute code
- how MCP tools are integrated into the same tool model
- how system prompts, tools, and conversation history shape each request
- how the agent loop manages context and tool calls
- how to build a basic coding agent in C#

Consists of presentations and hands-on live demonstrations using this open-source project of mine: https://github.com/tndata/CodingAgentExplorer