How Does a Coding Agent Work?
summary
Explains a working coding agent built as a while loop that sends a system prompt (~25,000 chars), tool descriptions (about 90,000 characters of tool descriptions) and conversation history to a Claude model via the Anthropic SDK, using a local proxy to intercept API calls. Demonstrates file tools (read/write/list), MCP access through getmcp.io MCP proxy, a real C# example under <200-line C# agent that writes and runs a FizzBuzz, and extensibility with pre- and post-tool hooks.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:56Mechanical sympathy for agents
- 3:40Agent-model communication
- 7:12Accessing source code
- 11:50MCP in agent APIs
- 16:47Inspecting MCP traffic
- 18:52Coding agent request breakdown
- 20:20Build your own agent
- 24:15Skills extend agents
- 28:50Conditional tool activation
- 29:11Extending agents with hooks
- 35:21Resources and tools
- 40:26Inefficiency in agent context
- 41:53Q&A
related talks
01h 03mAI Coding Agents and how to code them
01h 28mAI Coding Agents and how to code them
40mAI Coding Agents and how to code them
48mAI Coding Agents and How to Code Them
01h 01mAI Agents In-Depth – Function Calling, MCP and Tool Use Under the Hood
48mAI Coding Agents and how to code them
30mBeyond the Hype: Using AI Coding Agents Today
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips