How an AI agent harness works (build one in Python)

What an agent harness is and how it turns an LLM into a coding agent: the tool-calling loop, tools, permissions, context compaction and guardrails, built step by step in about 60 lines of Python. Your coding agent is not the model. The model can't open a file or run a single test. It reads text, it writes text, and that's it. 0:00 What an agent harness is 0:39 The loop 1:43 Tools 2:50 Permissions 3:45 The context window 4:55 Guardrails 5:22 Fixing a failing test #aiagents #llm #ai #programming