The Anatomy of an Agent Harness

A comprehensive breakdown of the 'agent harness' concept — the complete software infrastructure wrapping an LLM that makes autonomous agent behavior possible. Covers 11 production components: orchestration loop, tools, memory, context management, prompt construction, output parsing, state management, error handling, guardrails, verification loops, and subagent orchestration. Draws on implementations from Anthropic (Claude Code), OpenAI Agents SDK, LangGraph, and CrewAI. Also covers a step-by-step walkthrough of a single agent cycle, how major frameworks implement the pattern, and seven key architectural decisions every harness designer must make — including single vs. multi-agent, context window strategy, harness thickness, and tool scoping.