Agent Harnesses provide the runtime infrastructure that orchestrates autonomous LLM-powered coding agents within a buyer's own codebase and development environment. They manage the agentic loop itself: assembling context from repositories and tools, dispatching commands and file edits, maintaining session state across multi-turn interactions, and enforcing permission boundaries. Delivered as installable or embeddable artifacts distinct from underlying model weights, these harnesses sit between the LLM and the development environment, handling tool integration, context windowing, and execution control. Common deployment patterns include self-hosted runtimes, IDE-embedded loops, and CI/CD-integrated agents. The category emerged as teams moved from single-shot code completion to persistent, multi-step autonomous workflows that require durable state management, environment awareness, and robust safety controls that general-purpose agent frameworks do not provide out of the box.