Glossary
Harness engineering
Designing the tools, memory, permissions, and checks around a model so a job can finish.
Harness engineering is the work of designing the harness: the loop, tools, files, permissions, and checks around a model. The prompt is one piece of that system.
OpenAI’s February 11, 2026 account uses the phrase for a team whose job became designing environments, specifying intent, and building feedback loops so coding agents could ship an internal product. LangChain’s March 10, 2026 explainer uses it for building the system around a model so the model’s output becomes work. Databricks’ June 17, 2026 explainer places it after prompt engineering and context engineering: the artifact is the harness itself.
The longer guide is What is an AI agent harness?.