Agent Frameworks, Runtimes, and Harnesses- oh my!

The post defines and differentiates between agent frameworks, runtimes, and harnesses for building applications with large language models. Agent frameworks like LangChain provide abstractions and a standard structure, making it easier to get started. Agent runtimes such as LangGraph are lower-level and focus on production concerns like durable execution, streaming, and human-in-the-loop support. Agent harnesses, exemplified by DeepAgents, are higher-level, "batteries-included" systems that build on frameworks by adding default prompts and opinionated tool handling. The author notes these are evolving concepts with blurry lines between them.