Context Engineering for Agents - Lance Martin, LangChain

In this episode of the Latent Space podcast, Alessio and Sweeks interview Lance Martin about Context Engineering. They discuss the definition of context engineering, its relation to prompt engineering, and the challenges of managing context in agents due to the flow of information from tool calls. Lance defines five categories of context engineering: offloading, reducing context, retrieval, context isolation, and caching, and shares his experience building Open Deep Research. The conversation touches on the Bitter Lesson, the importance of removing structure in AI engineering, and the role of frameworks and abstractions in building AI applications.