newsfilter
← Interviews
Sequoia Capital40 min

Context Engineering Our Way to Long-Horizon Agents: LangChain’s Harrison Chase

Harrison Chase, Sonya Huang, Pat Grady

Interview, Podcast

The shift from scaffolding to agent harnesses has enabled long-horizon LLM agents to function by relying on context engineering and file system access rather than custom cognitive architectures. Current applications, including coding and incident response, operate under a "first draft paradigm" where AI generates outputs for human review rather than aiming for fully autonomous deployment. To address non-deterministic behavior, developers now utilize trace analysis, online testing, and recursive self-improvement loops that allow agents to refine their own instructions based on historical execution data.