Bio
Harrison Chase is the co-founder and CEO of LangChain and one of the operators most directly responsible for shaping how teams instrument, observe, and evaluate LLM-based agents in production. His writing focuses on the seams between agent runtime and agent learning, what gets logged, how feedback gets attached, how teams actually compound improvements over time rather than shipping a brittle prompt and walking away. The recurring through-line is that observability infrastructure that captures traces without binding them to feedback is logging dressed up as learning.
Operating themes
- Operating thesis: trace + feedback is the minimum learning unit. Either alone is not.
- Agent observability, trace capture, span correlation, feedback binding.
- LLM evals, promotion-as-PR, evaluator agents, drift detection.
- Frameworks-as-experience, design choices in LangChain / LangGraph reflect what teams actually fail at.
Cards
- A trace alone teaches nothing; learning requires feedback attached to the trace, A trace alone teaches nothing; learning requires feedback attached to the trace [Tier A]
- Decision models outperform LLM judges in agent evals at 1,000 to 80,000 times lower cost because they output typed verdicts directly rather than generating reasoning text, Decision models outperform LLM judges in agent evals at 1,000 to 80,000 times lower cost [Tier B]
Sources captured
- 2026-05-05, Agent observability needs feedback to power learning (LangChain blog)
- 2026-09-20, Jev-as-a-Judge for Agent Evals (LangChain blog, with Daniel Shea and Seán Roche)