Harrison Chase on better models and the agent stack, 20VC podcast
Overview
LangChain's Harrison Chase joined Harry Stebbings on the 20VC podcast to discuss how better models change what developers build and where value sits in the agent stack, according to a LangChain post on X. An excerpt from the discussion says LangChain spends about 30% of engineering salary on coding tools, a share that fell over the last two months while token usage rose.
LangChain's Sydney Runkle, in a separate X post, summarizes the conversation: an agent is a model, context, and harness. She says optimizing context gives the highest return, followed by the harness and then the weights. She also lists agent accuracy, cost, latency, and security as recurring developer concerns, and says memory remains unsolved.
Written by AI from the articles below · updated Oct 19, 3:34 PM ET
Check the sources:
Developments
2 developments
- Oct 19, 3:22 PM ET · 1 articleLangChain's Harrison Chase discusses better models and the agent stack on 20VCLangChain: LangChain's Harrison Chase discusses better models and the agent stack on 20VC
- Oct 19, 3:21 PM ET · 1 articleLangChain's Sydney Runkle recaps agent-building takeaways from a Harry Stebbings interviewLangChain: LangChain's Sydney Runkle recaps agent-building takeaways from a Harry Stebbings interview
Article timeline
The articles in this story. Times are ET.
LangChain@LangChainOfficialLangChain's Harrison Chase discusses better models and the agent stack on 20VCAILangChain's Harrison Chase joined Harry Stebbings on the 20VC podcast to discuss how better models change what developers build and where value sits in the agent stack. The post links to the full episode on YouTube, and an excerpt from the discussion says LangChain spends about 30% of engineering salary on coding tools, a share that fell over the last two months while token usage rose.
LangChain@LangChainOfficialLangChain's Sydney Runkle recaps agent-building takeaways from a Harry Stebbings interviewAILangChain's Sydney Runkle recaps a conversation between Harrison Chase and Harry Stebbings on the 20VC podcast, arguing that an agent is a model, context, and harness. She says optimizing context comes first for the highest return, followed by the harness and then the weights. She also says agent accuracy, cost, latency, and security are recurring developer concerns, and that memory remains unsolved.
Heat Trend
Not enough continuous observations to show a trend yet.