Everything Gets Rebuilt: The New AI Agent Stack | Harrison Chase, LangChain

Harrison Chase, co-founder and CEO of LangChain, discusses the evolution of AI agents, highlighting the shift from early models to today's more sophisticated systems. He emphasizes the importance of the "harness," which dictates how a model interacts with its environment, including tools for file editing and code execution. Chase differentiates between conversational agents and long-horizon agents, noting the increasing prevalence of coding agents due to their versatility and the models' training on code. The conversation explores key components of modern agent architecture, such as detailed system prompts, planning tools, sub-agents for context isolation, and file systems for managing context windows. Chase also touches on the significance of memory, distinguishing between short-term and long-term memory, particularly procedural memory, which involves instructions and agent configuration.