NVIDIA AI Podcast

Harrison Chase of LangChain on Deep Agents, LangSmith, and Earning Trust | NVIDIA AI Podcast Ep. 297

25 min

Harrison Chase, CEO of LangChain, discusses the evolution and future of AI agents, emphasizing deep agents as a general-purpose harness for LLMs. He highlights LangSmith for observability and evaluation, crucial for building trust and enabling evaluation-driven development in enterprises. The conversation also covers the strategic integration of frontier and open models, with a focus on cost-efficiency and the emerging need for proactive, always-on agents with memory and distinct identities.

Summarized by Podsumo

🎙️ Ask about this episode

✨ Key Takeaways

1

Deep Agents as a General-Purpose Harness: LangChain's deep agents provide a model-agnostic, open-source framework that gives LLMs more autonomy and an environment for interaction, exemplified by OpenClaw.
2

LangSmith for Enterprise Trust and Evaluation: LangSmith is presented as the essential platform for the agent development lifecycle, offering observability and evaluation-driven development (eVals) to ensure agents perform as expected, crucial for enterprise adoption.
3

Strategic Model Mixing and Open Source Importance: The discussion emphasizes combining frontier and open models (e.g., orchestrator with frontier, sub-agents with open-source) for optimal cost and performance, noting that open models are increasingly capable of driving complex agent harnesses, especially for always-on scenarios.
4

Future Agent Capabilities: Key future trends include asynchronous subagents for long-running tasks, proactive event-driven agents (e.g., email drafting), agent memory for learning from interactions, and the concept of agent identity with dedicated credentials.
5

OpenClaw's Impact: OpenClaw significantly influenced the industry by setting a new "North Star" for agent capabilities, prompting enterprises to develop a "Claw strategy" while prioritizing security and control.

💬 Notable Quotes

"And so I think like these always on asynchronous event driven agents that will be a really big productivity unlock and especially enterprises There's so many events that are just triggering triggering triggering and so if you can have agents listening to those and firing off I think that will be a massive game"

— Harrison Chase
"I think a failure mode for enterprises is you have some idea of an agent, you take three months to craft a bunch of examples. You take another three months to build the agent. You take another three months to get humans to look at everything. But the space has just moved so fast. The whole idea you came up with, there's probably a better way to do it at that point."

— Harrison Chase
"So at the risk of signing a little broad, like it needs to be intelligent, it needs to be good. Another thing that is maybe underappreciated is it probably needs to be good at coding."

— Harrison Chase