How Podium Scaled their Agents with LangSmith and LangGraph

Julia Schottenstein, COO at LangChain, sat down with Walker Ward, Principal Software Engineer at Podium, to unpack how Podium built AI agents that responds to inbound leads for car dealerships, home services companies, and elective medical practices. Walker walks through Podium's path from brittle early prompt engineering and manual "golden eval" reviews in a Google Doc to a full agent engineering practice built on LangSmith, plus why Podium ultimately moved off their own hand-rolled runtime and onto LangGraph deployments.

Chapters:
0:00 Customers brought their agent cookies
0:18 Meet Walker, Podium's AI lead
0:35 Podium's business before AI
0:42 Going all in on a hundred million in AI revenue
1:22 The Y Combinator connection that led to early GPT-3 access
1:48 Why speed to lead changes everything
4:34 Manually reviewing 50 golden evals in a Google Doc
7:11 The moment they knew Jerry was working
8:53 Scaling one agent pattern into a half dozen agents
10:27 Why they sought out LangSmith in 2023
12:18 Ditching their hand-rolled runtime for LangGraph deployments
13:11 Inside Podium's agent engineering loop
18:40 What's next for Podium
20:23 Walker's advice for builders starting today

Resources:
→ LangGraph: https://www.langchain.com/langgraph
→ LangSmith: https://www.langchain.com/langsmith
→ LangChain Academy: https://academy.langchain.com