Lyft ML Engineer Akshay Sharma on Customer Support Agent Architecture
In this video, Lyft Machine Learning Engineer Akshay Sharma explains how the company rebuilt its customer support agent system using LangGraph and LangSmith. He outlines the shift from brittle single-call agents to a meta-agent architecture that allows product managers and operations staff to deploy new agents via configuration rather than coding, noting improvements in development speed and resolution rates.
ImportanceLocalEvidenceE2 unreplicatedWrite-upQuick
This episode, published by the official LangChain channel on September 15, 2026, features Akshay Sharma, a Machine Learning Engineer at Lyft. Sharma provides a detailed breakdown of how Lyft reconstructed its customer support agent technology stack, focusing on the architectural changes brought about by integrating LangGraph and LangSmith.
According to the video description, the Lyft team moved away from previous brittle, single-call agent models to adopt a 'meta-agent' architecture. A key aspect of this transition was empowering non-engineering roles: Product Managers (PMs) and operations personnel can now ship new agents through prompt and configuration changes without writing new code.
Sharma also highlights specific business metrics resulting from this overhaul, stating that agent build time dropped from six months to one or two weeks, and the resolution rate increased by 16%. The video chapters cover the scale of Lyft's support challenges, resolving observability blind spots, and applying LangSmith tracing and evaluation mechanisms.