No drift allowed: LangChain's context playbook with Hex
LangChain's analytics engineering team built a layered context system for their Hex agent, and self-serve questions went from occasional to routine.
Overview
LangChain integrated dbt, semantic models, and guides into a unified context framework, turning its Hex agent into an organizational reference source. Covers managing data context without fielding every inquiry, trust via endorsements, and Context Studio feedback loops to close context gaps.
What you'll take away
- 01
An agent is only as smart as the context it's given: LangChain layers dbt project docs, workspace guides, a semantic layer, and endorsed "source of truth" models inside its analytics agent.
- 02
Column descriptions had to change for an AI audience: a vague "status of the account" became an explicit definition of what active, churned, and prospect actually mean.
- 03
The payoff was a 40x increase in self-serve analytics questions, letting the data team shift from answering every question to owning the context that lets other teams self-serve directly.
Session speakers
Emily Hawkins
LangChainLogan Cochran
LangChain