How LangChain Built an 'Agent-First' Data Stack

Existing data stacks were mainly built around dashboards, reports, and SQL workflows, which limited their usefulness for agents. For agents to go beyond simply generating SQL and deliver trustworthy answers, they need a data layer that includes clear definitions, reliable sources, and business context.

LangChain moved away from a traditional BI-tool-centric approach and shifted its architecture toward a data stack optimized for self-service analytics, shared context, and agent usage. Through this, the company aimed to resolve data team bottlenecks and meet diverse user needs.

The data team chose Hex as its core tool, integrating dashboards, notebooks, and conversational analytics. Users can interact with the agent through a variety of touchpoints, including:

• Hex UI (including Threads and notebooks)
• Slack
• CLI workflows
• MCP (Model Context Protocol)
• LangSmith Fleet

As a result of this shift, the in-house data agent is now handling about 40 times more requests than the existing 3-person data team could handle directly. In addition, over the past 30 days, about one-third of all users have used the agent, recording an average of 23 conversations per user per month, showing a high adoption rate.