Operator Insights

Reported by others2

1. 

   Jordan FarrerDirector of Data Science & AnalyticsChime

   Organization-wide. 500+ people asking data questions. Snowflake and dbt.

   Hex, published Aug 4, 2026Read the case study (opens Hex in a new tab)

   Observation

   At Chime, a data domain opens to business users only when Hex's agent passes its eval set.

   Takeaway

   Treat each data domain like a new analyst. Write the questions it must answer, and launch it only when the agent passes them.

   What happened

   Chime organized its data into domains with semantic views and endorsed projects, plus guides synced from GitHub. Senior Data Scientist Connor Macdonald built a three-level eval set. Each domain launched only after Hex's agent passed all three levels. Dori Wilson later moved the evals from an in-house tool to Hex's CLI. More than 500 people now ask questions in Hex or Slack, or through MCP. Domain launch time fell from two and a half weeks to five days.

   • Vendor case study
   • Usage and launch-time figures reported by Hex and Chime

   Deployment details

   Organization-wide. 500+ people asking data questions. Snowflake and dbt.

   • Vendor case study
   • Usage and launch-time figures reported by Hex and Chime

   Team

   Chime data science and analytics

   Workload

   Self-serve business questions by data domain, in Hex, Slack, and MCP clients

   Product version

   Threads, Context Studio, and CLI evals (beta)

   Measurement

   Domain launch time, monthly insights, and Threads conversations without user doubt

1. LangChainJul 27, 2026LangChain blogAt LangChain, a three-person data team moved the whole company onto Hex, and now it curates context while the agent answers requests.

   

   Emily HawkinsHead of Analytics EngineeringLangChain

   Company-wide. Three-person data team. Six-week move off a legacy BI tool.

   LangChain blog, published Jul 27, 2026Read the write-up (opens LangChain blog in a new tab)

   Observation

   At LangChain, a three-person data team moved the whole company onto Hex, and now it curates context while the agent answers requests.

   Takeaway

   Write down the context before expanding access. LangChain credits clean models and written definitions for its agent's reliability. Endorsement stays with the data team, which still validates answers that need review.

   What happened

   LangChain's data team, one person when the work began, left its legacy BI tool for Hex in six weeks because it wanted dashboards, notebooks, and conversational analysis in one workspace. The agent reads dbt definitions, a semantic model, workspace guides synced from GitHub, and endorsements only the data team can grant, and people reach it in Hex, Slack, the CLI, and MCP. In the last 30 days nearly every provisioned user, a third of the company, held about 2,200 agent conversations. The team says the agent handles about 40 times the requests it could field directly.

   • Public engineering write-up
   • Usage figures reported by LangChain

   Deployment details

   Company-wide. Three-person data team. Six-week move off a legacy BI tool.

   • Public engineering write-up
   • Usage figures reported by LangChain

   Team

   LangChain data team of three

   Duration

   Six-week migration; usage measured over 30 days

   Workload

   Company-wide self-serve analysis in Threads, notebooks, Slack, CLI, and MCP

   Product version

   Hex agent with semantic model, workspace guides, endorsements, and Context Studio

   Measurement

   Agent conversations and active users, and request volume against the data team's capacity