Miro lifts AI agent accuracy from 40% to 90% with DataHub context

Miro connected Claude Code directly to Snowflake across 20,000+ datasets and measured under 40% text-to-SQL accuracy on a 900-question benchmark. Adding DataHub Cloud as a governed context layer and narrowing scope to documented data products raised accuracy to around 90%.

• Direct Claude Code-to-Snowflake link scored under 40% accuracy
• Wrong-table routing occurred in more than 30% of cases
• DataHub context layer lifted accuracy from ~50% to ~90%
• The 900-question benchmark was updated with model and pipeline changes