From connected data to a dashboard and action in one conversation
OpenAI introduced the Data agent for ChatGPT Work on September 10. It connects administrator-approved sources including Redshift, BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, plus documents in Google Drive and SharePoint, then turns natural-language questions into analysis and shareable dashboards.
Its distinction from plain SQL generation is the use of company metric definitions and relationships from semantic layers, dbt, GitHub and trusted BI assets. It can interact with Tableau, Power BI, Sigma and ThoughtSpot and, after approval, send follow-up work through Slack, email or connected tools. OpenAI says the connected account's table, row and column permissions remain enforced.
Conversational access does not repair weak data
Efficiency claims in the announcement come from OpenAI's own use and alpha customers rather than controlled independent comparisons. Conflicting definitions, stale pipelines and missing records can still produce a polished but wrong dashboard. Administrators must separately configure available connections and write actions.
Teams should start with read-only investigation and preserve tables, periods, filters and formulas so analysts can reproduce each result. Write access should follow role-based allowlists, confirmation, audit logs and tested rollback.