Ask a business question inside ChatGPT Work, and a new Data agent will go looking for the answer across company systems connected beforehand. OpenAI introduced it on September 10.
Support for the feature is unusually wide for a launch. Backers announcing support include AWS, ClickHouse, Databricks, Snowflake, MongoDB and G2; dashboards can be read or built in Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot. Microsoft Fabric is involved as well, which means a Power BI dashboard can be requested in plain language instead of assembled by hand.
Output arrives as something a colleague can edit, share and refresh, with brand guidelines available so charts match the organization’s look. OpenAI says cited sources back the numbers, reviews sit in the path, and role-based access control decides which data each employee can reach.
What people ask tends to be familiar. Why sales slowed. Where spending is climbing. Which large accounts are drifting toward renewal risk. Those answers have usually travelled through a queue at an analytics team first, and OpenAI’s claim is that the queue, not the math, was the bottleneck.
Its own adoption figures make that argument. Adoption inside OpenAI is broad: nearly the whole product team relies on it, and over two-thirds of the go-to-market org works with data agents in ChatGPT Work. An in-house data team laid the groundwork: shared business definitions, access rules and safeguards for sensitive data.
Beyond analysis and dashboards, OpenAI lists further capabilities. Beyond answers, it proposes next steps, names who to loop in, ships results to Slack or email and runs approved actions via connected tools.
