Most business questions with a clear data answer still take days to get answered. Someone submits a request, waits for an analyst, gets a static report, and by then the decision has already been made. OpenAI is betting that removing that bottleneck is a strong enough use case to embed AI directly into the enterprise data stack. On September 10, the company announced the Data agent for ChatGPT Work, a purpose-built agent that connects to company databases, investigates changes in metrics, and produces shareable dashboards, all driven by plain-language conversation.
What it actually connects to
The integration list is the first thing worth paying attention to. The Data agent supports direct connections to Amazon Redshift, Google BigQuery, Snowflake, Databricks, ClickHouse, MongoDB, and more. It also pulls files and documents from Google Drive and SharePoint. That covers most of the serious enterprise data infrastructure in use today.
But raw database access alone is not enough. The agent also reads semantic layers and business context from sources like dbt, Databricks Genie Ontology, Snowflake Horizon, and existing BI dashboards. This matters because the same number means different things in different companies. By ingesting your organization’s metric definitions and data relationships, the agent is supposed to give answers that reflect how your business actually defines its terms, not just what the SQL returns.
Permissions carry over from the connected accounts. Table, row, and column restrictions remain in place, and enterprise administrators control which data connections are available and to which roles. That answers one of the obvious objections before it gets raised.
What you can do with it
The core workflow is conversational. You ask a question, the agent investigates, and you follow up in the same thread. From there, you can turn the analysis into an interactive dashboard with built-in visualizations, which your team can edit, share, and refresh. You can also point it at your brand guidelines to get outputs that match your company’s design standards.
For teams already using dedicated BI tools, the agent also integrates with Tableau, Microsoft Power BI, Sigma, Omni, Oracle BI, and ThoughtSpot. Tableau’s Chief Product Officer noted that the connection brings ChatGPT Work into the same trusted data model teams already rely on, and lets users publish new views directly from a conversation. That’s a meaningful claim if it holds in practice.
Beyond dashboards, the agent can recommend next steps, identify who needs to be involved, and distribute findings through Slack or email. Actions go through approval before the agent carries them out, which is the right call for anything touching live business data.
Who is already using it
OpenAI says nearly all of its own product team and more than two-thirds of its go-to-market organization use the Data agent internally. That kind of internal adoption number is worth something, though it also reflects the obvious home-field advantage of building on your own tools.
In the Alpha program, NTT DATA, Thermo Fisher, and ServicePiston have used it to analyze sales trends, flag reporting errors, and make staffing decisions. NTT DATA’s head of AI specifically called out licensing costs and technical expertise as barriers the agent helped lower for non-engineering staff. That’s the real target user here: the sales manager or finance lead who has data access but no SQL skills and no analyst on call.
How to get it and why it matters
The Data agent is listed as ‘Data’ in the ChatGPT Work Plugins directory. Administrators can install it org-wide through Workspace settings, configure the relevant data-source plugins, and control access by role.
Compared to alternatives like Glean, Seek AI, or the analytics features inside Microsoft Copilot for 365, OpenAI’s version has a wider integration surface and the brand recognition to get past IT procurement faster. The real competition, though, is the existing BI stack. Tools like Tableau and Power BI have their own AI features now, and Databricks has been pushing its Genie product hard for exactly this use case. OpenAI is not replacing those tools, it’s positioning the Data agent as the conversational front end on top of them.
So the question is whether users will actually trust the answers enough to act on them without a human analyst in the loop. That’s a cultural shift as much as a technical one. But if the semantic layer integrations work as described, this is a serious attempt at making self-serve data analysis something more than a demo.
