OpenAI has made the Data agent available in ChatGPT Work, enabling it to analyze company data and generate answers, interactive dashboards, and action recommendations. Users can initiate analyses in natural language without writing queries and direct the results within the same conversation. The tool works with Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake, as well as files and documents in Google Drive and SharePoint.
The Data agent interprets an organization’s business terms, metric definitions, custom calculations, and data relationships using context from trusted sources such as Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon, and BI dashboards. Administrators determine the available connections and roles; queries comply with existing account permissions, including table-, row-, and column-level restrictions. Users can review the evidence behind findings, ask follow-up questions, and turn analyses into dashboards that can be edited and shared by teams.
The tool can also create dashboards in and interact with Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot. It can additionally recommend next steps and who should be involved in the process; it can execute approved actions through connected tools and share the results via Slack or email. Almost the entire product team and more than two-thirds of the GTM organization at OpenAI use data agents.
Alpha program organizations, including NTT Data, Thermo Fisher, and ServicePiston, use the tool to review sales and spending, identify reporting errors, and evaluate opportunities. Data can be installed from the Plugins directory; administrators can manage access and data sources through Workspace settings > Plugins.
Why it matters
This feature offers a way to bring together information from different enterprise sources in the same workspace, going beyond the reports and dashboards prepared by data teams. Natural-language support allows employees without query-writing expertise to review areas such as sales, spending, and reporting, while existing table, row, and column permissions define the boundaries of access. Manager oversight, the ability to review evidence, and the creation of editable dashboards within teams make the process of sharing and reviewing results part of the product. The ability to execute approved actions through connected tools also ties the transition between analysis and operations into the same workflow. It remains unclear to what extent business terms and metric definitions across different sources will produce consistent results, and how use will take shape at organizations outside the alpha program.