OpenAI introduced a data agent for ChatGPT Work on September 10, 2026, which evaluates corporate data from systems like Snowflake, BigQuery, or Databricks directly in the chat. Employees ask questions in natural language and receive ready-made dashboards without having to write database queries themselves. According to its own statements, OpenAI is already using the tool internally in almost all product teams.
Data agent connects seven analytics platforms with chat
The new data agent can be activated as a plugin via the workspace settings in ChatGPT Work; afterwards, employees can call it in the chat window with the command @data. Administrators determine, according to OpenAI’s announcement, which data sources are connected and which roles are allowed to access them. Supported systems include Amazon Redshift, Google BigQuery, Snowflake, Databricks, MongoDB, ClickHouse, and the monitoring platform Datadog, as well as files from Google Drive and SharePoint. Queries respect existing rights at the table, row, and column level, so employees only see data for which they are already authorized. OpenAI does not specify a separate price for the plugin; usage consumes regular credits from the respective work plan.
Results can be immediately transferred to existing analysis tools: The agent builds interactive dashboards that can be reused in Power BI, Tableau, Sigma, Oracle BI, Omni, or ThoughtSpot. An ontology layer from systems like dbt or the Databricks Genie Ontology transfers company-specific metric definitions so that the agent correctly interprets terms like “revenue” or “active customers.” Evaluations can be shared via Slack or email, and follow-up questions remain part of the same conversation.
First customers report faster evaluations
The IT service provider NTT Data is among the pilot customers from OpenAI’s alpha program. So far, licensing costs and a lack of technical knowledge have hindered the expansion of their own dashboards; with the data agent, sales and administrative employees without programming skills are now building their own evaluations, reports NTT Data in OpenAI’s customer reference. The pharmaceutical and laboratory equipment supplier Thermo Fisher is testing the agent with data from its own system landscape to evaluate supplier data, recognize opportunities in the supplier network earlier, and prepare teams more effectively for tenders. Other alpha testers include, according to OpenAI, the logistics company Zipline and the software provider ServiceTitan, which have been participating in the pre-test phase since spring 2026; the company has not yet disclosed public details about their use cases.
Internally, OpenAI claims to use the data agent in nearly all product teams; more than two-thirds of the sales organization are said to be working with it – a number that has not been independently verified. OpenAI has not yet disclosed a user number outside its own company, such as among paying work customers.
Data agent expands OpenAI’s push into enterprise software
The data agent builds on ChatGPT Work, which OpenAI launched in July 2026, and complements the variant ChatGPT for Financial Services introduced a few days earlier with a cross-industry analysis tool for all customers with work access. Microsoft links similar functions to corporate databases with Copilot in Power BI and Fabric, while Google is pursuing a comparable approach with Gemini in conjunction with Looker. Both providers also advertise that specialist departments will be able to evaluate metrics in the future without their own query language.
For Germany or the EU, OpenAI does not specify any special restrictions or a separate launch date in the announcement; ChatGPT Work is already running regularly in European accounts, and a regional restriction for the data agent is not yet known. Larger corporations with their own data infrastructure are likely to be the first European users for the time being, while the competition over whose assistant evaluates corporate data most reliably and securely will intensify in autumn 2026 – especially since all three providers are addressing the same customer base of corporate IT departments.
It remains open how companies will reconcile extensive data access with their own security policies when a chat assistant combines metrics from multiple core systems simultaneously. OpenAI has not yet specified a date for when smaller business customers without their own IT department will gain access to the data agent – for them, setting up roles and permissions is likely to remain a greater hurdle than for large customers like NTT Data or Thermo Fisher.
