On September 10, 2026, OpenAI introduced the Data agent in ChatGPT Work in its official announcement, “Now everyone can put data to work.” It can connect to company-approved data sources, investigate metric changes, build editable interactive dashboards, and—after user approval—turn findings into actions through email, Slack, or other connected tools.
The update changes the entry point for business analytics. People can ask a question, inspect the evidence, refine the analysis, and adjust a dashboard in one conversation instead of waiting for an analyst to write a query or prepare a fixed report.
Data connections go beyond file uploads
OpenAI lists Amazon Redshift, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake, and Datadog among the supported sources. Files from Google Drive and SharePoint can also be included. The Data agent can use an organization’s metric definitions, custom calculations, and data relationships, with semantic context from sources such as dbt, Databricks Genie Ontology, Snowflake Horizon, GitHub, and existing BI dashboards.
That makes it more than a tool for inspecting an isolated CSV. For questions about slowing sales, rising costs, or renewal risk, it can follow the relevant data relationships and continue the investigation instead of stopping at a static chart.
Permissions determine whether it is production-ready
Connecting the Data agent does not grant unrestricted database access. Administrators choose which connections and roles are available, while queries continue to enforce the connected account’s table-, row-, and column-level restrictions. Without consistent metric definitions, an agent may only produce conflicting answers faster.
Before rollout, teams should define data owners, metric logic, sensitive fields, and publishing permissions. Finance, HR, and customer data should begin with least-privilege access and mandatory human review.
BI tools gain a conversational front end
OpenAI also names Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot. The Data agent can create or work with dashboards through natural language, while teams can still edit, share, and refresh the results. Existing semantic layers, permissions, and audit controls therefore become more important, not less.
Sales, operations, product, and finance teams with reliable warehouses and shared metrics are likely to benefit first. Organizations with messy data or unclear access rules should fix those foundations before asking an agent to recommend actions.
