OpenAI has released a Data agent for ChatGPT Work that can connect to approved company data sources, investigate business questions, and build interactive dashboards in a conversation. The launch moves ChatGPT Work closer to a governed analytics front end, but the announcement supplies no accuracy benchmark, pricing detail, or independent evaluation.
What changed
The new Data plugin can query approved sources including Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake. It can also incorporate files and documents from Google Drive and SharePoint, and use organization-specific metric definitions, calculations, data relationships, semantic layers, and trusted business context.
OpenAI says the agent can turn an analysis into a dashboard with built-in visualizations or work with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot. Users can refine the work through follow-up questions, inspect the evidence behind a finding, and ask for recommended next steps. Connected tools can share findings or carry out actions only through the organization’s configured permissions and approval flow.
The practical consequence
For organizations already using ChatGPT Work, the product can put warehouse queries, documents, business definitions, analysis, and presentation in one interface instead of requiring a separate handoff for each step. Administrators choose which connections and roles are available. OpenAI says queries enforce the connected account’s existing table-, row-, and column-level permissions, so the agent does not replace the source system’s access controls.
Deployment is not automatic. A workspace administrator must install or enable the Data plugin, configure the relevant data-source plugins, complete any required account connections, and decide who may use them. Useful answers also depend on the quality of the underlying data, semantic definitions, schemas, and access policy.
Limits and evidence
OpenAI does not identify the model configuration behind the Data agent, disclose query or refresh limits, publish error rates, or state whether the plugin is included in existing ChatGPT Work licensing. The announcement also does not quantify latency, analyst time saved, dashboard correctness, or total operating cost. Customer testimonials and OpenAI’s own adoption figures are first-party or partner reports, not controlled measurements.
Permission enforcement limits which records a query can retrieve; it does not guarantee that the agent interprets a metric correctly, chooses a sound method, or avoids a misleading visualization. High-consequence findings still need human review against source data and established business definitions. DrComps did not connect the product to a live warehouse or independently reproduce OpenAI’s capability claims.