OpenAI has introduced Data agent for ChatGPT Work, a new plugin intended to make company analysis more accessible without bypassing the controls that surround enterprise data. The product connects users to approved sources, helps them investigate a question in conversation and can turn the resulting work into interactive dashboards.
The announcement matters because it puts a familiar analytical workflow inside ChatGPT Work rather than asking every employee to learn query tools or wait for a specialist report. A sales leader can ask why a pipeline slowed, a product team can explore adoption, and an operations team can follow a spending change with additional questions. The aim is not merely a one-shot answer: the agent can refine the analysis as the user develops a line of inquiry.
Data access remains an administrative decision
OpenAI says the agent connects to company data and context that an organisation has approved. Supported sources include Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, alongside documents from Google Drive and SharePoint. It can also work with semantic layers, business definitions, custom calculations and data relationships from systems such as dbt, Databricks Genie Ontology, GitHub, Snowflake Horizon and BI dashboards.
That framing is important. The agent is designed to use the permissions already associated with the connected account, including table, row and column restrictions. Enterprise administrators choose which connections are available and which roles can use them. For companies weighing an internal rollout, the product is therefore closer to a controlled interface over selected data than a blanket connector for all corporate information.
OpenAI also positions the capability as a way to make governed context reusable. A metric name, a calculation rule or a relationship between data sets can be difficult to communicate consistently across an organisation. By drawing on trusted definitions, the agent is meant to reduce the risk that a fast natural-language answer uses an unfamiliar or misleading interpretation of a business measure.
From question to a shareable view
Once a user has explored a result, Data agent can create interactive dashboards with built-in visualisations. Teams can edit, share and refresh those dashboards, while the conversation remains available for follow-up investigation and evidence review. OpenAI says the agent can also build and interact with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot.
This makes the launch more than an integration catalogue. It brings analysis, an explainable trail of questions and a presentation layer into one workflow. The practical test for teams will be whether that hand-off remains reliable when a dashboard is refreshed, shared or reviewed by someone who did not start the conversation.
OpenAI says users can ask the agent to recommend next steps, identify who should be involved and, after approval, share findings through connected tools. Those capabilities create clear governance questions for administrators: which downstream tools are enabled, what review is needed before an action, and how are permissions kept aligned as staff and data sources change.
Availability and rollout considerations
Data agent appears in the ChatGPT Work Plugins directory as Data. Administrators can install it for teams through Workspace settings and configure the relevant source plugins. Organisations should start by identifying a small set of high-value, well-understood data sources, confirming that their semantic definitions are current and testing the agent against existing reports.
The launch reflects a broader shift in enterprise AI from drafting text around a business to working with the systems that describe it. Its value will depend on data quality, permissions and review practices as much as the model's ability to answer a question. Teams that establish those foundations can use Data agent to shorten the gap between a question and an evidence-backed operational view.
For buyers, the key distinction is that the new product combines connectors, governance and dashboard creation in ChatGPT Work. Rather than replacing every analytics tool, it offers a conversational layer that can move between analysis and the tools a company already uses. That makes careful pilot design, permission testing and clear ownership of business definitions central to a successful deployment.
How organisations can prepare
Successful deployments will start with data stewardship rather than a broad invitation to connect every system. Administrators can map the sources that have clear owners, stable permissions and agreed definitions, then test representative questions against established reports. Any difference should be traceable and resolved before the agent becomes a default route for decision-making.
Teams should also set expectations for dashboard review. An interactive view can accelerate discovery, but it does not replace the judgement needed to interpret an unexpected change or approve an action. Combining the new conversational experience with documented data owners and clear escalation paths will help organisations use it as a reliable analytical tool.