OpenAI’s Data Agent Brings Governed AI Analytics to ChatGPT Work

OpenAI has introduced the Data agent for ChatGPT Work, a plugin intended to let employees investigate business questions in natural language, turn findings into interactive dashboards, and act on approved next steps. OpenAI’s launch post is dated September 10, 2026; its data-focused ChatGPT Work solution page followed on September 11. 86

The important distinction is governance: the agent is designed to operate on administrator-approved connections and to preserve the access restrictions already attached to the connected account. 8

What the Data agent does

The Data agent connects company data to a conversational workflow. A user can ask a business question, investigate a change, refine the analysis with follow-up prompts, and create an interactive dashboard or report without writing queries or learning a separate analytics interface. 18

OpenAI says the agent can:

• investigate changes in business data;
• surface evidence and allow users to refine the analysis in the same conversation;
• create editable, shareable, and refreshable interactive dashboards with built-in visualizations; and
• recommend relevant next steps and stakeholders, then share findings or carry out actions through connected tools when the user approves. 8

This is a workflow that spans analysis and follow-through rather than stopping at a static answer.

Approved sources, files, and organizational context

OpenAI lists Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake among the approved data sources the agent can connect to. It can also incorporate files and documents from Google Drive and SharePoint into an analysis. 8

Raw access is not the entire model. OpenAI says the agent can use an organization’s established business terms, metric definitions, calculations, and data relationships. That context can come from semantic layers and trusted sources including Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon, and BI dashboards. 8

In practice, this means a prompt such as “Why did retention fall?” is meant to be interpreted against the company’s own definition of retention and its approved metric logic, rather than relying solely on generic language.

Permissions remain under administrator and source controls

The Data agent does not provide a blanket new data entitlement. Workspace administrators decide which connections are available and who can use them. OpenAI says queries retain the connected account’s existing table-, row-, and column-level restrictions. 8

For Enterprise and Edu workspaces, administrators can select Data in Workspace settings → Plugins to review its installation policy and manage access by role or group. Individual data-source plugins and associated apps may need their own configuration and authorization. 4

A practical setup path is:

1. Open Workspace settings → Plugins and make Data available under the workspace’s installation policy.
2. Enable and configure the required data-source plugins or apps.
3. Assign access to the appropriate roles or groups and complete any required account authorization.
4. Have eligible members begin a conversation with @Data. 24

Dashboard creation and BI integrations

OpenAI positions the Data agent as a layer that can create and explore dashboards in existing analytics products. It says the agent can create and interact with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot. 8

It can also generate interactive dashboards directly as part of the ChatGPT Work analysis flow. The stated goal is to let a user move from a question to a shareable visual explanation, then continue to interrogate the result conversationally. 8

For organizations already invested in BI tooling, that integration model matters: Data is presented as an additional way to query, create, and interact with established systems rather than a new warehouse or semantic layer.

From insight to a user-approved action

The agent can recommend follow-up actions and relevant people based on an analysis. Through connected tools, it can share findings or perform actions using Slack or email—but OpenAI says these actions require user approval. 8

That approval step is consequential. The product’s scope is not merely natural-language querying; it is also action orchestration with a person remaining responsible for the final decision to execute.

Who is using it

OpenAI reports that nearly all of its product organization and more than two-thirds of its go-to-market organization use data agents to analyze company data. The company attributes that internal adoption to shared business definitions, access rules, and safeguards for sensitive data. 8

OpenAI also named NTT Data, Thermo Fisher, and ServicePiston as alpha users. According to OpenAI, these organizations have used the capability for work including sales and spending analysis, reporting-error detection, opportunity decisions, and staffing choices. 8

These are vendor-reported adoption examples, not independently audited performance results. OpenAI’s broader ChatGPT Work page separately says Work is deployed at thousands of businesses and names Virgin Atlantic, BNY, and Estée Lauder, but that does not establish that those companies use the Data agent specifically. 6

What it means for BI and analytics vendors

The launch is best understood as a challenge to the analytics interface and workflow, not as an announced replacement for data warehouses or BI platforms.

The Data agent still depends on approved data systems, organizational definitions, semantic context, existing permissions, and integrations with established BI products. 8 Its potential pressure point is the work traditionally required to get from a business question to an analysis, dashboard, stakeholder handoff, or approved follow-up action.

That could make self-service analysis more accessible to non-specialists. But the quality and trustworthiness of the outcome still depend on the underlying data sources, semantic definitions, access policies, and the organization’s review practices. OpenAI has described the product’s capabilities; it has not claimed that it eliminates the need for data platforms, BI tools, or human oversight. 8

The bottom line

The Data agent makes ChatGPT Work a conversational front end for governed business analysis: it connects to approved databases and files, incorporates company-defined metrics and semantic context, respects existing access controls, and can turn analysis into dashboards and approved actions. 8

For data teams, the opportunity is not simply faster dashboard generation. It is a way to expose established data governance and business definitions to more employees through a single, conversational workflow—provided administrators configure the connections and permissions carefully. 48