The new agent connects approved sources, semantic layers, and BI tools to investigate data and create dashboards. See governance requirements.
Direct answer
On September 10, 2026, OpenAI presented the ChatGPT Work Data agent, capable of querying approved business sources, interpreting metrics with semantic layers, investigating evidence and creating interactive dashboards. The proposal reduces dependence on manual queries, but does not replace data quality, shared definitions, access control or human review before high-impact decisions or actions.
Semantic layer is worth more than natural language
Asking in Portuguese simplifies the interface, but a reliable answer depends on consistent definitions for revenue, active customers, margin and other metrics. Without this contract, the agent just makes the ambiguity faster.
Permissions must accompany the source
The query must respect the table, row, and column constraints of the connected identity. The correct design avoids copying data to a parallel repository and preserves the trail of who could see each result.
Evidence needs to remain visible
AI-generated dashboards should show origin, period, filters, transformation and limitations. A conclusion without a verification path should not trigger a budget, price, credit, or critical operation.
Action requires an additional border
Recommending a next step is different from taking it. Explicit approval, impact threshold, reversibility, and registration are required when the agent sends messages or changes systems.
Nexus Reading
The gain does not come from replacing BI with conversation, but from connecting natural language to a governed basis. A pilot should start with a few metrics, recurring questions and objective criteria for accuracy and time saved.
FAQ
Does the Data agent eliminate the data team?
No. It expands access to analytics, while experts remain essential for modeling, quality, governance and complex decisions.
Can it access any table?
No. Administrators choose connections and queries must respect the permissions of the connected account.
How to validate a pilot?
Compare responses to baseline reports, measure time to insight, and review errors, sources, and actions before expanding the scope.
Essential guides to delve deeper into the decision
This editorial analysis was produced by Nexus from the official sources below, consulted on September 15, 2026. The text is original and interprets practical implications for companies.
- OpenAI — Now everyone can put data to work: Data agent ad, connectors, semantic layers, controls, and analysis flows.
- OpenAI — Business data privacy, security, and compliance: Privacy commitments, administrative controls, and security features for enterprise data.
Date reported by the main source: September 10, 2026.