Summary: OpenAI launched a Data agent inside ChatGPT Work on September 10, 2026 to answer natural-language questions using approved enterprise data sources. It connects to warehouses such as Redshift, BigQuery, Snowflake, Databricks, and MongoDB while respecting existing semantic layers and permissions. The article argues the semantic layer, metric ownership, and governance decide whether the agent produces trustworthy answers.
Most business questions already have an answer stored somewhere in a warehouse. What slows things down is the queue: a request gets filed, an analyst picks it up later, and the report that comes back reflects last week’s figures. On September 10, 2026, OpenAI launched a Data agent inside ChatGPT Work meant to clear that queue — people ask in plain language and get the analysis back in the same conversation.
What the agent actually connects to
The Data agent hooks into approved sources such as Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, and Snowflake, and it can pull files from Google Drive and SharePoint into an analysis. It also reads your organization’s business terms, metric definitions, custom calculations, and data relationships, drawing that context from semantic layers and trusted sources like Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon, and BI dashboards.
That second capability is where things get interesting. The agent is not guessing at what “active customer” means; it is reading the definition your team already settled on. Administrators still decide which connections exist and which roles may use them, and queries enforce the table, row, and column restrictions that already apply to the connected account.
From question to dashboard to action
You can ask follow-ups, inspect the evidence behind a finding, and turn the result into an interactive dashboard with built-in visualizations that teammates can edit, share, and refresh. Brand guidelines can be applied to outputs. The agent can also build and interact with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot, so the work can land in the BI tool a team already uses. It can suggest next steps, identify who needs to be involved, and share findings through Slack or email, carrying out actions you approve through connected tools.
According to OpenAI, data agents in ChatGPT Work are used to analyze company data by nearly its entire product team and by more than two-thirds of its GTM organization. The company’s own data team made that possible by establishing shared business definitions, defining access rules, and putting safeguards around sensitive data. Organizations in the Alpha program using the agent include NTT Data, Thermo Fisher, ServiceTitan, and Zipline.
The part that is actually your job
Peel back the interface and the dependency underneath is distinctly unglamorous: the agent can only be as good as the definitions it is able to read. When your metrics live in three conflicting spreadsheets and a dashboard nobody trusts, natural-language access simply makes the disagreement travel faster. When your semantic layer is clean, the agent turns into a new front end onto work you already did.
This is the same lesson that shows up whenever an agent is handed a whole system: the integration surface, not the model, decides whether the result is trustworthy. That pattern is worth revisiting in what it takes to hand an agent the whole system, because the Data agent inherits exactly those constraints.
For builders, the practical move is to treat the semantic layer as a product with users rather than as plumbing. Assign owners to metric definitions. Settle which roles get which connections before anyone has to ask. Run the agent against questions whose right answers you already know, so that a confidently wrong number stands out from a correct one. The supplied material does not specify pricing, rollout limits, or how conflicts between two definitions are resolved, so those are open questions rather than settled ones.
The queue does not disappear on its own. It moves from “waiting for an analyst” to “waiting for someone to fix the definitions.”