On September 10, OpenAI added a Data agent to ChatGPT Work, listed in the app directory simply as "Data." Enterprise admins install or enable it from Workspace settings. Employees ask questions in plain language, and the agent connects to the company's already-approved data sources, checks where business metrics have shifted, and produces a shareable, interactive dashboard. The direction of analysis can be refined and revised within the same conversation — no query language to write, no separate analytics tool to learn.
The examples OpenAI cites are everyday business questions: why sales slowed, which spending category is rising, and which issues among the biggest accounts could threaten renewal.
ChatGPT Work is the agent hub OpenAI introduced in July alongside its new desktop app, sitting next to Chat and Codex, positioned for carrying a single project forward across apps and files. Data agent is Work's plugin for data-analysis scenarios. On the same day, OpenAI also released a public beta of its Agents API.
What it connects to
On the data-source side, the announcement lists Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, and files or documents from Google Drive and SharePoint can also be pulled into the same analysis.
On the dashboard side, it can build and edit dashboards directly inside six tools: Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot.
Metric definitions are the biggest headache in this kind of analysis: a metric called "active customers" can be calculated differently by sales and finance. Data agent's approach is to read the definitions a company already has in place; the sources OpenAI's public materials mention include metric definitions written in Databricks, dbt, GitHub and Snowflake Horizon.
On permissions, admins decide which data connections are opened up and which roles can use them; every query runs under the permissions already attached to the connected account, including table-, row- and column-level restrictions.
After an analysis is done, it can also suggest next steps, flag who should be looped in, share the findings via Slack or email, and — once a user approves — carry out actions through the tools it's connected to.
Who's using it first
OpenAI says nearly all of its own product team and more than two-thirds of its sales and marketing staff are already using the Data agent in ChatGPT Work to analyze company data themselves. The announcement names 11 alpha customers, including NTT Data, Thermo Fisher, ServiceTitan, Zipline and Empower.
As one NTT Data executive put it:
"many non-engineers, particularly in sales and corporate functions, have been able to build and update their own dashboards using plain language"
A Zipline co-founder noted that some of the conclusions the agent surfaces would take even the company's most capable analysts hours to dig up.
How it compares to the query tools built into data platforms
Asking data questions in plain language isn't a new category. Microsoft has put Copilot inside Power BI, Databricks has Genie, Snowflake has Cortex Analyst, and Tableau is also moving toward agent-style features. What these tools have in common is that they live inside their own platform and mainly see their own data and semantic layer.
OpenAI's version sits on top of those platforms instead: the data stays in Snowflake or Databricks, the dashboards stay in Power BI or Tableau, the metric definitions still come from what the company has written in dbt and each platform, and ChatGPT handles the questions, the connecting and the follow-up actions. For enterprises that already have a full BI stack, this position doesn't require replacing any existing system; for BI vendors, their own AI assistants and ChatGPT are now competing for the same entry point on the same dashboard.
A few details are still missing. The announcement's headline says it's "now available to everyone," but it doesn't specify which ChatGPT subscription tiers get access or whether it's billed separately; the named customers are still described as alpha users, and no timeline has been given for a wider rollout.
Sources: OpenAI's official announcement and developer community notes, Unite.AI, CocoLoop, Techzine; the data-source and BI-tool list and internal usage figures follow OpenAI's own announcement.