OpenAI Just Gave ChatGPT Access to Your Company's Database

On September 10, 2026, OpenAI added a Data agent to ChatGPT Work, letting employees ask plain-English questions about company data and get back interactive dashboards instead of a wall of SQL. It's the same week Salesforce, Microsoft, and a wave of enterprise vendors were all pushing their own agent lineups — but this one is aimed squarely at a job nearly every company has and few have automated well: turning a question about the business into an answer someone can act on.

ChatGPT Work itself launched in July 2026 on GPT-5.6, the same week Anthropic shipped Claude Cowork and Microsoft expanded Copilot Cowork — all three built around the same idea of an agent that plans and executes multi-step work rather than just replying to a prompt. The Data agent is OpenAI's move to extend that into business intelligence specifically. If you're evaluating where this fits alongside the rest of your enterprise AI stack, our Enterprise AI Agents pillar is the right starting point.

What It Actually Does

The Data agent connects to a company's existing data sources — OpenAI names Amazon Redshift, Google BigQuery, Databricks, MongoDB, ClickHouse, Snowflake, and Datadog among approved connections — without requiring any data migration. From there, an employee asks a question in plain language (why did sales slow down last quarter, where is spending rising, which accounts are at renewal risk) and the agent investigates what changed, builds an interactive dashboard, and lets the person refine the analysis conversationally rather than learning a query language or waiting on an analyst.

Early access customers give a sense of who this is actually for. NTT Data reported that non-engineers in sales and corporate functions were building and updating their own dashboards without licensing a separate BI tool. Thermo Fisher Scientific used it against its existing data environment to prepare for supplier negotiations. Estée Lauder deployed it to 3,000 employees and quickly expanded to 11,000. A business intelligence analyst at the San Antonio Spurs said work that used to take hours now takes minutes.

How It Compares to Existing BI Copilots

The Data agent isn't the first attempt at natural-language business intelligence — it's entering a category Microsoft and Tableau already compete in, just from a different starting point: it brings the AI to your existing warehouse rather than embedding inside a BI tool you already pay for.

Tool	What It Does	Where It Lives	Pricing
Data agent (ChatGPT Work)	Connects directly to warehouses such as Redshift, BigQuery, Snowflake, Databricks, MongoDB, and ClickHouse, then builds interactive dashboards from plain-English questions.	Inside ChatGPT Work, OpenAI's enterprise agentic workspace	Enterprise workspace pricing; no standalone fee disclosed
Power BI Copilot	Generates a report or dashboard from a natural-language prompt and can write DAX measures on request.	Built into Power BI / Microsoft Fabric	Included with qualifying Power BI / Fabric licenses
Tableau Pulse	Monitors metrics you've already defined and pushes personalized, plain-language insights and anomaly alerts to subscribers.	Tableau Cloud	No extra license cost on Tableau Cloud

 The practical difference: Power BI Copilot and Tableau Pulse both sit on top of a BI platform you're likely already using, inheriting its access controls and semantic model. The Data agent instead connects OpenAI's agent directly to your raw data infrastructure, which is more flexible if you don't want to standardize on one BI vendor, but puts more weight on how carefully you scope what it can see and query.

Where the Real Risk Is

Multi-source business questions are where natural-language analytics tools tend to break: independent research on AI-powered analytics has found meaningfully higher error rates when a query has to identify the correct source, schema, and business logic across several systems, compared to a single, well-defined table. A model with no built-in understanding of your schema will still answer confidently — the answer just won't necessarily be accurate. Before rolling this out past a pilot group, verify its output against a report you already trust, not just against how plausible the dashboard looks.

This is the same governance problem we cover more broadly in Enterprise AI Agent Governance: an agent with query access to live company data needs the same access reviews, audit trails, and human sign-off you'd apply to any system touching financials or customer data — arguably more, since a wrong-but-confident dashboard is easy to act on before anyone questions it.

Should Your Company Use It?

•         Already on ChatGPT Work and want self-serve dashboards without adding a BI license → worth piloting now, starting with a low-stakes reporting use case.

•         Already standardized on Power BI or Tableau → evaluate Power BI Copilot or Tableau Pulse first; you'll inherit governance you've already built rather than layering on a new access path to raw data.

•         Considering it for financial or compliance-sensitive reporting → wait for a broader track record and pair it with the access controls in our governance guide before trusting its output unsupervised.

And before committing budget to any of this, run the same math we walk through in How to Calculate ROI on Enterprise AI Agents — a self-serve dashboard tool only pays off if it actually replaces analyst hours rather than just adding a new tool nobody fully trusts, which is exactly the pattern our Agent Washing piece covers.

FAQ

1. What data sources does the Data agent connect to?

OpenAI lists Amazon Redshift, Google BigQuery, Databricks, MongoDB, ClickHouse, Snowflake, and Datadog among its approved connections, with more being added.

2. Do I need to migrate my data to use it?

No. It connects to data sources where they already live rather than requiring a migration into a new system.

3. Is this the same as Power BI Copilot or Tableau Pulse?

No. Power BI Copilot and Tableau Pulse sit inside a BI platform you already use and inherit its access controls. The Data agent connects OpenAI's agent directly to your underlying data infrastructure instead.

4. Can non-technical employees really use it without SQL?

That's the core pitch, and early customers like NTT Data report exactly that — non-engineers building and maintaining their own dashboards. Accuracy on complex, multi-source questions is the part worth verifying yourself before relying on it.

5. What's the biggest risk in rolling this out?

Confidently wrong answers on questions that span multiple data sources, and under-scoped access to sensitive data. Treat it with the same governance rigor as any other system with query access to company data.