What external business systems does the OpenAI ChatGPT Data agent support?

The Data agent plugin allows users to connect ChatGPT Work directly to major business systems such as Snowflake, Tableau, Power BI, Slack, and Google Drive. It enables employees to query complex corporate databases using natural language without needing specialized analytics software.

Is there a public accuracy benchmark for the new ChatGPT Data agent?

OpenAI has not released any public accuracy benchmarks for this tool, which raises concerns about enterprise trust. Organizations must currently rely on their own internal testing to validate whether the AI's interpretations of complex business logic are reliable enough for critical decision-making.

Why did OpenAI create the ChatGPT Data agent?

The tool was originally developed internally at OpenAI to solve data bottlenecks, with nearly all product staff and over two-thirds of go-to-market employees now using it. This internal adoption serves as a signal of confidence in the technology's ability to handle complex tasks across multiple business systems.

How does the ChatGPT Data agent handle user permissions and security?

The plugin maintains existing table-, row-, and column-level permissions enforced by the connected data systems. This ensures that access controls are preserved when users interact with corporate databases through natural language queries.

How does the ChatGPT Work Data agent differ from previous search features?

The Data agent shifts focus from simple information retrieval to actionable workflows by creating interactive dashboards and executing follow-up actions. It aims to become a central hub for workplace data analysis, allowing users to synthesize insights across different platforms like Slack and Snowflake.

OpenAI Launches ChatGPT Work Data Agent Without Benchmarks

TL;DR: OpenAI launched its ChatGPT Work Data agent on September 10, 2026, enabling natural language queries across systems like Snowflake and Tableau without providing public accuracy benchmarks. This lack of external validation poses a significant hurdle for enterprise adoption, as organizations must rely on internal testing to trust the AI’s analytical output for critical business decisions.

Key facts

OpenAI launched the ChatGPT Work Data agent plugin on September 10, 2026.
The tool connects to major systems including Snowflake, Tableau, Redshift, BigQuery, Databricks, Power BI, Slack, Google Drive, and SharePoint.
Nearly all OpenAI product staff and over two-thirds of its go-to-market employees use the agent internally.
Existing table-, row-, and column-level permissions from connected systems are enforced by the plugin.
The launch lacks any public accuracy benchmark for enterprise trust validation.
Users can generate insights, create interactive dashboards, and execute approved follow-up actions via natural language queries.

OpenAI brings enterprise data to ChatGPT Work, but leaves accuracy unverified

OpenAI has launched a new “Data agent” plugin for its ChatGPT Work platform, allowing users to connect corporate databases and generate insights through natural language. Introduced on September 10, 2026 [1][2], the tool aims to let employees query complex business data without needing specialized analytics software or waiting for manual reports from IT teams.

The Data agent connects directly to a wide range of external business systems, including Redshift, BigQuery, Databricks, Snowflake, Tableau, Power BI, Slack, Google Drive, and SharePoint [2]. By integrating these diverse sources into a single conversational workflow, OpenAI positions ChatGPT Work as a central hub for workplace data analysis. The agent is designed to investigate changes in business data, expose supporting evidence, create collaborative dashboards, and execute approved follow-up actions through connected tools [2].

From internal bottleneck to public feature

The development of this public-facing tool stems from an internal workflow OpenAI created to address its own data bottlenecks. According to the company, nearly all product staff and more than two-thirds of its go-to-market employees now use these data agents internally [2]. This internal adoption suggests that the tool is designed to handle questions spanning multiple business systems rather than replacing dedicated data platforms entirely [2].

OpenAI emphasizes that existing table-, row-, and column-level permissions remain enforced by the connected systems, ensuring that access controls are maintained when using the Data agent [2]. The plugin allows users to turn company data into answers, interactive dashboards, and actions via natural language queries [1][2].

The missing benchmark problem

Despite these capabilities, the launch has drawn attention for the absence of a public accuracy benchmark. Industry analysis notes that while the tool generalizes internal analytics workflows for broader use, its adoption in enterprise settings will depend on whether organizations trust its answers enough to widen permissions beyond basic queries [2].

The external product grew from an internal tool OpenAI built to address its own data, permissions, and workflow bottlenecks [2]. However, the lack of external validation metrics remains a notable feature of this initial release. Without a public accuracy benchmark, enterprises must rely on their own testing to determine if the AI’s interpretations of complex business logic are reliable enough for critical decision-making.

Shifting from search to actionable workflows

This launch marks a significant shift in OpenAI’s enterprise strategy. Previous iterations of ChatGPT focused primarily on simple search and conversational interfaces, such as Etsy’s app which allowed users to find products through chat [Editor Note Context]. The new Data agent moves beyond simple information retrieval into actionable data workflows.

By enabling the creation of interactive dashboards and direct integration with tools like Tableau and Power BI, OpenAI is attempting to embed AI directly into the daily operational rhythm of a company. This approach contrasts with earlier attempts where AI acted as an external layer over existing software. Instead, ChatGPT Work aims to become the primary interface for accessing and interpreting internal corporate data.

The tool targets questions that span multiple business systems rather than replacing dedicated data platforms entirely [2]. For example, a sales manager might ask the agent to compare Slack communication trends with Snowflake sales figures, resulting in a synthesized dashboard. This capability reduces the friction between asking a question and getting an answer, potentially accelerating decision-making processes.

What this means for enterprise adoption

The introduction of the Data agent highlights OpenAI’s growing focus on the “Work” segment of its business model. By leveraging internal success stories—where most product staff use the tool—the company is signaling confidence in the technology’s reliability for complex tasks [2]. However, the absence of a public accuracy benchmark remains a significant hurdle for widespread enterprise adoption.

Organizations will likely need to establish their own rigorous testing protocols before deploying the Data agent for critical business functions. The reliance on existing permission structures helps mitigate some security concerns, but trust in the AI’s analytical output is a separate challenge that requires validation [2].

As OpenAI continues to expand ChatGPT Work’s capabilities, the Data agent represents a bold step toward making corporate data accessible to non-technical users. Whether this accessibility comes with enough accuracy guarantees for high-stakes business decisions remains to be seen in the coming months.

Sources

OpenAI (@OpenAI) on X (x.com) — 2026-09-10
OpenAI Launches a Data Agent for ChatGPT Work Without a Public Accuracy Benchmark (superpowerdaily.com) — 2026-09-10