What is the Data agent in ChatGPT Work?

The Data agent is a plugin for ChatGPT Work, announced by OpenAI on September 10, 2026, that connects to a company's approved data sources, investigates what changed in a metric, and builds interactive dashboards — all from a conversation, with no SQL or BI tool required. You install it from the Plugins directory as 'Data', connect sources such as Amazon Redshift, Google BigQuery, ClickHouse, Databricks, Datadog, MongoDB or Snowflake, then start a message with @Data and ask a business question. It can also pull files from Google Drive and SharePoint into the analysis.

Which data sources and BI tools does the ChatGPT Data agent support?

Sources listed at launch: Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake 'and more', plus documents from Google Drive and SharePoint. For business context it reads semantic layers such as Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon and existing BI dashboards. On the output side it builds and interacts with dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot, and can share findings through Slack or email.

Who can use the ChatGPT Work Data agent and how do permissions work?

It requires ChatGPT Work, which is available on Plus and above; enterprise administrators choose which data connections are available and which roles can use them via Workspace settings > Plugins. Queries run under the connected account's existing permissions, including table, row and column restrictions, so the agent cannot see more than the user already could. Early customers in OpenAI's Alpha program include NTT Data, Thermo Fisher and ServicePiston.

Is the ChatGPT Data agent a replacement for Tableau or Power BI?

Not as positioned in September 2026. OpenAI built it to sit on top of existing warehouses, semantic layers and BI tools — it writes to Tableau, Power BI, Sigma, Omni, Oracle BI and ThoughtSpot rather than replacing them. The replacement target is the wait: instead of filing a request for an analyst to run a query or update a report, an employee asks @Data and gets an investigated answer and a shareable dashboard in one thread.

Quick Answer

What Is the ChatGPT Work Data Agent? Explained (Sep 2026)

Published: September 11, 2026

The Short Answer

The Data agent is OpenAI’s September 10, 2026 answer to “why did sales slow down?” — a ChatGPT Work plugin that connects to your warehouse and semantic layer, investigates a metric change, builds an interactive dashboard, and shares it, from one conversation. It enforces your existing data permissions, reads your business definitions from dbt/Snowflake Horizon/Databricks Genie, and pushes results into the BI tools you already run.

	Detail (as of September 11, 2026)
Product	”Data” plugin in ChatGPT Work’s Plugins directory; invoked with `@Data`
Requires	ChatGPT Work (Plus and above); admins enable via Workspace settings → Plugins
Data sources	Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake, more; files from Google Drive and SharePoint
Business context	Semantic layers: Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon, BI dashboards
Outputs	Interactive dashboards (editable, shareable, refreshable, brand-styled); dashboards in Omni, Oracle BI, Power BI, Sigma, Tableau, ThoughtSpot; Slack/email sharing; approved actions via connected tools
Permissions	Enforces the connected account’s table/row/column restrictions; admins scope connections by role
Model	ChatGPT Work runs on GPT-6 Astra for paid plans (Astra rollout began September 3, 2026)

What it does, step by step

OpenAI’s launch post frames the problem as latency: getting an answer from company data usually means “waiting for a report or asking someone else to run the analysis.” The Data agent collapses that into a thread:

Connect. Install the Data plugin, complete the account-connection steps for your sources (admins can pre-install and configure Databricks, Snowflake and the rest).
Ask. @Data Diagnose why weekly active users changed last week. Identify likely drivers, compare against prior periods, and recommend the next checks. — one of OpenAI’s own sample prompts.
Investigate. The agent uses your organisation’s metric definitions, custom calculations and data relationships (pulled from semantic layers) to interpret the data, then lets you ask follow-ups and review the evidence behind each finding.
Visualise. Turn the analysis into an interactive dashboard with built-in visualisations; your team can edit, share and refresh it, and you can supply brand guidelines for the look.
Act. Ask for next steps and who needs to be involved; share via Slack or email; carry out actions you approve through connected tools.

OpenAI says it built the feature from internal tooling: nearly all of its product team and over two-thirds of its go-to-market organisation use data agents in ChatGPT Work, made possible by a data team that set shared business definitions, access rules and safeguards for sensitive data. That is the honest prerequisite — the agent is only as good as your semantic layer.

Why the semantic-layer integrations matter

The failure mode of “chat with your database” tools is that the model guesses what revenue means. The Data agent reads that definition from where it already lives — dbt models, Snowflake Horizon, Databricks Genie Ontology, GitHub, or existing dashboards — so a question about churn uses the same calculation your finance team uses. If your definitions are inconsistent across tools, the agent will faithfully reproduce the inconsistency; clean the semantic layer first.

Security and governance

Enterprise administrators choose which data connections exist and which roles can use them.
Every query runs under the connected account’s existing permissions — table, row and column restrictions included.
This is a ChatGPT Work feature, so it inherits ChatGPT Enterprise controls (SSO, SCIM, role-based access, retention settings) on Business and Enterprise plans.

How it compares (September 2026)

	ChatGPT Work Data agent	Databricks Genie	Snowflake Cortex / Intelligence	Microsoft 365 Copilot + Power BI
Lives in	ChatGPT Work chat, cross-source	Databricks workspace	Snowflake	Microsoft 365 apps
Cross-warehouse	Yes — Redshift, BigQuery, ClickHouse, Databricks, MongoDB, Snowflake in one thread	Databricks data	Snowflake data	Power BI datasets
Uses semantic layer	dbt, Horizon, Genie Ontology, BI dashboards	Genie Ontology	Horizon	Power BI semantic models
Writes dashboards to	Tableau, Power BI, Sigma, Omni, Oracle BI, ThoughtSpot + native	Databricks dashboards	Snowflake dashboards	Power BI
Notable	Reads Genie’s and Horizon’s definitions rather than competing with them	Deepest inside Databricks	Deepest inside Snowflake	Best for M365-native orgs

The positioning is unusual: rather than compete with Genie or Cortex, OpenAI consumes their semantic layers and writes back to the BI tools next to them. For a company with data in more than one warehouse, that cross-source thread is the differentiator.

Availability and limits

Announced September 10, 2026; listed as “Data” in the Plugins directory. Alpha-program customers include NTT Data, Thermo Fisher and ServicePiston.
Requires ChatGPT Work — Plus, Pro, Business and Enterprise. Free users are excluded.
Source list is “and more” — check the plugin directory for connectors beyond the seven named.
Actions are gated behind explicit approval and connected tools; it will not modify data on its own.