OpenAI's new data agent turns business metrics into actions

Every business runs on numbers, but getting answers usually involves an all-too-familiar drag: open a ticket, wait days for a data team to write queries, and hope the resulting chart actually answers your original question.

OpenAI wants to bypass that bottleneck entirely. The company just released a new Data agent inside ChatGPT Work, built to let business users interrogate company data, generate live dashboards, and trigger follow-up actions entirely in plain English.

Plugging directly into your data warehouse and metrics

Unlike standard chatbots that attempt to guess SQL queries in a vacuum, this agent hooks directly into enterprise data stores.

Out of the box, it connects to Amazon Redshift, Google BigQuery, Snowflake, Databricks, MongoDB, ClickHouse, and Datadog. It can also pull unstructured files from Google Drive and SharePoint into the same analytical loop.

Crucially, it taps into your company's existing business logic rather than making up calculations. The agent reads metric definitions and data relationships from semantic layers and repositories like dbt, Databricks Genie Ontology, Snowflake Horizon, and GitHub.

Enterprise administrators determine which connections and roles are enabled. The queries strictly respect existing permissions down to the table, row, and column level, so sensitive data stays locked behind company boundaries.

Once an analysis runs, the agent can display the underlying evidence, build interactive dashboards directly in the conversation, or sync changes back into BI tools like Tableau, Power BI, Sigma, ThoughtSpot, Oracle BI, and Omni.

It can also execute next steps, such as alerting teammates over Slack, sending an email summary, or kicking off actions inside connected software once you approve them.

OpenAI notes that early alpha testers include NTT Data, Thermo Fisher Scientific, and ServicePiston. Internally, nearly all of OpenAI's product team and more than two-thirds of its go-to-market organization are already using it.

Why it matters

For non-technical business operators, this turns what used to be a week-long analytics ticket into a two-minute chat. Being able to ask why weekly active users dropped and get a verifiable answer with drill-downs shifts business intelligence from passive retrospective reporting to active problem-solving.

For the broader tech landscape, OpenAI is pulling off a clever platform play. Instead of trying to build a rival data warehouse to displace Snowflake or Databricks, or replacing established BI tools like Tableau, OpenAI is inserting ChatGPT as the conversational universal interface on top of everyone's existing stack. That puts serious pressure on native enterprise AI assistants built by those same warehouse providers.

However, we think the real bottleneck will be internal data cleanliness. A tool like this is only as dependable as the underlying semantic definitions and dbt models feeding it. If a company's data architecture is fragmented, an AI agent will only produce faster, more confident misinterpretations.