Data teams that can’t query their own datasets in real time are already losing ground to the ones that can.

Analysts are drowning in data they can’t actually use

The painful reality for most data-heavy teams is a two-step trap: waiting days for an engineer to write a query, then waiting again for a human to interpret the output. Those two delays compound into decisions made on stale information.

GPT-5 and Codex are doing the querying now

Inside OpenAI’s in connects to internal datasets, accepts plain-language questions, and returns reasoned, cite-able answers within minutes by using GPT-5 to interpret intent, Codex to generate and run the query, and a memory layer to retain context across sessions. You type a business question, and the agent decides which data it needs, writes the code to pull it, runs that code, and hands back a structured finding. The memory component means follow-up questions don’t start from scratch.

Data engineers are not the only ones who needed this

This is most immediately useful for professionals who make decisions on numbers but can’t write SQL to save their careers.

• Business intelligence leads who wait 48 hours for a one-line query to confirm a budget assumption
• Product managers who need usage-pattern data to justify a roadmap call but have no direct database access
• Strategy analysts who lose hours reformatting exported CSVs before any actual analysis can begin

The common thread is proximity to data without the technical access that makes it usable.

The enterprise data tool market just got a credible internal benchmark

Competitors like Databricks and ThoughtSpot have spent years pitching AI-assisted analytics to enterprises, and OpenAI building this for internal use signals that the bar for what counts as fast and reliable has moved sharply upward. If the model that powers the agent is also the one OpenAI ships commercially, enterprise buyers will start holding every vendor to the same standard.

What you can actually do with it

• Ask a plain-language question and get a SQL-backed answer in minutes
• Run follow-up queries without re-explaining dataset context
• Pull cross-table insights without touching a database interface
• Export reasoned findings with the query logic attached for audit trails

Pricing not listed — check our directory.

One real constraint worth knowing

Because OpenAI’s data agent was built for internal infrastructure, public availability, integration scope, and data privacy boundaries for external enterprise use have not yet been detailed.

The alternatives are older and slower

ThoughtSpot offers natural-language analytics but requires significant setup and schema mapping before it returns anything useful. Tools like Mode or Hex give analysts a notebook environment, but they still require someone who can write code to be in the loop.

Enterprise AI analytics is being redefined from the inside out

What OpenAI built internally sets a new expectation for how fast data questions should get answered, and that expectation is about to hit every vendor pitch deck in the analytics space. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.