The new Data agent connects ChatGPT Work to warehouses and BI tools, investigates what changed, and builds shareable dashboards. No CRM connector is named.
Until now the rep's pipeline question went to whoever owned the BI tool, and the answer came back when it came back. This puts the question in the chat window the rep already uses, against the same data, with the permissions the admin already set. The seller who can ask "which of my accounts slipped a stage this month, and what changed in them" on Monday morning builds the week's plan before the team meeting; the seller waiting on a dashboard request is still guessing at it. The catch is that the data has to be reachable first, and OpenAI's connector list stops short of the CRM.
A "data agent" is a chat assistant that runs queries for you. You ask in plain words; it writes the query, reads the result, and asks itself follow-up questions before it answers. "Connected data" means your company's databases and reporting tools, plugged in by an administrator. A "semantic layer" is the company's dictionary of what words like "pipeline," "bookings," and "active account" mean in the numbers, so the agent uses your definition instead of inventing one. An "interactive dashboard" is a set of charts you can click and filter rather than a screenshot.
The short version: if your revenue numbers already sit in a warehouse or a BI tool, an admin can switch this on and you can question them in ChatGPT instead of waiting for a report.
OpenAI introduced a Data agent in ChatGPT Work on 10 September 2026. In OpenAI's words, it "connects to your company data, investigates what changed, and builds interactive dashboards you can share." The questions OpenAI uses to frame it are the ones a revenue team asks every Monday: "Why did sales slow down? Where is spending rising? Which issues threaten renewals in our largest accounts, and what should we fix first?" It appears as "Data" in the Plugins directory, and it "uses your organization's business terms, metric definitions, custom calculations, and data relationships to interpret the data." Findings can go out through Slack or email, and it can "carry out the actions you approve through connected tools."
Announced: 10 September 2026, on openai.com; the ChatGPT release notes carry a same-day entry for a "Data plugin in ChatGPT Work and Codex."
Availability: ChatGPT Work, via the Plugins directory. "Administrators can make it available or install it for their teams through Workspace settings > Plugins." Which paid tiers include it: Not stated.
Price: Not stated.
Regions: Not stated.
Connectors named: data sources "Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake, and more"; BI tools "Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot"; files and documents from Google Drive and SharePoint; and business context from "semantic layers and trusted sources such as Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon, and BI dashboards." OpenAI says "over two-thirds of our GTM organization use data agents in ChatGPT Work to analyze company data themselves."
What it doesn't do: The announcement does not name Salesforce, HubSpot, or any CRM as a connector. The pipeline questions it describes only work if your CRM data already lands in a warehouse or BI tool on the list. It does not state pricing, regions, or which ChatGPT Work tiers get it. It analyzes and charts; it is not a place to write outreach.
SDR — Your sequence and reply data usually lives in an outreach tool, not a warehouse, so this touches you only if RevOps pipes it in. If they have, keep the question narrow: which of your accounts had activity or a stage change this week that you did not cause. Do not expect it to write the follow-up; that stays in your prompt library.
AE — The renewal question OpenAI quotes is yours. Ask which of your largest accounts show usage or support signals that threaten the renewal, and take the dashboard into the QBR instead of a spreadsheet you assembled by hand. Check one number against the source system before you trust the chart; the agent is only as right as the metric definitions behind it.
Manager — Pipeline review can stop being a slide-building exercise. Ask why stage conversion moved this quarter and which reps it moved for, then share the view so the team argues with the same chart. The risk is a room full of people disputing a number nobody can audit, so agree the metric definitions in the semantic layer before you run the first review off it.
RevOps — You are the gate. The agent sees only what the admin connects and what the roles allow, and the connector list covers warehouses and BI tools, not the CRM directly. If Salesforce or HubSpot is not already syncing to Snowflake, BigQuery, Databricks or a BI tool on the list, the sales questions in OpenAI's pitch do not work yet. Your first job is deciding which tables and definitions the sales org gets; your second is watching which questions actually get asked, because that is the reporting backlog you no longer have to build.
The agent answers "why did sales slow down" only once the data is connected. This prompt runs the same diagnosis on the pipeline metrics you give it, with a health score, stage bottlenecks and acceleration moves, so you have a baseline now and a comparison the day the agent is switched on.
Two minutes, once a week. What changed in AI, and what to run because of it.