ChatGPT's New Data Agent Builds Dashboards From One Question

On 10 September 2026 OpenAI launched a Data agent inside ChatGPT Work that turns a typed question into a finished chart or dashboard, no spreadsheet formulas or database queries needed. For anyone drowning in sales reports and WhatsApp order lists, it is a preview of how fast reading your own numbers is about to become.

What did OpenAI actually launch?

The Data agent connects to a company's approved data sources, including Amazon Redshift, Google BigQuery, Databricks, Snowflake, MongoDB and ClickHouse, and can also pull in files from Google Drive and SharePoint. Instead of asking someone to run a report, a staff member can type a question such as why sales slowed last month, and the agent investigates the numbers, explains what changed, and builds a shareable dashboard on the spot. It uses the company's own business terms and metric definitions, drawn from tools like dbt and BI dashboards, so the answers match how the business already talks about its numbers.

The results are not stuck inside ChatGPT either. The agent can build and update dashboards directly in Tableau, Power BI, Sigma and other business intelligence tools, and administrators decide which data connections staff can use, with the same row and column permissions already in place.

At Zipline, we're using the Data agent in ChatGPT Work to explore our company data and build dashboards just by asking in natural language. In early testing, it surfaced profound findings that would have taken one of our best people hours of digging through the data to uncover.from Ryan Oksenhorn, Co-founder, Zipline (OpenAI, 10 September 2026)

Does a small business even need this?

Honestly, not this exact version. The Data agent is built for companies that already run a data warehouse and pay for ChatGPT Work, which is aimed at larger organisations, not a single-outlet shop. OpenAI's own customer examples in the announcement, names like NTT Data and Thermo Fisher Scientific, are large enterprises with existing IT teams. But the underlying idea, describing your numbers in plain language instead of building formulas, is already something every business can try today, and it is the direction every major AI tool is heading.

What can you already do with ChatGPT or Claude right now?

• Export a sales, expense or inventory report from your accounting or point-of-sale system as an Excel or CSV file.
• Upload that file into a ChatGPT or Claude conversation on a proper business plan, not a free personal account.
• Ask a specific question in plain English or Malay, such as which product sold the least last month compared with the month before, and ask it to explain why.

Also today

• OpenAI opened a public beta Agents API to all developers on 10 September 2026, giving businesses a standard way to build AI agents that carry out multi-step tasks such as logistics tracking or customer support without custom engineering (OpenAI).
• Anthropic's September 2026 threat report describes scammers running fake discount websites that pose as Claude resellers to steal buyers' login credentials, a reminder to only buy AI subscriptions from the official Claude, ChatGPT or Gemini sites (Anthropic).

What should you do this week?

Pick one number you check every week by hand, weekly sales, top-selling items or outstanding invoices, and try asking an AI tool to explain it from a spreadsheet instead of building the report yourself. It will not be as polished as OpenAI's enterprise dashboard, but it is the same skill: asking a clear question and trusting the AI to do the digging, then checking the answer against what you already know about your own business. That is exactly the habit we build hands-on at NextGen Training Academy in Kota Kinabalu, in the Master Claude AI for Business Productivity program, using real business documents instead of toy examples, so staff leave knowing how to question their own numbers, not just how to click a button.