ChatGPT's New Data Agent Lets You Build Dashboards by Just Asking Questions

OpenAI just shipped a Data agent inside ChatGPT Work that connects to live company databases and turns a plain-language question into an interactive, shareable dashboard. No SQL. No waiting on an analyst. No learning a separate BI tool.

For designers who are starting to build their own products, this matters a lot. Data has always been one of the walls between "I have an idea" and "I can prove it works." That wall just got a lot shorter.

What the Data Agent Actually Does

The agent connects to data sources your organisation has already approved: things like Google BigQuery, Snowflake, Amazon Redshift, MongoDB, Databricks, and more. It can also pull in files from Google Drive or SharePoint. Once connected, you ask it a question in plain English, and it investigates the data, surfaces findings, and builds a visual dashboard you can share with your team.

A semantic layer (a set of business definitions and metric rules your company has already set up) sits between the agent and the raw data. That means when you ask "why did signups drop last week?" the agent interprets "signups" the same way your data team does, not just whatever column name happens to be in the database.

Enterprise admins control which connections are available and which team members can use them. Existing permission rules, right down to which rows a user is allowed to see, carry through to every query the agent runs. So it is not a free-for-all with sensitive data.

Why This Is a Genuine Shift for Designer-Builders

If you are a designer building a vibe-coded app or internal tool, your biggest blind spot is usually feedback loops. You ship something, and then you either wait for someone to pull usage numbers or you fly blind. Neither is great.

The Data agent changes that loop. You can connect to whatever data your product generates, ask conversational follow-up questions to dig into specifics, and then turn the results into a dashboard your stakeholders can actually read. The agent supports integrations with Tableau, Power BI, Omni, Sigma, and ThoughtSpot as well, so if your organisation already lives in one of those tools, you are not starting from scratch.

That last point is bigger than it sounds. Being able to hand a non-technical stakeholder a live, refreshable dashboard and say "here is how the feature is performing" used to require either a dedicated analyst or a lot of your own time wrangling spreadsheets. Now it is closer to a conversation.

A Practical Workflow: From Design Question to Dashboard

Say you have shipped a new onboarding flow and you want to know whether it is reducing drop-off. Here is how a designer-builder could use the Data agent today.

First, make sure your product's event data is flowing into one of the supported sources. If you are using a vibe-coding tool that writes to a database, check whether it is compatible (Redshift, BigQuery, Snowflake, MongoDB, and Databricks are all on the list).

Once connected, open ChatGPT Work and install the Data agent. Then ask a direct question: "Show me completion rates for each step of the onboarding flow over the last 30 days." The agent investigates, surfaces a result, and you can ask follow-up questions the way you would in a normal conversation: "Which step has the highest drop-off?" or "How does that compare to the previous version?"

When you are happy with the analysis, ask it to turn the findings into an interactive dashboard. You can even share your brand guidelines so the output matches your organisation's visual style. Then share the link with whoever needs to see it.

The whole loop, from question to shareable artefact, happens inside a tool most designers already have open.

What to Watch For

A few honest caveats before you get too excited.

The Data agent is part of ChatGPT Work, which is OpenAI's paid business tier. Individual or free-plan users are not the primary audience here.

The quality of what the agent produces depends heavily on what context your organisation has set up in its semantic layer. If your data is messy, or if your company has not yet defined its key metrics in a tool like dbt or Databricks Genie, the agent will be working without a map. Garbage in, confident-sounding garbage out.

And because the agent enforces existing permissions, you will only ever see data you are already authorised to see. That is a feature, not a bug, but it does mean you need to be in the right access group before you start.

The Grounded Takeaway

The Data agent is not magic. It still depends on clean data, sensible permissions, and someone having done the groundwork of connecting the right sources. But for a designer who is building their own tools and wants to close the loop between shipping and learning, it removes a very real barrier. The skill you now need is knowing what question to ask, which, honestly, is what good design thinking is already about.