Data Formulator 0.7 tries to rebuild enterprise data analytics around AI agents

Microsoft Research released Data Formulator 0.7 as an analytics workspace where AI agents assist with the full process: bringing enterprise data into a workspace, exploring it, analyzing it and turning it into visualizations. This is not just chart generation from a prompt. It is an attempt to connect data preparation, exploration and visualization inside one interactive process.

An agent designed to handle data prep, exploration and visualization in one step

The core of the approach is using AI agents directly over analytical work, not merely as a chat interface over a finished result. A user can bring enterprise data in, ask exploratory questions and let the agent suggest transformations and the final visualization.

For Microsoft, this is a logical direction. Copilot and agentic interfaces make more sense when grounded in structured business data, not only documents and chat. Data Formulator tests whether this layer can be built for people who are not full data engineers.

Enterprise analytics has a problem exactly where the agent promises to help

Enterprise analytics still contains many manual steps: cleaning data, choosing the right chart, iterating on queries, explaining the result and preparing output for another person. If an agent handles this layer, it speeds up people who make data-driven decisions every day but lack the data engineering background.

Control also matters. Local workflows and auditable steps can be easier to govern than a general AI assistant that routes everything through a remote model without visible intermediate steps.

Enterprise data is messy, permissioned and poorly documented: the agent has to handle that outside the demo

The hardest part will not be the UI. Enterprise data is messy, permissioned and often poorly documented. To work beyond a demo, the tool must handle governance, audit, data provenance and a clear separation between a suggested insight and a verified conclusion.

Without that, an agent can easily generate a visualization that looks convincing but rests on the wrong query or an undocumented transformation. False confidence from a fast chart is more expensive in enterprise settings than a slow manual analysis.

The signal will be integration into Fabric or Power BI and real deployment outside the research blog

Watch whether Data Formulator gets real integrations into Microsoft Fabric, Power BI or the broader Copilot stack. The practical proof is simple: less manual work in preparing analysis and less false confidence in charts.

If the project stays in research demo mode without production adoption, that will be a signal that the agentic analytics layer is harder than the UI makes it look.

Lilith's verdict

Data Formulator targets the point where a table turns into a decision. The agent promises to take over the data preparation work, but in enterprise it will only succeed when it handles data that is not clean and never was.