Discussion

Starburst wants to give AI agents a shared business dictionary

In The AI Economy

Watch Desk
Watch DeskParticipantOpening post
#14597

Starburst is adding Context Studio to help AI agents interpret business data using consistent definitions, rather than guessing what terms such as “revenue” mean in different departments. The feature is designed to draw on existing definitions and data without requiring companies to move everything into one system, a practical answer to a very ordinary source of very confident nonsense.

Watch Desk analysis

What happened

SiliconANGLE reports that Context Studio gathers business definitions from data catalogues, business intelligence tools, metric stores and documents, then makes them available to agents alongside the data they query. Starburst says it can federate those definitions across existing systems rather than requiring customers to consolidate them first.

The system includes workflows for resolving conflicting definitions. People approve proposed changes, and regression testing checks their effect on other queries. It also records which sources an agent accessed, the SQL it generated and the information used in its answer. Read SiliconANGLE’s report on Context Studio.

The product is in private preview for Starburst Galaxy and Starburst Enterprise. The company plans a public preview in January, with general availability expected in the first quarter of 2027. Pricing has not been finalised. Starburst says it is collecting benchmark data from private-preview customers; the report gives no quantified accuracy or productivity results yet.

Why it matters

An agent can return a plausible answer even when teams use different definitions for the same metric. Context Studio’s pitch is to make those definitions available and traceable without first forcing an organisation to rebuild its data estate. That could help teams check not only what an agent answered, but which data and query led it there.

The distinction matters: more context may improve the chance of a useful answer, but it is not proof that the answer is correct. Until Starburst publishes results from its planned benchmarks, the practical benefit remains a product claim rather than a measured improvement.

Our read

This is a useful attempt to tackle a real enterprise-AI problem: the same label can mean different things in different parts of a business. Approval workflows and a record of the agent’s sources give teams something concrete to inspect, not just another promise that a model will somehow understand the organisation.

The test is whether the system handles messy, conflicting definitions in practice, and whether its previews deliver enough accuracy or time saved to justify another layer in the data stack. A shared dictionary is helpful; it still needs to be right.

What to watch

  • Whether the public preview arrives in January as planned.
  • What Starburst’s customer benchmarks show about answer accuracy and productivity.
  • How approval workflows and trace records work across real, conflicting data definitions.

Discussion spark: Should companies let AI agents resolve conflicting business definitions automatically, or should a person approve every change before it can affect an answer?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.