AI Business

Starburst brings business context to AI agents to reduce hallucinations and build trust

Starburst Data is rolling out Context Studio, a new feature that feeds AI agents consistent business definitions from across an enterprise to help them deliver more reliable answers.

·3 min read
Starburst adds context support to give AI agents consistent access to business data definitions
Starburst adds context support to give AI agents consistent access to business data definitions

Starburst Data Inc. is broadening its data federation platform to include business context management through Context Studio, a fresh capability aimed at enabling artificial intelligence agents to deliver more dependable and credible responses by tapping into information spread throughout enterprise infrastructure.

Context Studio gathers business definitions from existing data catalogs, business intelligence platforms, metric stores and documentation, then makes these definitions available alongside data when AI agents process requests. The tool is built to prevent the conflicting definitions that frequently lead AI agents to generate inaccurate output.

The capability emerged from challenges Starburst observed as its customers began running Aida, its natural-language data assistant, in live production settings, according to Starburst co-founder and Chief Executive Justin Borgman. "We've found that as you get into production, the biggest issue is context because you need to trust the answers the agent gives back," he said.

The issue becomes more severe in large enterprises where separate business units may define the same metrics—such as revenue or active customers—in different ways. When AI models lack a unified grasp of these definitions, they can generate plausible yet incorrect answers.

"When those definitions conflict, the [large language model] doesn't know any better; it's just going to guess," Borgman said. "That's what leads to bad answers and an erosion of trust."

Data products

Context Studio sits on top of Starburst's current data products framework, which lets organizations construct filtered perspectives that span numerous data sources. These products embed business definitions, metadata and permission settings that apply uniformly when agents pull information.

In contrast to traditional semantic layers that demand organizations consolidate definitions into a single system, Context Studio federates existing definitions across platforms without forcing customers to transfer or reconstruct their data infrastructure.

The system includes built-in workflows to settle conflicting definitions, plus a feedback channel through which agents can recommend enhancements. Human teams authorize modifications before regression testing validates their effects on other queries.

Borgman framed the move as a natural continuation of Starburst's established practice of accessing distributed data without consolidating it into one location.

"We think context is going to live everywhere, just like we think data will live in multiple places," he said. "Our approach treats [context] as a first-class principle in our design."

Starburst has not yet released numerical results demonstrating Context Studio's effect on AI accuracy or output speed. The organization stated it is gathering performance metrics from early-access customers in private preview.

The company is also prioritizing transparency to strengthen trust in AI-produced answers. Context Studio logs which data sources agents connect to, the SQL queries they write and the facts they use to create responses. Permission controls stay attached to the underlying data products.

"Context is going to increase the probability of a good answer, but there's also a need to trace how it got to that answer," Borgman said.

An agent can engage with subject-matter experts to gather business knowledge and turn it into preliminary data products. Starburst is also widening its Aida data assistant to handle queries spanning multiple data products and launching an agent that builds dashboards on its own.

Further platform additions include Mission Control for tracking consumption and expenses, automated Apache Iceberg upkeep, a customer-supplied cloud option and graphics processing unit-powered query acceleration. Starburst reports its newly unveiled Torin query engine achieves up to sevenfold speed gains in its own testing.

Context Studio enters private preview for Starburst Galaxy and Starburst Enterprise starting now. A public preview launch is scheduled for January, with full availability targeted for the first quarter of 2027. Final pricing is still being determined.