Blue Planet Studio - stock.adobe

Starburst's latest targets agentic AI development

The data lakehouse vendor continues to expand beyond its roots in data mesh, adding features such as an MCP server and access to vector stores that enable users to build agents.

Starburst Data on Thursday unveiled new features targeting agentic AI development and fostering collaboration between humans and agents.

The capabilities -- which support what Starburst terms the "agentic workforce" -- were revealed during AI & Datanova, the vendor's annual user conference in New York City, and will be generally available (GA) in the Starburst Data Platform before the end of 2025.

Features are designed to provide an infrastructure for multiple agents, open and interoperable access to vectorized data, model usage controls including governance, and agent-powered data exploration and analysis.

Starburst unveiled initial AI development capabilities in May, demonstrating its evolution beyond solely data management. The new features, though in line with competitor offerings, add depth, according to Kevin Petrie, an analyst at BARC U.S.

"The announcement is no surprise because every data vendor wants to enable agentic AI," he said. "But Starburst has put some heft into its release."

Specifically, quality and control measures such as agent observability and enabling federated architectures are significant, Petrie continued. Federated data remains in its original system rather than a central repository. This reduces movement while maintaining sovereignty and improving security and compliance.

"I like the focus on compliance-sensitive, distributed data environments," Petrie said. "Compliance-sensitive organizations host nearly half of their data sources, data preparation and feature engineering on-premises for AI projects. These organizations cannot afford to consolidate everything, so [they] need help querying and governing these elements in a consistent, federated fashion."

New capabilities

Like many data management vendors, Starburst has expanded beyond its roots to include AI development capabilities.

Kevin PetrieAnalyst, BARC U.S.

OpenAI's Nov. 2022 launch of ChatGPT marked a substantial improvement in generative AI (GenAI) technology, which now has the potential to make workers better informed and businesses more efficient. As a result, enterprise investments in building chatbots and other GenAI tools surged.

Now, development has evolved to include agents, which are applications programmed with contextual awareness and reasoning capabilities, enabling them to autonomously undertake specific tasks.

In response to customer demand for agent-building capabilities, many data management vendors shifted their AI development focus away from GenAI to agentic AI. For example, Databricks and Snowflake now provide tools specifically geared toward agent development and management.

Starburst is now doing the same in a move driven by customer feedback, according to Matt Fuller, the vendor's co-founder and vice president of AI/ML products.

"This launch is a direct response to what customers have been telling us," he said. "They are ready to scale AI, but they cannot do it responsibly without context, governance and control."

While not referring specifically to an agentic workforce, Starburst's users were asking for the capabilities that enable humans and agents to work together, Fuller continued.

"They were asking how to move faster without losing trust, how to let business users ask questions in natural language without cloud costs running wild, how to ensure AI agents can act while maintaining explainability, privacy and policy enforcement," he said. "These new capabilities [address] those challenges directly. They were not built in a vacuum."

One of the main challenges organizations face when developing and deploying agents is insufficient governance, according to David Menninger, an analyst at ISG Software Research. As a result, Starburst's inclusion of AI governance capabilities makes its new feature set significant for users. 

"Given the current state of the market, any updates that address AI governance should be considered significant," he said. "Our research shows that more than two-thirds of AI platforms have shortcomings in their data governance capabilities. In addition, we have seen increasing concerns about data sovereignty, which is an issue that the Starburst platform helps address." 

Specific new capabilities include the following:

  • A Model Context Protocol (MCP) server and APIs that enable users to build and deploy agents working in conjunction with Starburst's agentic AI-fueled natural language interface to complete increasingly complex tasks.
  • Open, unified access to vector stores to enable retrieval-augmented generation and search across data table formats, including Apache Iceberg and PostgreSQL.
  • AI model monitoring to track usage and control expenses, along with governance to ensure compliance and confident deployment.
  • An improved version of Starburst's conversational analytics agent that lets users query data across different data products and organizational domains.

While the components aimed at enabling the agentic workforce seemingly make sense together, whether they truly lead to successful agent production depends, in part, on the data quality used to train the agents, according to Petrie.

"The Achilles Heel of compliance-sensitive organizations is data quality," he said. "We find that nearly half of them have inconsistent data quality controls when supporting AI projects."

Therefore, it's imperative for Starburst to not only provide development capabilities but also integrate those capabilities with data quality monitoring tools so users can trace AI pipeline errors and model hallucinations to their source, Petrie continued.

"I'll be interested to see how Starburst ties its data quality observability features into its model and agent governance framework," he said.

Meanwhile, because they not only add tools for building agents but -- as Menninger noted -- include governance capabilities, the new features could help Starburst better compete with more established vendors, according to Petrie. Fellow data lakehouse vendor Dremio -- one of Starburst's closest competitors -- already integrates development and governance, as do broad-based data platform vendors such as Databricks and Snowflake.

"These new capabilities, once GA later this year, might help Starburst recover some growth and compete more effectively with Dremio," Petrie said.

A graphic lays out the key attributes of agentic AI and generative AI.Informa TechTarget

Next steps

Looking ahead, Starburst plans to continue focusing on improving multi-agent workflows and providing users with flexible infrastructure options, according to Fuller.

"Our goal is to help organizations move faster from AI experimentation to execution," he said.

Focusing on multi-agent workflows is wise, according to Menninger. Toward that end, he advised Starburst should add support for Agent-to-Agent Protocol (A2A) in addition to MCP.

"While support for MCP is getting a lot of attention to date, it doesn't fully address the issue of multi-agent orchestration," he said. "Support for A2A would complement MCP and facilitate direct interactions between multi-agents." 

Petrie, meanwhile, suggested that Starburst work with partners to ensure straightforward and simple AI development.

"While Starburst gets positive reviews for ease of use, some customers complain about implementation complexity and overall learning curve," he said. "AI projects will make these more complex. I recommend that Starburst work more closely with partners to simplify things and guide customers along their AI journey."

Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.

Dig Deeper on Data governance