June 2026 Summaries

Unity Catalog and Starburst have collaborated to address the challenge of enforcing fine-grained access control policies across multiple data engines, specifically focusing on how Starburst queries can respect the governance policies defined in Unity Catalog without re-implementing policy logic. This solution leverages the Iceberg REST catalog API, which allows for server-side scan planning, ensuring that data is sanitized according to user-specific policies before it reaches the Starburst engine. This approach centralizes governance, eliminates redundant policy layers, and enhances compliance by enabling a single source of policy definition that is respected across different data engines. Unity Catalog acts as the central governance point, while Starburst ensures policy adherence, demonstrating the power of interoperability and open standards in modern data lakehouse architectures. The implementation is designed to work seamlessly across various engines, ensuring that organizations can maintain data privacy and compliance while benefiting from the flexibility of choosing the best tools for different workloads.

Building an agentic control plane is crucial for enabling AI systems to transition from merely providing insights to executing actions, ensuring governance and mitigating risks associated with AI-driven decisions. Traditional business intelligence systems often leave the interpretation and action-taking to humans, but AI can streamline this process by removing the analyst from the middle. However, this requires a control plane to manage and govern AI actions, ensuring they are based on the correct context and definitions across different business areas. The agentic control plane acts as a governance boundary, coordinating agents, tools, and complex workflows while enforcing policies and setting guardrails to prevent errors and maintain trust. Context is paramount, as definitions and business rules must accompany data, certified and reviewed by humans before agents can act on them. The control plane also facilitates integrating external services and codifying repeatable workflows, maintaining auditability and compliance. This approach is essential for transforming AI from a tool that informs to one that acts, providing a pathway to operationalizing AI safely and effectively within enterprises.

Starburst Galaxy emphasizes the critical role of a robust data foundation for the successful deployment of enterprise AI, arguing that the shift from experimental to production AI requires data that is accurate, consistent, and auditable, rather than merely focusing on model quality. The release introduces enhancements like Infrastructure as Code with the Terraform Provider for Galaxy and Data Products as Code, which bring software engineering discipline to data management, making the data foundation reproducible and manageable as code. Additionally, features such as serverless Icehouse Table Maintenance and Smart Load Balancing ensure that data remains healthy and stable under unpredictable loads, supporting continuous and efficient data ingestion. These improvements aim to provide a managed, cloud-native platform capable of underpinning scalable and trustworthy AI operations, highlighting that the true breakthrough in enterprise AI lies in a deliberately engineered data foundation rather than in AI model advancements alone.

In the context of enterprise AI, the major bottleneck is not the quality of AI models but the accessibility and understanding of the underlying data, particularly structured data that provides the necessary business context. Structured data is considered the "ground truth" of AI, as it determines the accuracy and governance of AI outputs. The challenge lies in bridging the gap between having data and extracting meaningful insights, which requires curation, precision, and appropriate governance. Companies like Starburst are addressing this by providing data products that ensure consistent and certified definitions across business units, and by enabling flexible data access that accommodates the distributed nature of enterprise data environments. This approach allows AI systems to move beyond merely providing information to enabling actionable insights, thereby enhancing trustworthiness and scalability in AI applications.

Dremio's recent acquisition by SAP raises uncertainties about its future strategic direction, especially as it transitions from being an independent company with its own engineering priorities to becoming part of SAP's broader portfolio. While Dremio has been recognized for its capabilities in data lakehouse architecture, it has faced challenges with concurrency and federation breadth, which are crucial for enterprise-scale operations. The acquisition introduces potential changes in product focus and roadmap, with immediate effects such as the suspension of Dremio's Cloud Enterprise edition and the implementation of telemetry data tracking on Kubernetes deployments. This shift poses compliance issues for organizations with strict data governance requirements. Furthermore, Dremio's position on open formats and ecosystems may be compromised under SAP's proprietary umbrella, which could lead to increased vendor lock-in. Existing customers have a limited window to evaluate alternatives before potential integration-related complexities arise, while prospective customers might consider other options like Starburst, which offers broader connector coverage, better concurrency handling, and a robust governance framework.

Starburst and Qlik have partnered to tackle the challenge of accessing AI business context, a critical bottleneck in AI project success. This partnership leverages Starburst's ability to act as a universal context layer, accessing data across various platforms, including cloud data lakes, warehouses, and on-premises systems, to provide the necessary business context for AI applications. Qlik complements this by offering integration, replication, transformation, and operational support, with capabilities such as change data capture (CDC) and robust visualization tools that facilitate a smooth transition to agentic AI workflows. Together, they aim to improve AI accuracy and enterprise production environments by providing universal access to business context, reducing the operational risks of deploying AI-generated SQL without proper structural awareness. This collaboration not only enhances data accessibility but also transforms data engineering workflows by automating SQL generation and execution, enabling enterprises to scale operations efficiently without accumulating technical debt.

Starburst's 480-e LTS release introduces significant advancements in multi-cluster data architecture and AI readiness, enhancing performance and usability for modern enterprises. Key features include Coordinator High Availability (HA) for minimizing downtime, Load Balancing for horizontal scalability, and Workload Isolation to allocate resources effectively. The release improves query performance by enabling CTE reuse and optimized execution modes, supports advanced integrations with Databricks Unity Catalog, and introduces the OpenAPI Connector for seamless API querying. Enhanced capabilities of the Starburst AI Agent allow for improved natural-language-to-SQL accuracy and cross-cluster data collaboration, while the Iceberg v3 integration offers write isolation workflows and interoperability improvements. The update also brings new approaches to data management, such as Data Products as Code, which streamline data asset management through engineering best practices. The changes support a robust, high-availability platform for managing modern data lakehouses and are designed to reduce engineering friction, enhancing the deployment and governance of complex data architectures.

AIDA Skills is a new feature in private preview designed to extend the analytical capabilities of the AIDA platform by allowing users to incorporate domain-specific knowledge into their workflows through reusable, parameterized Skills. These Skills are human-readable files created in the open agentskills.io format and can include SQL statements for data operations. Users can author Skills either through a structured form within AIDA or by uploading a pre-written SKILL.md file, with both methods resulting in the same artifact. AIDA matches user queries with the appropriate Skills based on natural-language descriptions and example prompts, allowing for both implicit and explicit invocation. Skills execute under the user's permissions to ensure security and are treated as untrusted input to prevent prompt injection. The private preview aims to validate the Skill format, matching loop, and safety model, with future enhancements planned for AI-assisted authoring and governance. Skills empower users by enabling them to address specific business needs without waiting for new product releases, positioning AIDA as more than a chatbot but as a federated query engine enriched by customer-authored domain knowledge.

Artificial intelligence (AI) has the potential to transform organizational decision-making, but its success hinges on the presence of comprehensive business metadata and context. Without these elements, AI systems may produce unreliable results, leading to a loss of trust and hindering AI adoption. The principle of "garbage in, garbage out" underscores the need for data enriched with business context to ensure consistent and accurate AI outputs. Enterprises can prevent disillusionment by implementing robust metadata frameworks, which include data lineage, governance policies, and quality metrics, to provide AI systems with the necessary context for reliable operation. Modern AI architectures, such as data lakehouses and open table formats like Apache Iceberg, facilitate this by combining analytics and metadata management to maintain data integrity and support AI reliability. By prioritizing these foundational elements, organizations can avoid prolonged AI skepticism and foster sustainable adoption, gaining a competitive edge through trustworthy AI systems.

In the context of mid-2026, understanding query plans is crucial for leveraging AI tools to enhance data performance and scalability, particularly with technologies like Trino and Starburst. Despite the capabilities of AI in analyzing query plans, a firm grasp of how modern MPP engines function is necessary to interpret technical jargon and make informed decisions on the recommendations provided. Starburst offers resources such as a three-part video series and hands-on workshops to deepen knowledge in areas like parallel processing fundamentals, query plan syntax, and detailed plan analysis, covering topics like task decomposition, optimizations, and join strategies. These educational tools are designed to empower users to effectively analyze and optimize query plans, ensuring they can fully harness AI's potential in improving data operations. Additionally, Starburst provides comprehensive support through its community forums and professional services, reinforcing the importance of understanding the underlying mechanics of distributed compute engines to maximize the benefits of AI-driven insights.

Context engineering is a critical practice in designing the information inputs for large language models to ensure they produce reliable and accurate outputs, acting as an intermediary between data platforms and model prompts. This engineering discipline is increasingly vital as AI expands beyond traditional roles into areas like customer support and sales, where models rely on external, often untrained data. The core challenge lies in maintaining data freshness, fragment completeness, governance, and schema stability to prevent models from generating outdated or incorrect responses. Effective context engineering enables models to function as fluent analysts, allowing them to integrate diverse data sources and provide accurate, contextually rich answers. By treating context as a foundational platform rather than an isolated feature, organizations can enhance AI applications across various domains, leveraging tools like Starburst's Icehouse architecture for a unified, governed query experience that supports both analytics and AI workloads.

Dynamic connection passthrough is a feature developed by Starburst that allows a single catalog to accept runtime connection contexts at query time, thus eliminating the need for static catalog properties and reducing the complexity of managing multiple data environments. This approach enables organizations to handle multiple data sources more efficiently as it allows the injection of extra credentials at query execution, which dynamically determines the appropriate remote account or configuration to use. By treating the catalog as a reusable template and shifting the complexity from the platform layer to the application layer, dynamic connection passthrough facilitates the support of ephemeral or sandbox environments, multi-tenant analytics, and reduces the operational burden associated with managing numerous catalogs. This feature is supported by connectors such as Snowflake, PostgreSQL, BigQuery, and Iceberg, each requiring specific runtime credentials but adhering to a consistent operational pattern. As the concept expands, it aims to provide a more scalable and adaptable architecture for data virtualization and platform scaling, allowing a more streamlined and predictable interface for applications across diverse data ecosystems.

David Azaria discusses the limitations of the cloud era's promises of elasticity and managed operations, highlighting the overlooked challenge of operational portability in data management. The cloud solved compute scalability but often trapped users within a single vendor’s ecosystem, making true portability elusive. While data format portability has been achieved through open table formats like Iceberg and Delta Lake, the real challenge lies in operational portability, where managed services can operate where the data resides, such as with Starburst's Bring Your Own Cloud (BYOC) model. This model allows organizations to maintain managed operations within their own cloud infrastructure, accommodating diverse regulatory and security needs without forcing data to relocate. Azaria argues for a shift in architecture that prioritizes the proximity of compute to data, reflecting the industry's need for platforms that adapt to data's location rather than requiring data to conform to platform constraints. This approach is increasingly relevant in today's fragmented data landscape, where enterprises face diverse data residency and regulatory challenges that demand flexible deployment models.

Starburst has integrated NVIDIA's cuDF, an open-source data processing toolkit, into its SQL query engine to enable GPU-accelerated query processing, significantly improving performance without requiring changes to SQL or application code. This integration allows for up to 6x speedup in TPC-H benchmark queries and 3.4x speedup in ClickBench queries, with some individual queries executing over 11 times faster. The enhanced throughput is achieved at a cost comparable to or lower than CPU-only instances, challenging the assumption that GPU infrastructure is more expensive. This development reflects a technical collaboration between Starburst and NVIDIA, aiming to make GPU-accelerated SQL a staple in enterprise data platforms. The integration specifically targets the physical operator level, allowing data to remain on the GPU across multiple execution operators, thereby reducing data transfer overheads and compounding performance benefits. While still in active development, with plans for further optimization and expanded operator coverage, the current implementation demonstrates significant potential for reducing latency in agentic AI systems and improving efficiency in analytics workloads.