The Evolution of the Lakehouse: Databricks, Snowflake, and Starburst as Agentic Control Planes
The Evolution of the Lakehouse: Databricks, Snowflake, and Starburst as Agentic Control Planes
Summary
A fundamental shift is underway among leading data lakehouse vendors: they are transitioning from being simple data storage and query engines into agentic enterprise control planes. Platforms like Databricks (with Genie One, Omnigent, and Unity Catalog), Snowflake (with CoWork/SnowWork), and Starburst Galaxy are introducing runtime and context layers to govern, secure, and coordinate autonomous AI agents directly on enterprise data stores. This transition marks the beginning of the “Agentic Enterprise” era, where data and AI orchestration merge.
What happened
Over the past few weeks, major data platform players have announced updates positioning the Data Lakehouse as the foundation for agentic AI workloads:
- Databricks introduced its “Agentic Enterprise Control Plane” vision at its summit. Core pillars include an expanded Unity Catalog to govern agents, models, and tools; Genie One, a data-smart AI coworker designed to automate workflows; and Omnigent, an open-source (Apache 2.0) meta-harness allowing developers to compose and swap agent frameworks.
- Snowflake officially rebranded Snowflake Intelligence to Snowflake CoWork. Built on the autonomous capabilities previewed in Project SnowWork, CoWork functions as a natural language personal work assistant executing multi-step business workflows while maintaining security policies via Cortex Sense and RBAC.
- Starburst Galaxy launched a federated data foundation designed for agentic AI, utilizing a built-in Model Context Protocol (MCP) Server that allows agents like Claude Code or ChatGPT Enterprise to securely access distributed data products without complex ETL pipelines.
Why it matters
For developers, data engineers, and enterprises, the requirements for running AI agents in production are changing. The main bottleneck is no longer raw model capability, but providing precise, high-fidelity context (“Golden Context”) and real-time governance. By transforming lakehouses into control planes:
- Security & Compliance: Agents inherit existing fine-grained access controls (RBAC, column masking) directly from the data layer.
- Data Sovereignty & Federation: Starburst Galaxy demonstrates that agents can leverage data where it resides across hybrid/multi-cloud environments, reducing infrastructure and regulatory friction.
- From Analysis to Action: Systems like Genie One and CoWork show that data platforms are moving beyond generating dashboards to executing multi-step actions across business tools (Slack, Gmail, Jira).
Evidence
The market’s movement toward agentic control planes is backed by official launches and industry reports:
- Bain Insights analyzed the Databricks Data + AI Summit, highlighting the transition of lakehouses into orchestrators of autonomous enterprise agents.
- Starburst published technical architecture details showing how Starburst Galaxy provides federated data products to AI agents using the Model Context Protocol (MCP).
- Snowflake launched Snowflake CoWork at its annual summit, outlining the path from the Project SnowWork preview to a production-ready enterprise agent.
Analysis
This trend highlights that LLMs in enterprise settings are ineffective without a structured context and security layer. Databricks’ launch of Omnigent is a strategic move to prevent vendor lock-in at the agent layer while cementing Unity Catalog as the default governance standard. Snowflake’s transition to CoWork underscores a shift from passive Business Intelligence to active human-machine collaboration. However, standardization remains a critical challenge. Starburst’s adoption of MCP is a step in the right direction, but widespread industry alignment is needed to prevent fragmented agent ecosystems.
Practical Takeaways
Enterprises and developers looking to adopt agentic workflows should:
- Govern Tools via Catalogs: Leverage Unity Catalog or Snowflake CoWork to manage AI agent tool permissions with the same rigor as human credentials.
- Prioritize Context Engineering: Ensure enterprise data is curated as clean “Data Products as Code” to provide high-fidelity context and minimize agent hallucinations.
- Explore Federated Agent Access: Use MCP (via Starburst Galaxy) to allow agents to interact with multi-cloud databases without the need for expensive data consolidation.
Open Questions
- Will open standards like MCP become the universal connector between lakehouses and agents, or will proprietary agent integrations from Databricks and Snowflake dominate?
- How effectively can traditional data governance models adapt to unpredictable, multi-step agent behaviors without degrading agent utility?