Written by William Guo, CEO of WhaleOps
Translated by Debra Chen
After attending both Databricks Summit 2026 and Snowflake Summit 2026 in San Francisco this year, my biggest takeaway is that if you still see them simply as a company building a lakehouse and another building a cloud data warehouse, you are already behind the curve.
Both companies are moving upward in the stack. They are no longer satisfied with being the place where enterprises “store data and process data.” Instead, they are competing for a much more valuable position in the future of enterprise AI.
The most interesting part of this year’s events was that Snowflake invited Daniela Amodei, Co-Founder and President of Anthropic, to its opening keynote, while Peter Steinberger, the creator of OpenClaw, entered Snowflake Dev Day through the OpenClaw ecosystem. Databricks, meanwhile, brought OpenAI President and Co-Founder Greg Brockman directly onto its keynote stage.
Is this simply the convergence of Data + AI? I don’t think so. What I see is a strategic exploration between model companies and data platform companies around one fundamental question: who will own the entry point to enterprise AI in the future?
Will the primary gateway to enterprise AI belong to large language model Agents, or will it remain in the hands of enterprise data platforms? What will the future trend look like?
I would like to share some of my observations and insights.
Data Platforms Are Evolving Upward into the Core Hub of Enterprise AI
1. Looking Back at the 2025 Summits: Get Data Ready for AI
At the 2025 Snowflake Summit and Databricks Summit, both companies focused heavily on the theme of “Get Data Ready for AI.”
Snowflake’s 2025 announcements around Snowflake Intelligence, Semantic Views, Openflow, Trusted AI, and business user-facing experiences were fundamentally centered on one question: the hardest challenges in enterprise AI exist beneath the model layer — data definitions, access control, business semantics, and governance boundaries.
Without these foundations being properly managed, even the most powerful models can only deliver limited value.
The overall philosophy behind Snowflake’s 2025 Summit was clear: how far enterprise AI can go ultimately depends on the maturity of enterprise data governance. The stronger AI models become, the higher the requirements for enterprise data management.
In the past, poor data governance might have resulted in teams simply debating whose report numbers were correct. In the future, when Agents make incorrect decisions or execute wrong actions, the consequences will no longer be a disagreement over metrics — they could become production incidents and business-impacting events.
Databricks took a more engineering-driven approach in 2025. When looking at products such as AI/BI Genie, Databricks One, Databricks Apps, Lakebase, Agent Bricks, Lakeflow, and Unity Catalog together, the underlying goal becomes clear: bring data, analytics, AI, applications, and transactional capabilities together within a unified platform as much as possible.
Databricks’ overall perspective in 2025 was also highly practical: enterprise data, applications, and transactional systems have historically existed in silos. Data has been repeatedly copied between systems, definitions have become inconsistent, and business states have become difficult to synchronize.
Once AI enters the enterprise, these historical limitations will become major obstacles for Agents.
In my view, although Snowflake and Databricks followed different product paths in 2025, their strategic direction was aligned: the first step toward enterprise AI is building solid foundations across data, semantics, permissions, governance, and application connectivity.
2. The 2026 Summits: From Data Foundations to Agent Foundations
By 2026, neither company was satisfied with simply being a data infrastructure provider.
Snowflake repeatedly emphasized concepts such as Agentic Enterprise, Governed Data, Actionable AI, Horizon Context, and Agentic Control Plane. The company has shifted its vision from helping users gain insights from data toward enabling data-driven actions.
In the past, data platforms helped people understand data. Now Snowflake aims to enable Agents to directly execute actions within governed data environments and trusted contexts.
This is a major transformation.
Once AI moves from answering questions to executing business operations, capabilities such as permissions, auditing, identity management, compliance, rollback mechanisms, and accountability boundaries become critical.
Snowflake is positioning itself as the control layer before Agents enter enterprise production environments.
Databricks, meanwhile, is going all-in on becoming an Agent system platform.
Genie One, Genie Ontology, Genie Agents, LTAP/Lakebase, Omnigent, Agent Bricks, and Unity AI Gateway together form the infrastructure required for Agents to truly operate: context, semantic layers, runtime systems, harnesses, deployment capabilities, governance, and multi-Agent collaboration.
Snowflake is defining the rules that govern how Agents enter enterprises.
Databricks is building the operating environment where Agents perform work after entering enterprises.
One focuses more on the control plane; the other focuses more on the runtime stack.
3. This Year, the Product Evolution of Both Companies Shows Remarkable Similarity
Although Snowflake and Databricks launched different products and features, their evolution as data platform companies shows striking similarities in several key areas. In my opinion, these shared directions reveal the future trajectory of enterprise data platforms.
First, both companies are increasingly embracing a model-agnostic approach.
If you can integrate Claude, others can integrate Claude as well. If you can integrate OpenAI models, others can do the same.
The long-term competitive advantages for enterprises will not come from access to a specific model, but from their own data assets, business semantics, permission systems, governance capabilities, and closed-loop execution processes.
Second, both companies are pushing AI from answering questions toward taking actions.
Asking about reports, searching knowledge bases, and querying business metrics are only the first steps.
The areas where enterprises are truly willing to continue investing are whether Agents can participate in business processes, call enterprise systems, trigger actions, and operate under full governance and control.
Third, both companies are placing governance at the center of their strategies.
In the past, data governance was often viewed as a backend IT function. In the Agent era, it becomes a prerequisite for production deployment.
What data AI can access, what tools it can call, what level of actions it can execute, and how accountability is handled when something goes wrong — all of these require a comprehensive governance framework.
Fourth, both companies are competing for the new central hub of enterprise AI.
Snowflake wants to become the control plane, while Databricks wants to become the runtime stack.
The terminology may differ, but their strategic positions are increasingly similar.
Neither company wants to remain only a foundational data resource layer. Both are moving upward toward the layers of tasks, context, permissions, and action boundaries.
Data+AI or AI+Data? The Real Enterprise AI Battle Is About the Entry Point
What I saw at these summits was a preview of a larger battle over the future entry point of enterprise AI.
Model companies need enterprise data platforms, and enterprise data platforms need the strongest model capabilities.
The collaboration is real, but so is the competition for strategic positioning.
The history of enterprise software shows that enterprise-level entry points rarely coexist peacefully for long.
In the AI era, whoever captures user tasks, gains access to enterprise context, decides which systems are called, and defines the boundaries of Agent actions will control the primary entry point of next-generation enterprise software.
Model Companies and Data Platform Companies Are Moving Toward Each Other
Anthropic and OpenAI must move deeper into platforms like Snowflake and Databricks.
A foundation model without enterprise data, permission systems, business semantics, audit trails, and complex organizational boundaries cannot truly operate inside an enterprise environment.
No matter how intelligent the model becomes, it still needs the enterprise ecosystem to support its operation.
At the same time, Snowflake and Databricks also need Anthropic and OpenAI on their platforms.
Enterprise customers are pursuing the most advanced intelligence capabilities, and platform companies need leading models to unlock the value of their data and governance systems.
Short-term collaboration makes perfect sense, but long-term boundaries will inevitably create friction.
Model companies will not be satisfied forever with being only intelligence providers.
Data platform companies will not be willing to become merely the data warehouse behind foundation models.
Two Possible Paths: Agents Growing from Data Platforms or Agents Consuming Enterprise Data
The future of the Agentic Enterprise may follow two different paths.
The first path is Data + AI: building AI Agents from enterprise data foundations.
This is the more natural direction for Snowflake and Databricks.
Enterprises first organize their data, permissions, contexts, semantic layers, governance rules, workflow interfaces, and accountability boundaries. Then Agents grow from this structured enterprise environment.
This approach is slower. It requires more foundational work and involves many difficult tasks that are not always visible in impressive demos.
However, it aligns better with the reality of enterprise production systems.
At the 2026 Summit, Snowflake emphasized concepts such as “Your Data Is Your Competitive Moat,” Governed Data, and Control Plane.
Databricks highlighted Genie Ontology, Unity Catalog, Agent Bricks, Omnigent, and Unity AI Gateway.
Although their product strategies differ, both companies are emphasizing the same fundamental idea:
Enterprise Agents should grow from an organization’s own data governance framework and business operating model.
The second path is AI + Data: general-purpose AI Agents moving downward to consume enterprise systems.
OpenAI and Anthropic naturally represent this direction.
The strategy is to first build the most capable general-purpose Agent, become the primary user interface, and then connect downward to enterprise data, APIs, SaaS applications, workflows, and execution systems.
If this approach succeeds, the most valuable enterprise entry point in the future may belong to model companies.
The difference between these two paths lies in the order of development — and ultimately, in control.
One approach establishes enterprise order first, then grows intelligence from that foundation.
The other establishes intelligence first, then absorbs enterprise systems and processes from the top down.
Task Distribution Rights Are the Real Entry Point
Because the true value of an entry point is not simply about who speaks to users first. It is about who controls the distribution of tasks.
In the future of enterprise AI, the most valuable asset will not be a chat interface. It will be task distribution rights.
Whoever receives user tasks first, captures enterprise context first, decides which systems should be called, defines what actions Agents are allowed to perform, and controls the execution boundaries and workflows will define the primary entry point of next-generation enterprise AI.
Snowflake and Databricks are aggressively moving upward because they understand that if they remain only data platforms, they may eventually be abstracted away by the AI layer above them.
Anthropic and OpenAI are aggressively moving into enterprises not only because they want to sell more tokens. They want to become the first recipient of enterprise tasks.
Today, many people still evaluate enterprise AI from a relatively superficial perspective: who integrates more models, who builds more demos, or whose Agent appears more intelligent.
But the deeper question is:
Where will the center of power for future enterprise software actually exist?
Will it gradually emerge from the underlying layers of data, governance, context, and workflows, where controlled Agents are built on top?
Or will general-purpose AI Agents at the model layer become the new center and reshape how enterprise software is accessed and operated?
Once you understand this question, you will realize that the keynotes at these summits were no longer simply product announcements.
They were early attempts to define the future power structure of enterprise AI.
My Conclusion: The Organizational Model of AI-Native Enterprises (Agentic Enterprise) Must Be Rewritten
Today, Databricks, Snowflake, OpenAI, and Anthropic are not competing merely over who can provide enterprises with another AI tool.
They are competing over who will define the future organizational model of enterprises.
In the past, the fundamental units of enterprise operations were people.
A salesperson, a finance specialist, a data analyst, or an implementation consultant — each role had its own responsibilities, permissions, workflows, KPIs, and accountability boundaries.
Enterprise software was designed around these human roles, which is why systems such as CRM, ERP, BI platforms, OA systems, ticketing systems, and workflow platforms emerged.
The fundamental transformation brought by the Agentic Enterprise is that the core nodes within enterprise processes are shifting from simply “humans” to “humans + Agents.”
In some scenarios, certain processes may even become fully Agent-driven.
This means enterprise architecture can no longer simply be the old system landscape with a Copilot added on top, nor can it be just another Chat-BI feature built into existing BI platforms.
Instead, enterprises must redefine fundamental questions:
Which tasks should remain human decisions?
Which tasks should be executed by Agents?
Which actions require approval?
Which workflows can operate autonomously?
How should data permissions be inherited?
How should accountability boundaries be defined?
Therefore, the future battle over enterprise AI entry points is not merely a competition over software interfaces.
It is a competition over who gets to define the structure of enterprise organizations.
Foundation model companies want general-purpose Agents to become new working units that move downward into enterprise data and processes.
Data platform companies want to grow controllable Agents upward from enterprise data governance, business semantics, and permission systems.
Whoever can define how these new organizational units work, collaborate, and remain governed will define the next generation of enterprise software.
The ultimate destination of the Agentic Enterprise is not that companies simply gain a smarter assistant.
It is that enterprises evolve from “human-driven processes” toward “processes jointly created by humans and Agents.”
The enterprise AI battle has only just begun.
About the Author: William Guo
William Guo is an Apache Software Foundation Member, Co-Chair of the DataOps Forum at ApacheCon, Open Source Innovation Committee Member of the United Nations Consultative Organization of the China Association for Science and Technology, Open Source Committee Member of the China Communications Industry Association, Data Intelligence Committee Member of the China Software Industry Association, DataOps Expert at the China Academy of Information and Communications Technology (CAICT), and PMC Member of Apache DolphinScheduler and Apache SeaTunnel. He is also the founder of the Chinese ClickHouse community.
William Guo graduated from Peking University, where he studied under Professor ShiWilliam Tang, a leading scholar in data warehousing.
With more than 20 years of experience in data warehousing and big data technologies, he previously served as General Manager of the Data Department at Wanda E-commerce, Head of Lenovo’s Big Data Platform, CTO of Analysys, and held key big data leadership roles at Teradata, IBM, and China International Capital Corporation (CICC).
As an Apache Foundation Member, William Guo founded two Apache Top-Level Projects — Apache DolphinScheduler and Apache SeaTunnel — and built China’s ClickHouse open source community from the ground up.
He has spoken at multiple international DataOps conferences and made significant contributions to research and practice in the big data ecosystem.
William Guo is also the CEO of WhaleOps.
His mission is:
“Making data easier and more efficient for everyone to use.”
Top comments (0)