Three truths about enterprise AI

It’s no secret that AI is accelerating innovation. In a very short amount of time, we’ve moved from the foundational building blocks of LLMs and RAG workflows to the advent of specialized AI agents. 

Every enterprise is talking about AI adoption, and it’s not hard to understand why. These agents are creating a competitive advantage for the enterprises deploying them, driving greater productivity and improving efficiency. In the near future, they will run the enterprise, creating massive opportunities. But not every enterprise is benefiting equally. While some are surging ahead, others are stuck in the planning stage. This speaks to an AI implementation gap at the heart of enterprise AI. 

To understand what’s driving this, you need to follow the data and realize that data access is the new frontier of AI.

AI has one essential truth: it’s only as good as the data that fuels it. That’s as true of AI models as it is of Agentic AI. For AI agents to achieve the critical mass of value that businesses demand–true Agent Intelligence–they need access to high-quality, contextual data across entire organizations. 

How do we solve this? I have three answers. 

 

1. Saas applications are dead as we know them; long live enterprise application data

For years, the Software as a Service (SaaS) model has been highly successful at navigating the interface between human business needs and the technological needs of data systems. For many people, their main day-to-day experience of both their organization’s data and the insights attached to that data is entirely refracted through the lens of these applications. 

What happens when the disruptor is itself disrupted? 

AI rewrites the playbook around how humans access, collaborate, and govern their data. The future is vendor-neutral AI Agents operating on customers’ data-centric architectures led by Agent Intelligence. 

Here’s how it will unfold. 

Business applications were designed for an old paradigm

Your enterprise data architecture is sprawling. To help make sense of this, business applications deploy user-friendly graphical interfaces (GUI) around what are essentially proprietary data silos. 

This made sense as long as the only users were humans. But with agentic AI, we’ve reached a tipping point. With AI, the data inside these applications is increasingly more important than the applications themselves. SaaS is dead because its primary mission no longer exists.

AI Agents make business application interfaces redundant 

Ironically, SaaS applications’ human-centric user interfaces used to be a functional benefit. But in the AI era, they provide little or no value to an AI agents. AI is being deployed in these applications to create more customer value than the applications can hope to achieve. These agents are able to access the same data using MCP and A2A without human interaction. 

SaaS data in the Agentic AI era

These agents will not be embedded in the data platforms themselves, but will cut across all data platforms, offering smart, personalized, conversational interactions. 

Most importantly, this isn’t happening in a distant future; it’s happening today, and it’s growing. McKinsey estimates that by 2030, 30% of enterprise work will be performed by agents, and Deloitte estimates that 50% of enterprise businesses will deploy agents by 2027. The result is superior insights with less direct human-to-application interaction. 

Starburst + Agent Intelligence

Starburst provides a unified data layer allowing AI agents to access multiple SaaS applications at once. We provide superior data access, data collaboration, and data governance – the three foundations of AI that I’ve written about before–allowing businesses to better discover, trust, and activate data. 

What’s more useful than accessing Salesforce data efficiently? The ability to access data from all SaaS applications simultaneously using AI agents. Imagine a unified access layer that provides true Agent Intelligence across all your data sources, regardless of where that data resides. 

Starburst is there today. 

Universal access will profoundly disrupt SaaS application environments, breaking open their data silos and opening them up to interoperable data access. 

 

2. The world’s most critical data is on-premises 

On-premises data is growing, and this growth will continue. Not only is this a good thing, it also reflects the ongoing needs of enterprise customers when adopting AI. In this context, on-premises data is actually an accelerant to the speed of change, lowering the bar for enterprise adoption of AI. 

Not what you expected? Let me explain. 

The death of on-premises data has been greatly exaggerated

Since the growth of the cloud, on-premises data has seemed like the dinosaur in the room. 

Not only is that not the case, the inverse is true: on-premises data powers some of the most critical data on the planet, and that role is only going to increase with AI. As Arthur Lewis, President of Infrastructure Solutions Group, Dell Technologies, March 2024, notes, “83% of the world’s data sits on-premises.” 

All of this means that on-premises data isn’t going anywhere. Far from dying, it’s actually growing faster than ever. This is the truth of our industry, and we need to accept it as both a positive and desirable outcome as we look to understand what enterprise businesses need to adopt AI successfully.  

Whole industries are on-premises for a reason

Why will on-premises data continue to exist? The answer is that the customers and businesses that use it require it. In fact, whole industries operate either entirely or mostly on-premises. These include some of the most important industries in the world: 

• Financial services
• Insurance
• Healthcare
• Public Sector 

On-premises by choice

Organizations in these sectors aren’t simply on-premises out of inertia or a failure to embrace the cloud. They’re there because it makes sense to be there. The data they handle is some of the most sensitive in the world, and they want to be able to access, manage, and govern that data in controlled and predictable ways. 

Towards on-premises AI

As part of this, they want ways to access their data wherever it lives. Increasingly, as AI takes hold, they also want a way to run AI on-premises too, using data and models that exist on site. Not only does this accord with the data compliance expectations and needs of their industries, it also makes sense for those businesses. 

That need isn’t going to go away. It’s going to keep growing as the datasets themselves keep growing. 

Starburst powers agentic AI on-premises and in the cloud

Starburst meets your data wherever it lives. Our hybrid lakehouse model is built using Icehouse architecture that leverages the power of Apache Iceberg and Trino. It connects the full power of the cloud alongside the security of on-premises solutions. More importantly, it works for both analytics and AI workloads, meaning organizations can begin with analytics as they build towards AI.  

As organizations in these industries embrace the power of AI, they will increasingly look for solutions that operate on-premises, too. On-premises AI will be a driving force in key industries, especially those operating in high-compliance environments. 

On-premises AI will be a major trend in the years to come because it fits customer needs.  

 

3. Knowing who really owns your data is going to become increasingly important 

AI brings new opportunities, but it also brings new rules and new risks. For those of us in the data industry, many of these rewards and risks are not new. AI merely shines a spotlight on old tensions between the value and speed of insight versus the potential cost of data governance. 

All of this means that data security and governance are about to become increasingly important for organizations of all sizes as they adopt AI. Seen this way, AI will be an accelerant. Data governance was already critical. Now, it’s essential. 

Bring your AI to your data, not the other way around

Ownership matters. Choice matters. It matters for analytics, and it matters even more for AI.

The ability to access data across multiple data sources, without moving it, is about to become more critical than ever. The ability to operate AI in a hybrid environment will make true AI data governance possible. In addition, data sovereignty will continue to grow as a requirement for many organizations. To satisfy demand and to allow organizations to operate AI compliantly and internationally, data governance will need to grow at the same pace that AI scales. 

For many organizations, this will not only make the difference between compliant and non-compliant AI. It will be the difference between projects that can go forward and those that become stalled or never get off the ground due to compliance and regulatory hurdles. All of this will be due to data access.  

This is where Starburst rises to meet this challenge. Let’s look at two trends in AI data governance. 

Air-gapped AI helps manage risk

Air-gapped AI brings the data compliance guarantees of on-premises data solutions to the cloud, using solutions like Google Distributed Cloud. Other on-premises solutions bring similar benefits. 

Overall, these solutions appeal to organizations needing data governance at a time when data compliance is becoming increasingly important. Like on-premises AI, this trend will continue to grow. 

Data products + AI

The other trend is data products. These curated data sets make data collaboration and data governance easy by applying attribute-based access control and role-based access control. This was already a great solution for analytics, and it’s an even better solution for AI. It provides the governance that organizations need when feeding data to their models, in a way that is both human-centric and intuitive. My prediction is that data products will continue to rise alongside the need for AI governance. For this reason, they represent the centerpiece of Starburst’s AI strategy. 

 

Starburst: A data architecture that fuels Agent Intelligence 

Where do these predictions leave us? We’re at an important inflection point, and it amounts to a holistic reevaluation of data architecture to take account of AI. As I’ve written before, I believe this will have three key characteristics: 

• Access
• Collaboration
• Governance

Analytics was already driving towards a consensus position on the value and power of data lakehouses. We see this constantly with our enterprise clients’ interest in Starburst’s Icehouse architecture. In fact,  we’re seeing it across every industry and every enterprise. And the two worlds–AI and data lakehouse are colliding. 

What’s more interesting than lakehouses for analytics? Lakehouses for AI workflows. Lakehouses that push build a seamless bridge from the data architecture of today to the AI of tomorrow.  

Lakeside AI: Data Lakehouse + AI

We call this concept Lakeside AI–AI powered by the data lakehouse—and we think it’s going to be the way that AI becomes achievable for enterprises already adopting data lakehouses for other reasons. 

Extending the lakehouse architecture to deliver AI has several advantages. First, it builds on the strength of Icehouse architecture. It uses Iceberg as the center of gravity for your data world, creating AI agents capable of securely accessing data from any data source and any SaaS application. 

This is AI as it was meant to be–helpful, insightful, and connected. It uses a dedicated Starburst Agent to help facilitate AI workflows involving both SQL generation and natural language processing (NLP). 

Everything you liked about data lakehouses for analytics, now for AI 

The beauty of this approach is how iterative it is. Many organizations were already making moves towards the data lakehouse for their analytics or data applications. Now, the same approach also works with AI. 

 

Going from BI to AI 

We are at an important tipping point for many organizations. These enterprises might have AI as a roadmap item, but haven’t been able to make these plans concrete because other technologies haven’t met them where their data lives. 

Starburst helps you approach AI architecture as you would any other data architecture project. Since very few organizations will be using only analytics or AI in isolation, a single data foundation can power both ventures. 

This isn’t just good data engineering, it’s good business management. It builds a sound foundation for AI tomorrow and long into the future. 

Join us for Starburst Launch Point

One thing is certain: Starburst is uniquely positioned to address each of these three truths for our enterprise customers, seamlessly connecting to data sources inside SaaS applications, on-premises, and across clouds. 

And we’re holding a Launch Point event to unpack all of the Starburst + AI story: 

• May 28, 2025 – EMEA/NORAM (Virtual)
• May 29th, 2025 – APAC (Virtual)