Meet Snowflake Intelligence: A Personalized Enterprise Intelligence Agent with Sridhar Ramaswamy
The AI Handover
Eighteen months ago, the torch passed from Frank Slootman to Sridhar Ramaswamy. It wasn't just a change in leadership; it was a fundamental pivot from being the world’s best data warehouse to becoming the definitive AI Data Cloud.
"The company was a little slow to reacting... Frank felt presciently that we were headed into a time that was a lot more tumultuous."
— Sridhar Ramaswamy
The market's reacted incredibly well lately, but it’s been a journey. Walk me through the first six months. How do you re-orient a rocket ship?
Snowflake had specialized at every layer. There were seven to ten layers between an engineer and the customer. That works when you have perfect product-market fit and you're just driving a truck through the win. But in AI? We can't even tell what’s coming next month.
The Dictum: Speed Wins
Accountability
Organized into distinct areas like AI or Core Warehousing. No more "distance" between the code and the customer.
Iteration Over Strategy
"Ability to iterate always trumps carefully laid out strategies." Realizing gain requires a high-frequency feedback loop.
Finding the Sweet Spot
We are not a CSP. We are not a foundation lab. We are the AI Data Cloud.
The Humility to Pivot
Early last year, we actually built a credible MOE foundation model. But we had to look hard at ourselves.
"We simply did not have the capital to compete with OpenAI or Anthropic. So we asked: How does AI accelerate the data already in Snowflake?"
Half of the Fortune 2,000 have their most valuable data on Snowflake. The strategy became systematic: Invest in search, invest in text-to-SQL, and add value to what people are already doing.
Snowflake Intelligence
An "opinionated" agentic platform designed to kill the 2D dashboard.
A lot of agentic platforms from CSPs offer an infinity of options—you can bring data from anywhere, do any workflow. But when you can do anything, it's hard to figure out what you *should* actually do.
Snowflake Intelligence (SI) is focused. We started internally with a tool called "Raven"—a Sales Data Assistant. Instead of staring at ten different dashboards, our sales team just asks questions.
"A dashboard is a 2D view of a complex surface. It has no easy answers to the many questions a reasonable person will have. We wanted to free people of that style of thinking."
Early Adopters
- • Cisco
- • Fanatics
- • USA Bobsled Team
The Vision
"Not a general purpose agent to do it all, but a foundation to realize value from data really quick."
Next: How do we actually build these agents? Sridhar breaks down the Snowflake Intelligence user experience and where we draw the line between the data, the agent, and the app.
Democratizing the Data Cloud
Moving beyond the SQL barrier to reach every single employee.
"Our aspiration was for this product to be used by every single employee in the company. It’s not for people who can write SQL."
Building on our evolution into an AI-integrated platform, the user experience of Snowflake Intelligence (SI) is designed to break the traditional dashboard bottleneck. Historically, Snowflake was a tool for the data team—they’d build the dashboards, and everyone else would just look at them. We’re flipping that script.
Imagine an interactive interface where you don't need to know where the data lives. You ask questions. You use "Raven," our Sales Data Assistant, to prep for a meeting in seconds. What’s our relationship with this customer? What does their consumption look like? Any outstanding tickets? It's about breadth of value.
The "No YOLO AI" Policy
We place a massive emphasis on trust. I tell my team: we need to think of AI the same way we think about software engineering. There is a right and a wrong answer. We cannot operate in a "YOLO AI" mode where you get a mix of good and terrible answers and it's the user's problem to figure out which is which.
New Motion
Identity provider integration means no more individual Snowflake accounts for every single user. It’s seamless.
Model Health
If you change the model, you must run an eval. Don't blow up what's already working.
"The line between agentic systems
and software is going to be bloody."
Interviewer
How do you draw the line between a data system and an application like Salesforce or SAP?
Sridhar Ramaswamy
I tend to be completely emotionless about where the strongest current is. We aren’t pretending to be SAP. But if my sales team needs to update Salesforce, why shouldn't they do it via an agentic API? If you need to file for vacation on Workday, our HR agent should handle that. We are being opportunistic—operating from a position of value, not just naked ambition.
The Speed of Transformation
Leadership Alignment
We rolled out clearer accountability and leadership changes within a few quarters. It's easier when you're dealing with a smaller group at the top.
The Pod Model
We integrated product, engineering, and go-to-market functions into "pods." Small groups moving fast without disrupting the entire ship.
Bottoms-Up Adoption
Our founder, Benoit, fell in love with coding agents. That did more for engineer adoption than any top-down memo from me ever could.
Next: How Sridhar’s past as an investor and entrepreneur
built the foundation for this high-velocity leadership. →
The Accretive Leader
Transitioning from the precision of a PhD and the scale of Google to the "heartbreaking" hustle of Neva. How every failure and abstract-writing session built the CEO Snowflake needs today.
The PhD Years
"A waste of time?"
It teaches you to focus on ideas and convey them crisply. Spending hours on a four-line abstract is actually a superpower for high-scale leadership.
The Google Era
"Immediate Scale"
My first launch—one person, three months—was in the New York Times. At Google, you take distribution for granted. You forget what it's like to fight for air.
The Neva Hustle
"Heartbreaking but Vital"
A startup makes you realize success isn't a given. I learned more about marketing, social, and "the hustle" than I ever could in a corporate cocoon.
"Thank you for inviting me to your rock concert."
— Recalling the two-block line at the Javits Center.
Honestly, both the Google and the Neva experiences make me a lot more grateful for my job today. We talked earlier about dealing with the "stuff" you don't want to deal with when running something big—I’m much more gracious about that now. It is a privilege to be at Snowflake, having this kind of impact.
When I saw that line of people waiting to get into our conference in New York, I realized: there's nothing ordinary about this. I don't take the platform for granted anymore.
Product-Market Fit is Magical.
I think product-market fit continues to be magical. It’s the only reason Snowflake exists. Think about it: the three hyperscalers would love to own the data space entirely, yet here we are. There’s Snowflake, there’s Databricks. That redeeming value is unique, and we should have humility about what it takes to create that "lightning in a bottle."
"I joke to people that these are like empires that have not met their oceans just yet."
Look at OpenAI or Anthropic. They are in a phase where they don’t think they *can’t* do anything. But you have to pay attention to what is likely to be in their immediate path. Coding agents, for example? They are laser-focused on that. If your "startup" is just a set of prompts on top of their models, you're in a problematic space.
Innovation Paranoia Index
Hyperscalers have infinite budgets and patience. Our only defense is staying—not just being—ahead.
The "Right to Win"
Google stopped at information. We tried to get into shopping, travel, hotels... we didn't really succeed because we lacked the core competence beyond the world of information. Discovering where that boundary lies for OpenAI and Anthropic will be the defining story of this era.
The Orientation
"We need to continue to earn it. No software company can feel secure about their position in the sun right now."
As we look toward Snowflake’s strategic positioning and our upcoming partnership strategies, this mindset of constant innovation isn't just a corporate slogan—it's a survival requirement in a world of infinite-budget competitors.
The Architecture of Defensibility
Moving beyond the "unassailable" software moat. Why Sridhar believes defensibility in the AI era isn't a strategy you draw on a whiteboard—it’s a muscle you build through daily execution and higher levels of abstraction.
I’ve seen this play out: even the giants feel the ground shifting. When the technical environment is this fluid, you realize that defensibility isn't a static wall; it’s momentum.
At Snowflake, our core strength is the data platform layer. I tell my team—and our customers—that we are there from inception to insight. Think about the difference between the 90s and now. If you built Adobe Photoshop in 1995, you did research, burnt CDs, and waited months for feedback. Data was a slow afterthought.
But look at Google or Meta. They aren't just product companies; they are data-first companies. The moment you interact with a search ad, that behavior feeds back into the system within minutes. That’s the magic we want to bring to every enterprise.
The Feedback Latency Shift
"AI is a massive accelerant because it elevates the value of data. It’s no longer about just running queries; it’s about influencing how your entire business operation functions in real-time."
"Defensibility is BUILT,
not strategized."
The Satya Lesson
Learning from Microsoft: How to create winning partnerships—and how to adjust them when the competition gets real.
SAP + Snowflake: A bidirectional data share that turns 1+1 into 3.
The Multi-Cloud Reality
Azure + Snowflake is a strict positive for the customer.
Platform Maturity
Moving from "Snowflake-centric" to a collaborative ecosystem mentality.
With the data foundation laid and the partnerships secured, the conversation shifts to the ultimate metric of the current era: The ROI of AI.
Moving from the high-level ecosystem of partnerships, the conversation shifts to the pragmatism of the P&L. For every enterprise leader, the question isn't 'if' AI, but 'where' the money is.
The ROI
Stack Rank
Sridhar breaks down where the immediate value lies—and where companies are over-investing too early.
1. Coding Agents
"The easiest ROI. It's about making projects faster and demystifying technology. We use them heavily at Snowflake to make Snowflake itself easier to navigate."
2. Support Repositories
"AI effortlessly taps into human knowledge bases. It handles voice and text seamlessly, with a human backup if it fails. It's a guaranteed win."
3. Democratized Data
"Moving away from $50/user/month licenses toward seamless data access via tools like Snowflake Intelligence. That is the next frontier."
"You should not spend a lot of money on AI with Snowflake. You should do it a thousand bucks at a time."
— Sridhar Ramaswamy
Don't Take the 100-Foot Step
I place a lot of emphasis on how many shots you take on goal. Obsessing about massive ROI too quickly is a trap. You don’t want your first step to be a 100-foot leap; you want to do a lot of little things that prove value.
Look at our Sales Data Assistant. It didn't just appear. It was the result of three versions that came before it. We started with basic enablement, moved to customer information, then built a Python app called Customer 360. Only after those iterations did it culminate into the Sales Assistant.
I prefer not to take big bets. Iterating and creating value every step of the way is the key. When you have significant value that you feel good about, then you wrap it up.
The Future of Ads & Agency
"What happens to the '10 blue links' when we move to direct chat interfaces?"
— The Host
The ad model is here to stay. It’s an incredibly powerful medium, but it will reinvent itself. My fear? That it becomes more insidious. Imagine a digital psychiatrist who has a subtle, undisclosed affinity for prescribing one medication over another because of an ad deal. That’s the creepy territory we have to watch for.
But there’s a silver lining. We are seeing a massive premium on citations and sourcing. The fact that people want to go back to primary sources, even with AI reasoning, is a positive indicator. At Neva, we were obsessed with citations in 2023; today, that relevance is stronger than ever.
"You and I can get an expert paper literally on any topic in seconds. We just have to have the brainpower to digest it. That’s pretty amazing."
If AI can generate expert papers, the next question is how it verifies that truth. Next, Sridhar dives into why LLMs, for all their power, still desperately need Search.
Beyond the Model: Why Indexing Still Matters
After dissecting the ROI of AI and the shifting sands of the ad model, we arrive at the architectural endgame. Is traditional search dead, or is it the secret weapon LLMs are missing?
"There is a contingent of folks that believe traditional information retrieval and indexing is less and less relevant as more data is available through the model... How do you think about that?"
— SarahThe PageRank Era
Harnessing the entire internet to figure out what was popular. It was the original "vibe check" for the web.
The "Juice" Trajectory
Note: PageRank "ran out of juice" around 2004. The feedback loop of click behavior became the real engine of value.
The "Smart Person" Approach
It is tempting to trivialize search as just "retrieval." But think about the "Math Problem" in AI.
Should LLMs be able to do math? A maximalist would say yes. But a smarter person is going to say no. They shouldn’t "do" math; they should write the two lines of Python needed to solve the problem and then run it.
I think of Trust in a very similar way. There are well-known, reliable solutions for figuring out what is most trustworthy when answering a question. Why would you not use that as a tool?
"You cannot be so smart that you don't use the computer. There's no bravery in just like hard work. If something can be done easily, you get to focus your energy on other things."
"Maximal intelligence will use
reliable tools wherever
they are available."
The Pragmatist's Manifesto
Sridhar’s point is clear: Search APIs aren't just legacy tech; they are the externalized memory models benefit from because they provide information that is not easily "internalizable" yet. As we close this chapter, the future looks less like a single monolithic model and more like a highly intelligent agent reaching for the right tool at the right time.
Next Chapter
Conclusion & Credits
Thank you for listening to No Priors.
Subscribe for a new episode every week.
Transcripts and more at no-priors.com