Everyone wants AI that works like magic. But Sridhar Ramaswamy wants AI that works, period.
“AI should not be a Big Bang,” the Snowflake CEO told me in a sit-down. “It should be a series of little projects that show value every step of the way.” But as Ramaswamy noted, while that may sound like caution, it’s actually strategy.
In the interview, Ramaswamy laid out a simple but radical roadmap for enterprise AI. “Don’t start with flashy demos or massive model investments,” he said. Start with data. Start small. Prove value. Then build.
The Agentic AI Hype — And The Hidden Work Beneath It
Ask 10 vendors to define “agentic AI” and you’ll likely get 10 different answers. But when I asked Ramaswamy what he really thought about agentic AI, his response was that we must move past semantics into doing actual work that makes AI work indeed.
What Ramaswamy sees is the growing desire for AI that not only retrieves and summarizes, but acts. From automating pre-meeting research to updating internal systems, agentic AI promises to reduce the time humans spend stitching data across platforms. But that only works if the data is accessible, connected and trustworthy in the first place.
“Step one is making information easier to access,” he explained. “Step two,” he continued, “is letting models decide what to pull. Step three is chaining those components together. That’s where the orchestration begins.”
Still, he warned that enterprises can’t skip the groundwork, as that would be a costly mistake. That’s one important message that industry experts are starting to propagate across the industry today, especially since it’s easy to think of AI as a magical wand that just makes all your problems disappear.
But like a car, AI only goes where the driver wills its wheels. In this context, business leaders sit in the driver’s seat and must figure out how to cut through the hype and get real value from AI. As Loganandh Natarjan noted in an op-ed titled “Generative AI is not a magic wand, it’s a strategic tool” on YourStory, “the potential of AI can be tapped only when it is thoughtfully merged into the very core of the organisation’s functions.”
The Single Biggest Mistake In Enterprise AI Projects
As companies scramble to keep pace with AI trends, many make a costly misstep: starting with the model instead of the mission.
“A lot of folks went out and bought GPU capacity or model licenses without thinking about where that’s going to create value,” Ramaswamy noted. As I wrote before in a previous article, that’s a fast track to disappointment.
His solution? Rather than start with scale, start with your customers’ needs.
Snowflake’s own internal example — a lightweight chat interface for its sales enablement content — is a case in point. “It didn’t cost a lot of money to build,” Ramaswamy said, “but it’s getting a lot of use. That told us we were onto something worth growing.”
AI Is Only As Smart As The Data It Sits On
The phrase “AI is only as good as its data” gets repeated often. But what does that actually mean for the modern enterprise?
At Snowflake, where more than 100 SaaS apps are in use across the company, the answer is that unless your data is unified, it’s practically invisible. What that implies is that you can’t successfully deploy AI or extract actual value from your AI projects.
As Ramaswamy told me, you can’t even run a proper dashboard without integrating data from different sources — like Workday, Google Calendar, Qualtrics, or CRMs like Hubspot and Salesforce. “And if you can’t run a dashboard, “ he added, ”you definitely can’t build a useful AI application.”
The challenge is deeper than business intelligence, according to Ramaswamy. Most external tools like ChatGPT or Gemini have no access to a company’s internal systems. They can’t pull consumption metrics or sales rep activity unless those systems are centralized and accessible.
“That’s why data readiness isn’t just a technical project,” he noted. “It’s the foundation of whether your AI investments will even work.”
The SaaS Model Is Being Rewritten
Ramaswamy believes that AI will redefine how SaaS tools function at a core level.
“Most SaaS applications were built to help humans be more efficient,” he explained. “But the future is software that can actually handle a good chunk of the work itself.”
That shift — from decision support to decision execution — is why BI tools, dashboards and even customer support platforms will evolve rapidly. As natural language interfaces mature, the number of people who can directly query business data will expand beyond analysts and data teams.
“This technology will let anyone who understands the business ask questions,” he said. “That’s a big change.”
The Most In-Demand Skill Isn’t Technical
When I asked him about what roles or skills will be most valuable in the next 18 months, Ramaswamy did not point to coding or data science, which was surprising as those particular skills are often on the list of the most in-demand skills for the AI era.
Instead, he talked about malleability — the mindset to experiment, stay curious and question AI’s output.
“It’s the ability to understand what’s possible and what’s fanciful,” he said. “To try new things, but also to be critical when something doesn’t look right. That’s more important than any single technical skill.”
It’s also how Ramaswamy stays grounded. He still tests AI agents personally, building simple use cases just to keep his intuition sharp.
“You need to live and breathe this stuff,” he noted. “It’s the only way to separate hype from reality.”
The Data And AI Platform Era
As Snowflake doubles down on being an end-to-end data and AI platform — not just a warehouse — Ramaswamy sees clarity in its role.
“In a world where AI is thriving, Snowflake will thrive,” he said. “Because we are the layer underneath that powers this data access.”
The future may belong to agentic AI, outcome-first SaaS and open-source pressure on inference pricing. But none of that matters if enterprises can’t get their data act together. The AI promise begins — and sometimes ends — with what you feed it.