Enterprise software as we know it is being rewritten by data and AI.
In the latest installment of our series on how technology leaders navigate generative AI, Databricks CIO Naveen Zutshi shares how GenAI is transforming everything from development and security to organizational design. As he puts it, AI won’t just enhance software, “data and AI will eat software”: a shift that’s redefining how enterprises build, govern, and scale.
Asheem Chandna: How has GenAI changed how data companies operate?
Naveen Zutshi: Even before GenAI, data companies relied on ML and NLP for forecasting, churn models, and fraud detection. We’ve long applied ML to data cleaning and wrangling, but making data usable for insights has always presented a challenge until now.
The ability to build solutions more quickly with GenAI has transformed both engagement and intelligence layers. Agentic systems now handle discovery and insights, replacing rigid RPA bots with flexible probabilistic tools. And while reliability remains a challenge, the potential is massive. Organizations also see improved accuracy and better outputs when they build intelligent AI systems informed by their proprietary data, or what we call at Databricks data intelligence.
Where has GenAI reliably improved systems, and where is it still a work in progress?
GenAI is most effective where evaluation criteria are clear, like software development. At Databricks, tools like Copilot and Cursor are widely used. We track usage, stickiness, and code volume, and the numbers are way up. The underlying tech has improved a lot, and we’re actively training people on prompt engineering and white coding. As a result, we can expect to see more people managing teams of AI developers.
Sales, on the other hand, is more mixed. Top-of-funnel efforts, such as campaigns and lead scoring, are improving quickly, but the middle of the funnel proves to be a greater challenge. Data is sparse, quality varies, and success is harder to measure; therefore, reliability is still low.
That’s why we’ve focused on trust and building enterprise-grade GenAI built for reliability. If you’re replacing an RPA bot in an order-to-cash process, you can’t afford errors. It has to be accurate, auditable, and safe.
How advanced are your customers in using AI, and what’s your advice for those just starting out?
Let’s start with advice.
So, a lot of experiments never make it into production. It used to be 1 in 10; now it’s closer to 1 in 3. Still, the ratio is high.
Regardless, my advice is to not stop experimenting. Experimenting is how we learn. What’s important is to involve the business early and focus on the data. Most enterprises don’t have internet-scale volumes, so data quality, access controls, and governance matter even more. If your foundation is solid, the results will be too.
Customer maturity varies widely, from early-stage to highly advanced. Take MasterCard: they built a governed digital assistant that streamlines customer onboarding via chat and learns from feedback. Or Crisis Text Line, a nonprofit that handled 1.3 million mental health texts last year. They now use GenAI to simulate training conversations, cutting training time and expanding their reach. PicPay is another great example. As Brazil’s first fintech super-app with over 60 million customers, PicPay used Databricks to unify its massive data ecosystem to overcome scaling challenges and enable the company to leverage and build with AI, all while reducing IT operations costs by an incredible 20%.
As a CIO, how far along are we in the development of GenAI and agentic AI for enterprise use?
We’re still in the early innings. A year ago, agentic AI was just starting to get on our radars while we focused on things like RAG. Now, with innovations like MCP servers, we’re seeing new ways to connect agents and apps, opening the door to a significant shift in enterprise AI over the next two years.
What might that shift look like?
So far, we’ve focused on development, but much of what surrounds it is still manual.
Take voice of the customer: call transcripts can now be turned directly into product feature requests. GenAI can speed up product marketing, improve testing, and even help automate deployment, which still relies heavily on traditional SRE methods.
That’s just one workflow. Imagine applying that level of innovation across every workstream.
In finance, for example, ad hoc reporting used to require a data analyst. It could take weeks. Now, with Databricks Genie, anyone can ask a question in plain English and get answers, dashboards, and even full data rooms instantly. No data analyst required—you become the data analyst yourself.
What are the biggest obstacles enterprises will face in scaling GenAI?
Reliability is one piece, but security and governance are just as critical, especially when it comes to securing agents and preventing the exfiltration of data.
How do you protect PII and PCI data? How do you prevent exfiltration, especially when agents outnumber human users? Securing agents, preventing hijacking, and ensuring they act safely on behalf of humans will be a significant challenge.
From a governance standpoint, you need to audit what those agents did. For things like SOX compliance, it’s essential to show exactly what actions these agents took.
I expect we’ll see an entirely new wave of companies to solve these issues.
What has Databricks’ experience been like working with startups in the AI space?
It’s been fantastic.
We work with a range of startups, including Replit and Cursor, and we recently acquired Neon, a Postgres database company. Neon’s founders told us that agents, not humans, now create three to four times as many databases on their platform, often triggered by Replit. That’s agents building full-stack apps, including infrastructure, autonomously.
Cursor is another example. They’ve scaled rapidly without adding headcount at the same rate. These GenAI-native companies are growing exponentially.
Why is GenAI enabling that kind of growth without more people?
It’s automation. GenAI handles repeatable tasks that humans used to complete manually. For example, at Databricks, we developed an AI field assistant that summarizes and can fully complete tasks for 4,000 of our sales reps, like providing 360° insights and drafting emails that can then be edited if needed, so the team can focus on strategic initiatives.
Startups also have the greenfield advantage. Without legacy systems, they can build modern architecture from day one and focus limited human capital where it matters most.
And GenAI products often improve with use. More users mean more data, better performance, and a flywheel effect that fuels growth without linear hiring.
Are startups building for broad markets, or specifically for Databricks users?
Both. Through our Built on Databricks program, some startups build their entire business on our platform. With core infrastructure already in place, they can move fast and avoid reinventing the wheel.
Using architectures like Delta Lake and the Lakehouse helps them scale efficiently.
What advice would you give startups on which problems to solve next?
There are numerous opportunities in AI infrastructure. Bridging the dev-to-deployment gap, or translating voice of the customer into product feedback, remains difficult. Startups can solve that.
Reinventing traditional security with GenAI is a major frontier: authenticating agents, seamless DLP, or building a Gen-AI-native SIEM.
How should CIOs and governance teams manage GenAI experimentation? Do you focus on ROI or let ideas emerge freely?
We don’t emphasize ROI at the experimentation stage—chasing ROI early is a trap. People can fudge the numbers. Instead, we ask: Is it a good idea? Is it practical, secure, and feasible within our governance framework?
Our model is: experiment at the edge, scale at the center.
What’s the next most promising area for AI expansion?
I’m excited about how applications and agents interact, especially via MCP protocols.
We’re moving from rule-based systems to outcome-based systems. Instead of coding every rule, you define the outcome, and an “uber agent” coordinates with others—sometimes autonomously and sometimes with human help—to deliver it.
Decision-making is another area. Executives are overwhelmed by data, and GenAI can synthesize insights, reducing that burden.
And organizationally, we’ll likely see fewer middle managers. With more autonomous systems, employees can do more independently, and leaders can manage larger teams, flattening org structures without losing productivity.
Can you give an example of how AI could replace traditional middle-management functions?
Middle managers have long served as translators, conveying leadership goals to the front line and reporting back insights from the ground. But that communication layer is flattening. Leaders can directly access ground-level data, and innovation from the edge can rise quickly. That kind of bi-directional visibility reduces the need for interpretation layers and accelerates innovation in areas like product and sales.
If you were advising a public company board on GenAI, what would you say?
Boards today need an AI evangelist or a data and AI expert.
At Databricks, we believe every company is becoming a data and AI company. Traditionally, software was a set of deterministic rules coded by humans. Now, AI learns from data and generates software dynamically, a fundamental shift in how software is built, deployed, and used. In that sense, data and AI will “eat software,” and tomorrow’s leaders must be AI-first. That has major implications for how boards govern, how they challenge management, and how they think about long-term strategy.
What does that mean for the future of Databricks products?
We’re infusing AI into everything, from indexing to query planning to clustering in data warehousing. The goal is to make our products smarter and more adaptive, so they improve continuously based on usage.
Final question: What advice would you give startup founders working with enterprises?
Find the business problem that’s top of mind for the executive. Do your homework, challenge their current approach, and show how your product uniquely solves that issue.
Be proactive about security and governance; it builds trust.
Great founders lead with product, not a pitch. A crisp, two-minute demo that shows immediate value goes a long way.
And don’t chase every deal. If a request doesn’t fit your product vision, walk away. Staying focused is key to building something great.
