What prompted you to start Cast AI? What was your vision?

The idea started to surface when my co-founders — Leon Kuperman and Laurent Gil — and I were at our previous company, Zenedge, which was subsequently acquired by Oracle. As we were preparing for our quarterly board meetings, one of the line items that we had to highlight was the cost of the cloud. There was always a good reason why the cloud costs were growing: We were adding more customers. But the cost was still a very significant number. We embarked on many different routes to manually create efficiencies, which would produce temporary improvements. But then, everybody would get busy, and the improvements became less effective, while costs got even worse. We realized there was a huge opportunity to create an autonomous platform that could interact with an application the same way a human can, reasoning and understanding the application. It would be able to provide the infrastructure the application needed in real time, whether it was more CPUs, more memory, or other resources. It could determine if the existing configuration was optimal, and if not, it could reconfigure it automatically. At the same time, it would provide stability, better performance, and increased uptime for the application, reducing not only cloud costs, but also the human capital required to run the application.

What did it take to do this?

When we started, we had to make a bet on which cloud platform would be the dominant player in the future. It was 2020, and based on all our years of experience understanding cloud computing, we chose Kubernetes, or containerized applications. The world is moving to multi-cloud environments, and it's much easier to move a container across clouds than an application. Then, we realized that to build a platform that can think like a human, we had to build a model that could self-learn. We started by training the platform to look at different conditions and make decisions about the best outcome for an application at any moment in time. For the next two and a half years, we were heads-down building machine learning algorithms that could do that. Today, our algorithms are about three times smarter than a human. Three years from now, they will be 10,000 times smarter. Why? Because a machine is much more prudent in the way it makes decisions. Human beings apply emotions. Machines don’t. They look at application metrics and make decisions every two seconds. A DevSecOps engineer may know 50 different types of cloud systems, but there are more than 600 different kinds. Our automated system can look at all of them, analyze the different permutations, understand what inventory is available across the globe, and make decisions based on that to optimize for performance and cost.

Can you quantify the performance gains?

Our customers save up to 80% on the compute portion of things. We also have a database product where we increase performance by 30x. At the same time, we create stability in the database environment and reduce the cost, because you no longer need that many CPUs to manage the database.

That’s pretty remarkable. Do you think AI is overhyped? Underhyped?

I am a true believer that five years from now, it will be 10 times smarter. The speed at which the platform or the engines of what we are calling agentic frameworks can learn is simply unimaginable.

What are some of the biggest challenges you’ve had to overcome?

When we started, there were so many closed doors. The only reason you sometimes win is because you have experience, and you have seen something like this before. I can tell you, every five years, everything changes. People would say to me, “Yuri, you’ve built seven companies.” So? What I did 20 years ago doesn't count anymore. We are in a different world, with different technology. Everything is different. Every five years, you have to reinvent yourself. The only thing that doesn’t change is that you have to have a true belief that you will solve a problem. You just open one door, go to the second door, and so on. And it’s important for leadership to keep the team believing. That’s the hard part.

How do you instill that as a leader?

It is truly continuous work. You have to engage with the team on a daily, weekly, and even monthly basis, and instill belief. You need to focus on the little wins. And then little wins become more and more wins. And suddenly, you may not have a solution yet, but you see the light. You just have to find a way to get to that light.

You have argued that it’s critical for an entrepreneur to have direct experience with the problem they are trying to solve. Why?

If you don't live and breathe what you're trying to build, you’ll never get there. It’s just not possible. If you have internal knowledge of the subject matter, then you can start creating a thesis about what you're trying to build. Then you need to convince top performers to join you in making your crazy idea a reality. You need to be able to explain it to those people, and it’s very hard to convince them if you don’t have that direct experience with the problem. That’s why the vast majority of successful entrepreneurs build again in the sector they’re in. That’s because they can see an opportunity to fill a gap or to do something other companies are not doing.

As a serial founder, do you think entrepreneurship gets easier?

No way. It doesn't get easier. Technology evolves so rapidly, so when you’re starting a company today, you’re jumping off a much higher base. Sales are not the same. Technology is not the same. Marketing has evolved. Everything has changed. The only constants, I would say, are learning to be patient. You learn how to recognize issues much more quickly. You learn how to deal with human emotions much more sustainably. But still, you have to reinvent yourself. But I love entrepreneurship. I wouldn't be doing this after so many years if I didn’t.

Q&A

Yuri Frayman: Building application performance automation one step at a time

Seven-time founder and Cast AI CEO on why entrepreneurship never gets easier

The only thing that doesn’t change is that you have to have a true belief that you will solve a problem.Yuri Frayman, Founder & CEO, Cast AI

How serious is the problem Cast AI set out to solve?

First, if you look at your P&L, cloud costs are usually among the top five expense lines — top two for some companies, top three for others. The second issue is an application’s performance and stability. What are human beings doing? They take an application at a moment in time and create the infrastructure for it. But an application evolves continuously. As you add more users, the application’s needs change, and staying on top of that is humanly impossible. It would mean working seven days a week, 24 hours a day. So DevSecOps engineers optimize it for performance at a set moment in time, but that doesn’t help the application’s ultimate performance.

Performance automation can do that, while allowing the DevSecOps engineers to focus on higher-value problems. It reduces cost and opens up additional budget for other investments.

If you don’t live and breathe what you’re trying to build, you’ll never get there.Yuri Frayman, Founder & CEO, Cast AI

You’ve always relied on automation. Did generative AI help take it to the next level?

Yes. Before generative AI, we could automate app performance and how the app interacted with a database. Does it need more CPUs? Or different kinds of CPUs? Or more memory? Since we introduced a large language model, we can analyze all the metrics and suggest changes in the application’s source code to make it more efficient.

Generative AI allows us to create a test environment, where we can make the code changes, and test the hypothesis of how to improve that application. And it sends the results back to the developer for review. So now the application developer doesn’t have to spend time working on the maintenance of the application, and can be writing other code.

Feature

Variant Bio: Genomic discovery on the road less traveled

How Variant Bio is mapping highly differentiated data to accelerate drug development

Sōzō Pulse

Explore exclusive survey data

Track trends over time — filter by region, sector, and stage