theCUBE + NYSE Wired: AI Agent Conference | Itamar Friedman, Qodo

Clips
More from theCUBE + NYSE Wired: AI Agent Conference

Itamar Friedman

Co-Founder & CEO

Qodo

play_circle_outline AI agent conference in New York preview

play_circle_outline Qodo's responsible software for enterprise

play_circle_outline The Evolution of AI Code Writing and Acceleration of Technology: A Closer Look

play_circle_outline Expectations for AI agent conference in New York

play_circle_outline Revolutionizing Building Processes and Ensuring Safe, Responsible AI Code with Unified AI Agents

Info
Transcript

Itamar Friedman, Qodo

Itamar Friedman

Co-Founder & CEO Qodo

Exploring innovations in artificial intelligence: Insights from the AI Agent Conference 2025

In this in-depth discussion, Itamar Friedman, co-founder and Chief Executive Officer of Qodo, joins John Furrier on theCUBE for the NYSE Wired: AI Agent Conference 2025. Friedman shares expertise and insights on the evolving landscape of software development and AI, focusing on innovations set to redefine industry standards.

Friedman brings extensive experience as a leader in AI-driven software development to this engaging conversation. As the Chief ... Read more

explore Keep Exploring

Who is Itamar Friedman, what company does he lead, and why was he appearing on theCUBE at the New York conference? add

How does Qodo position itself in the AI/software market, and what role does it play in verifying the security, testing, explainability, and resilience of AI-generated code? add

How will advances in AI — including new IDEs, agent-driven tools across the software development lifecycle, and the prospect of advanced/superintelligent systems — change software development workflows and the role of human developers over the next few years? add

What developments can be expected in the next year, and how should teams prepare for them? add

What are you most excited about in the AI developer track, what developments should developers expect over the next year, and how should teams be organized to prepare (for example around code review, responsible/safe AI, quality code, and managing AI agents)? add

bolt Powered by CUBE AI

Itamar Friedman, Qodo

search

John Furrier

>> Welcome back everyone to theCUBE here in theCUBE Studios in Palo Alto, California for a special digital event previewing the AI agent conference in New York. Simon, and his community really kind of an elite set of diverse folks talking about what it takes to usher in the agent error. And that's going to be, obviously, it's all hyped up, but the real work is getting done from the folks in the industry to building the code, building the infrastructure, and kind of figuring out how AI will work. And we've got a great lineup here today. We got Itamar Friedman, who is the co-founder, and CEO of Qodo, which is doing an amazing responsible software with confidence. This is really changing the game for the enterprise on moving fast with great code, and moving the needle. Itamar, thanks for coming on theCUBE, appreciate you coming in all the way from Tel Aviv.

Itamar Friedman

>> Yeah, my pleasure. Thank you for having me here.

John Furrier

>> All right, so all the rage is on the Copilots, we've seen that a couple of years ago that ushers in code, coding is one of the top use cases right now where people are seeing immediate value, and again, seeing immediate value, and scaling it are two different things. That's the theme we're seeing in the industry. Take us through what you're working on, and what's different about what's going on now in the code world, because there is value, but there's a caveat, right? There's little gotchas. Share with us what you're working on.

Itamar Friedman

>> Sure. First, with your approval, I'll talk about on the market from a 30,000 feet. So, I think the first wave, or extensions were like assistants, small assistants that were helping developers in their ID where they write their code. The second wave that we saw, for example, with Cursor amazing success is that these IDs where you write your code were being forked, and rethought. The third wave, and that's by the way a higher level of automation because the entire entire environment development was rethought. And then the third phase is where not IDs are being forked, not the development environment. Rather, developers are being forked. That's one way it could happen. The other way that it could happen is that the agents, the AI agents are not going to live only inside IDs. Again, it's where developers write their code rather they're going to live throughout the entire software environment. That's the two option, probably it's going to be a mix. As you go from phase one to two to three, then you see higher level of automation. Now with that comes risk, risk that the AI code, the code that was written by the AI would have less review of human developers. Some of them have tens of years of experience avoiding outages that we saw last year, et cetera. And then we must also use AI on reviewing on software quality, and software development quality. And that's what we do at Qodo. Qodo is the red team for all of the other players in the market who are the blue team. The blue team are rushing to help everyone write more, and more line of code, Qodo is here to verify that that code work as expected with AI code review, AI testing, et cetera.

John Furrier

>> I love that whole cyber angle of red team blue team, because obviously in Tel Aviv it's very well known, cybersecurity is on everyone's mind there, and experts come out of there. But if you think about security, it's also the most active vertical around data, and resilience. So, you're getting at something that's really cool, which is moving fast, which is the speed of AI. We're seeing tons of acceleration, but the resilience is critical. Explainability, safety, these are paradigms that are security. I mean, in a way it's not necessarily a security product. They're principles.

Itamar Friedman

>> Yeah. So, I think the way we like to look at it, we call it integrity as a larger scope, a larger envelope for that. A big part of integrity of your software, of your code is security. But as AI write more, and more for code, we're also interested to know is it maintainable? Is it well tested? Is it according to our values? Think about the same feature that is being requested by a technical product manager, having AI completing a feature end to end can be developed in many ways. For example, one could be, yeah, I rewrote AI, rewrote the code completely, and actually on part of that, went to some social network, brought information into our company, and used that maybe as even part of the process of developing the feature. You don't know about it. So, it's not solely on security rather when AI is doing the human work, it's also the integrity of the code, and the process. And I think that's what we're seeing the growth of this sector security into integrity.

John Furrier

>> Yeah. I mean, integrity is a great point. I mean at the end of the day is that responsible AI is a topic, here you're just getting at safe, secure software. I mean it's not so much a security as a threat vector, which it can be for sure, but it's more of little things like whose license is it? Which open source license code is it? And that could really kill a software base. I mean if you've got some GPL in there, you never know. So, again, I'm going to bring up a weird example, but that's an example of what you guys would hit on, right?

Itamar Friedman

>> Right. So, the thing is that we do believe in future at Qodo. We do believe in a future where AI is going to write most of our code, and actually review most of the code. And we just want to make that with high integrity. So, there are solution for the problem. That's what we're doing. For example, right now, if the blue team AI is spitting code, the red team can at the same time, or pre, or post, sorry, during the code generation, or post code generation fetch information from the internet, from GitHub, and verifying that the code before it's being used by users do not infringe anything. So, that's the idea. So, infringement is part of what we see as integrity, and like we mentioned maintainability, security, quality, compliance, privacy, et cetera. So, what we need to make sure that we're treating as we're rushing, and I do think it's exciting, we're rushing to use AI to generate all our code.

John Furrier

>> What's great about this wave is it brings up old computer science memories that trigger in my brain, which is words like interpreter, or compiler come into my mind, because what you're basically doing is doing some interpretation in real time to essentially make the code safe, and runnable. In other words, out in the open, these are computer science principles. Are there other areas you're seeing come together that's being, I won't say repurpose, I'm making that example up to illustrate a point, but what you're doing there is for good reasons. You want to run good code, and that's what your mission is. What are some of the things driving your business from a tech perspective, some things you're borrowing from computer science, first principles, and what's new?

Itamar Friedman

>> So, I really agree that there's a lot to learn from the past to project here on the future. And I think that since now software is becoming a statistical, software development is becoming a statistical creature, right? Software development, and machine learning are being combined. I think another field that we should learn from is machine learning, and the equivalent that you can look at as verifying that code works as expected. You would call it testing, and verifying that a model work as expected, you would call it a benchmark. And now that software is being written by a model, we need to apply some of those best practices from machine learning into the software. It doesn't make sense anymore to use common traditional testing. You need to use the testing that was used for machine learning now also for software. So, I think you are seeing it actually happening right now. You're checking if an agent is working well according to a benchmark that is called SWE-bench, that is called Codeforces, if you're familiar with OpenAI, and Anthropic like releasing models for coding, they're showing how well they're doing on those statistical benchmarks for coding. So, I think that's extremely important. We need to do that as soon as possible. We need to say our software is changing, and actually the software that we're going to use is going to be rewritten as we use it. That's the near future. So, we can't have those traditional testing capabilities from the past. We need to apply the machine learning the benchmarking capabilities into the future. That's one of the, I think important notions.

John Furrier

>> That's a great example that we're living in an era where conventional wisdom isn't always the standard. You got to have situational awareness to what's happening. On that point, I want to ask you, it's come up in some conversations here, and on theCUBE, and today actually, and other times is that there seems to be a tech stack emerging for AI workloads. And if you look at say Kubernetes, Ray, VLLMs, PyTorch, each part of that stack have different people who care about certain things. For instance, VLLMs, they care about writing the models, making sure they're efficient. They run on this GPU, or whatever, PyTorch, slinging Python around. But if you're a platform engineer running Kubernetes, you have no visibility, you don't really care. Their application got to run on, but then same they got to run through. So, you start to see a whole nother end to end. How do you view that? Because as coding comes on, it opens up a question of, okay, what's happening in the operating system, or the operating environment of the tech stack? Do you have a perspective on this?

Itamar Friedman

>> Yeah, I think this world is moving really fast. I don't know if you remember back in the days we were talking only about three years ago, there's a lot of buzz, and I think just justifiable on ONNX that enables you to take models from different frameworks, and to have one inference, and it's gone, right? So, it'll be interesting for me to try to predict the future about this, but I still think that I'm going to try to do that, and say that I think that the most interesting framework that you actually need to consider these days are how are you going to orchestrate the agents? I know it's a bit higher level than the inference part, but I think that will, it's the most interesting because Jensen for example said that in the very near future he wants Nvidia to have 100 million agents working for Nvidia, and 100,000 people working for Nvidia. And when you're thinking about those agents that you need to handle them, it's much more than inference efficiency. It's like how do you do batch compute? How do you do contact switching? So, as part of, for example in DLM, there's some mechanism to try to deal with a few models running on the same GPUs, and different inferences coming in different times. This is going to happen big time when you need to have a full data center just running, and agents working for them. So, when you're thinking on that load, I think it's hard to predict if DLM, or other Kubernetes is going to actually going to hold that. And then that's why I think it's not a surprise that Nvidia are buying companies like Run:ai that are managing GPUs, and other companies that some of them were announced that are managing how you do inference, et cetera, because I think we didn't have the last word of how you're going to manage those agents, and the practically.

John Furrier

>> It's a great point because it's an unknown, but codices will probably be... Coding will be there with the agents may be in the future. It's funny, Dave Vellante, and I joke on theCUBE all the time, it's like we used to have full stack developers lamp stack back in the day, and then cloud became a half a stack because they did a lot of stuff for you, and now you are back to what you're kind of teasing out, which is full stack use case specific code, which looks monolithic, it looks maybe homogeneous, but we live in a heterogeneous world. So, if you have the distributed computing environment, there's a real conversation starting to emerge around what's homogeneous, and what's not, and do people care if it's homogeneous, if it works? So, again, what's your thoughts on that? I know it's little bit out there on the thought exercise, but what's your thinking on that? Because I wouldn't mind if someone was homogeneous if it didn't have to be heterogeneous. I can run great Nvidia all the time in this use case. It's like a processor, right? You harden hard to top down, who cares, or?

Itamar Friedman

>> Yeah, look...

John Furrier

>> I got to work across ecosystems. Whole nother discussion.

Itamar Friedman

>> Yeah, look, I'm going to be a bit romantic here maybe, but we're heading into a future where writing software is going to be very cheap. So, basically I could tell you who cares. Basically I can just ask for something I can ask for change this software to run for this environment, change this software to run for this hardware. The reason I'm saying a bit romantic here because I think it will take time not in the next year, but five years, not 20 years. So, when you're thinking about that future where software can be almost immediately being written, because you're going to have the blue team, and the red team working together, and making sure that software is being written is going to be high quality, by the way, that's why the reason for the purple rather than blue working together. Then basically I think that question would be a bit less important, and in a sense that you would care much more on your application layer, and that's why I think by the way, you could see all the players that are around infrastructure, whether they're clouds, or chip design, thinking about software development, and putting their resources, and effort, and software development, because when it reaches that peak, then it's going to disrupt a lot of the way we look on infrastructure.

John Furrier

>> Yeah. The utility of computing, and GPUs are certainly going to be the value. That's why everyone's coding lower in the stack, microcode level, similar kernel, a lot of the alpha engineers are going down, not up, because coding can be commoditized, but I love that call out. First of all, I agree with you. I'm a little bit romantic on, too. I brought it up for a reason because we don't care if it's working, we care if it's not working. But I love the five year 20 year distinction, because that brings up the fact that hey, we're all grinding, let the best company person win. But that's not a bad thing either because that speaks to the speed. You mentioned the framework that's no longer around, well goodbye, you're irrelevant. So, cool is one thing, but cool, and relevant becomes another one. So, we're in a five-year window of innovation, so we'll cross that bridge when we get to it, right? I mean this is kind of what's happening.

Itamar Friedman

>> Yeah. So, if I rephrase, recap, what I started with is that the first phase of coding started as assistant, as extensions, and existing integrated development environment, and then it was not enough, and we needed to reinvent the UX, and then you saw forks of IDEs, and integrated development environment like Cursor. Then now we're talking about the fork of people, and agents working all across the software development life cycle. And this is going to be really hard phase. The fact that we managed to pass through quickly, the first, and second makes us romantic about how fast a third one, and you're seeing the CEO of Anthropic, the CEO, I'm going to call them out just by their title, the CEO of Anthropic, the CEO of OpenAI saying that ASI is around the corner this year, and forget about being a developer, and quoting by the end of the year, it's going to be much better than developer. I think that's for monetization. The third part is actually taking a lot of the human out of the loop. And that's going to take time more than the first, and the second. And again, that's where I think like AI, we must have more fundamental workflows, and technologies using AI as well to also verify that what we're doing is proper. And I think that it's not this year, but it's unavoidable that's going to happen in five years. And when that happens, I do think we're seeing acceleration of many other fields that are happening. So, I think it's really exciting, and we need to think about it as society, et cetera, and we have time. That's my thinking.

John Furrier

>> Having really accurate responsible software with confidence at the speed of AI is what you guys do. Put a plug in for what you're working on, status of the company, where you're at, your focus is today, and what's going on?

Itamar Friedman

>> Yeah, we're only two, and a half years old. I think one of the youngest there in the field already in the magic quadrant. It matters because we are working for serving Fortune 100 enterprise. We already have multiple Fortune 100 companies, including Fortune 10 that are using us more than 10,000 developer weekly active user in some of them. Obviously use our spearhead product, which is the code review. And some of them also combine it with our co-generation, or with other co-generation like Copilot as an example. It doesn't matter. You're going to have many co-generation blue team, and they use us as the red team. We're scaling. We grew from 25 people to 80 in the last half year. We are scaling in the US specifically, I'm moving to New York from Tel Aviv, and really excited to partner up with our new partners worldwide, which include AWS, Google, Microsoft, and Nvidia. So, really excited about next year that is coming. We're full stack. We have our models, we have our frameworks, open source, et cetera. So, that's really exciting times for us.

John Furrier

>> Well, we'll certainly see in New York with theCUBE's new studio, the NYSE, of course the NYSE Wired that Brian Bauman put together. And of course Simon's got a great community. All open source, all for the community. And again, we'll do a deeper dive, and appreciate you coming on theCUBE at the late early morning tomorrow, California time. And what should people expect for the AI agent conference? What teaser would you give folks? What can they expect to see, and hear?

Itamar Friedman

>> Yeah, I know my audience would love to hear first, what are we going to see in the near future, and how we're going to get about that, and what it means for them, right? We're not trying to predict what happens in five years, but what's coming next year, and how should you prepare for that? We are seeing at Qodo, by the way, Qodo stands for Quality of Development. We're seeing teams from Fortune 100 to companies like Monday.com that are not there, but growing really fastly, rethinking how they're building theirself, and their dev team considering the AI era. So, I think that's what's exciting, at least in the dev-like channel, the track, the AI dev track, what's coming to you in the next year, and how you could build your team accordingly.

John Furrier

>> Congratulations, code review, certainly important, responsible AI, safe AI, quality AI code. And again, we're all going to have AI agents working for us. We'll make sure that they don't get forked if they do, they get in the right way. Thanks for coming on theCUBE. Really appreciate your time. And again, thank you.

Itamar Friedman

>> Thank you. My pleasure. Thank you for having me.

John Furrier

>> Okay, I'm John Furrier here as part of theCUBE, and the NYSE Wired preview of the AI agent conference happening in New York. Thanks for watching.