Solving the AI Confidence Gap: Kobi Tzruya on Intent Alignment and the Future of AI Agents

Hey everyone. Welcome back here to Tech Drunk tv. You know, I'm, I'm in my studio at the house today and I was really happy to have my next guest on here.

He's a friend of mine. I've had the pleasure of interviewing and speaking with him and presenting with him on stage a couple of times at the RSA conference, which will be next month. Um, I want to introduce you to my Frank Co.

Kobe Za. True. Yeah.

Say it better than me. Yeah. Kobe, how do you pronounce it?

Yes, And I got it now. Alright, we Got it. I gotcha.

Kobe, you know, the last we had checked in with you, of course, uh, you were at check marks in it, in it, you know, c-level, senior role there. But, you know, this ai, this AI bug, if you will, the excitement around it has so many of my friends who haven't, they haven't coded in, in some of them in 20, 25 years even, right? They've been executives, they've founded companies, they've had exits.

They are in love. I don't know if in love or in lust with the ability to just generate code again and make apps, whether, you know, using code code or, or any of Theis. And, and they all have great ideas.

You know, for many of them, they, they thought they were done. They had founded their last company, they had their last great idea, but they're, they've got new ideas. They're founding new companies.

This same, the same fever's gotten to you, it sounds like. Right? Tell us what you've been doing since you left check marks.

Yeah. Okay. So, uh, you're right.

Uh, kind of this, uh, fever hit me, hit me, and, uh, I know I need to do it in order to clean it out. You got No, you gotta, you gotta fulfill it, right? You gotta Exactly.

I, I, I have to fulfill it. Um, I founded a company named the Omni Guard ai. Uh, Omni Guard AI are the, the main, the main problem that we're solving for is to make sure that, uh, that AI agents or, uh, or applications that are using, uh, that are using ai, uh, LLMs, um, enterprises and companies that develop them will have confidence that they actually do what they intended.

So you will not, let's say you'll not develop an agent, and, uh, it, uh, if you're a car dealer, and it will sell a car for $1 or, or, or, or you'll not develop an agent that will, uh, that will, uh, that, that will send illegal advice to, to your, uh, uh, to, uh, to your, uh, to, to your customers or, or, or will be biased or, or anything like that. Meaning we look at the behavior, okay? Uh, at, at, at, at, at the behavior of, uh, of, of the agent or, or the agent application.

We see if the intent and we can identify the intent and the intent and also the workflow, uh, that the agent or the, uh, or the workflow needs to, to work by. It doesn't match the output. And if it, if there's a deviation, uh, we can actually either alert or stop the response or, or, or, or bring a man, uh, or, or, or, or, or, or bring a man in the middle in order to, uh, in order to look, and we can also identify the root cause, meaning we're not only giving you the headache, we're also telling you Okay.

Where it comes from. So, excellent. So that's, that's the vision.

This is what we're, uh, this is what, uh, this is what we're working on. And it's a real problem because what we see is that, uh, you know, when you ask people, are you building agents, uh, like 80% of them, uh, 80% of them, you know, pick a survey, uh, will pick a survey, 80% will say yes. And then when you say, okay, is it in production?

Then, you know, the percentage drops to 10 to 20%, and this at best, yeah, at best. And the confidence in these iGen applications or, or AI agents I is, I think the root cause, meaning is the root cause. Absolutely.

Co Kobe, I, I just wanna make sure we got the name of the company. Can you spell it for us? It's Omni Guard, a i omni guard.

It's, uh, O-M-O-M-N-I-G-U-A-R-D Guard. Like guardian, okay. Yeah.

Okay. Look, the way people talk about it, sorry for my accent today, I would No, no. The way people talk about God For my accent, you know, English is not my first language.

That's okay. No, but the way people talk about God, I wasn't sure if you spelled the GOD, you know, so I wanted to make sure we got it right. Yeah, yeah.

Um, you know, Kobe, we were talking before we got on camera here, before we started recording, and I told you about this, uh, article I wrote, and the, the shimmy says I'm gonna do comparing AI agents to parrots, Right? Yeah, yeah. Parrots talk.

And they, and they seem to talk great. One parent says hello, the other parent says, how are you? And in our minds, and I think it's human nature, we fill the gaps in and say, listen to that.

They're intelligent, they're talking to each other, right? We, we tend to give things human, uh, human uh, not feelings, but human identities, if you will, human sort of capabilities. And I think there all capabilities, I think, yeah.

You know, there's a, there's a fancy word for it. It's something like Anth anthropomorphizing or something like that is the scientific term where we put human kind of capabilities onto things or, or other beings. I think we're guilty.

We're guilty of doing that with our AI agents, be because they, they talk back to us nicely, right? You, you tell it to do something that says, oh, that's a great, that's a great suggestion. I'll get right on it.

Oh, yes, let's improve it. That, and we, and we, too many people give their agents names, right? And, and it's true.

It's true. And, um, so we, we, we tend to indu, ind them with intelligence or capabilities that they really don't have. You know, I'm, I'm, I'm not saying they're, they're dumb, but they're tools.

They're tools, right? And a tool does a job, and if you don't give it the right parameters, the right instructions, don't blame the tool. Blame the tool user.

One, 100%. Uh, 100%. Like, uh, what are the things, as I mentioned before that we do is we trace back the root cause, meaning let's say we have, we see deviation between the output and the intent.

We also trace back and, uh, like from what we see when we now work with our design partners and, uh, and, and also in our research lab, is that a very high percentage is, is because the task was not, was not, uh, was not given well defined in a, in a precise and a detailed enough way. Mm-hmm. Yep.

Okay. So at, at, at the end of the day, it's a machine. You need to tell it exactly what to do.

It's a tool. I, I agree. Now I just wanna let people know who we're watching, right?

This is sort of a, we're back in the farm. We're back in the kitchen here. Omni the guard is still a, a project that you are, you know, bringing to, it's, it's not even ready for market.

It's still in the design and, and market fit, you know, precede kind of, this, this is what you are working on real time right now, because you see the, the issue in the market. Do you envision that Omni Guard works with open claw or, or some of these agents that we see out there, or are you gonna have Of course. Yeah.

So it's not that you are developing agents, you are just not, not even think managing, but you're managing the pro, the workflow of the agents, if you will. Yeah. We're, we're managing and we're monitoring it, and, and we can Yeah.

We're, we're monitoring and, and, and managing it. And by the way, you mentioned open call. We, these day we're working, we're working on a plugin for open call.

Yeah. That, that, that, that, that will give, uh, that will give the, uh, the community, the, uh, you know, the, the, the, the ability to, to manage and control. Okay.

Because the open call is, is great, but, uh, it's also, it, it's, it's, it's also a headache because, uh, It is. But you know what, in some ways, Kobe, it reminds me of the early days of like Docker and Kubernetes, where it's a good piece of technology, right? They didn't think about security as a security person.

You know, you know that that drives us crazy, right? So what I'm saying, it's a headache. Yeah.

That's it. They, they didn't think about security. I'm a Security person, you know, I talk Yeah.

And, and they didn't think they didn't. It's not a complete game. It, it's a piece, but it's a big piece.

And I think what we're seeing, and we're already seeing at, you know, this is the beautiful, the beauty of the times we're living it, this thing's out a couple of weeks. It's had, it's, it's got the most stars ever in GitHub, the fastest, you know, I forgot what it was, 2 million stars or something. Um, but we're already seeing this sort of ecosystem coming in around it.

I, my, my inbox is flooded with pitches from companies that are securing open clo, right? That are helping you manage and, and these kinds of thing, or open clot. Um, I think, and that's maybe what we needed in the agent space, is the agent that was general enough, but was not the whole package that would allow other people, you know, other companies to come in and bring their vision to this whole agent.

Ai. Uh, yeah. So, uh, a lot of people kind of ask me, okay, what, what, what is the uniqueness that that, that, you know, that you, uh, that that, that you bring?

So kind of the uniqueness that we bring is kind of, we come, uh, orthogonally, uh, you, you, you, you can say, and we say we have here, um, an agent and, uh, an LLM piece, okay? Uh, an engine, which is an LLM based that all, all it does, and this is how we built and and trained it, is to look at the intent is and, and, and to see if you're aligned at the intent. We don't care what your business is, okay?

What, what, what, what we care is about intent. And kind of, we're kind of, we, we, we are, we're intent experts. And, and, and this is, uh, and this is the product that we're, uh, uh, that that, that we're building and, and kind of we're super encouraged with, with the outcome that we're seeing with the, uh, uh, with the design partners that, uh, we're al we're already working with.

Excellent, excellent. You know, I, I, we led off the conversation with so many of my friends are, are coding again, thinking about starting new companies. It, it's gotta be such a, uh, such an exciting time, right.

To, to do this. Give us a sense, Kobe, right? Because you've been in startups, you've been in, you know, companies that were a little bit more mature.

What's it like right now, sort of swimming in these waters and in, you know, the whole world is, so every day seems like new breakthroughs, things are accelerating, things that would take that I, we'll see this happen at six months or a year. All of a sudden. It happens in two weeks, right?

The, the MBOs, the open clause is a great example. What's it like for you, Kobe sitting where you are now? Yeah, well, uh, like a Moving target.

Great, great. Great question. You know, you know, all of my life I've been, as you said, in startups.

I built products zero to one. I see 2, 2, 2 main things right now. First of all, the level of uncertainty the unknown is, is huge.

Okay? Also before, like 2, 5, 10, 15 years ago. But it's nothing like now, because things develop so fast, and especially with ai, and you don't know how it'll turn, how, how it, how it'll turn.

And everyone also understands that we're at a historic moment, you can say with, with this ai, but no one knows how it will kind of, where it will go. So anything that you do, you can, you feel that you, that you gamble at, at, at, at the end of the day, kind of you, you, uh, you know, um, you put all of your intelligence and experience and, and everything, but at the end of the day, you know, it's like flipping a coin right now. What, what, whatever you do.

Second, I think that kind of the companies that will do great are the companies that will be able to adjust, okay? Meaning you need to adjust fast, okay? Even if you had a dream, you had a product, you built it in one way, you might see a month after you built it, or even the course, you might see that, okay, that the path that I took is not, uh, is, is, is not the right one.

You will need to adjust. So you, you need to be very, very attentive to what is happening and adjust very, very, uh, very, very fast. Yeah.

Uh, very good, Faster than we've ever had to adjust. I say, Sorry, I say faster, faster than we've ever had to adjust Faster, fa faster than that, that we never run. And by the way, laas, that we have ai that, that help us to, to, to, to, to, to build things fast.

Okay? So, So that, but that's the paradox. Yeah.

It's, that's the paradox. Exactly. It's stored and the shield, right?

Because at some level, that's what is forcing us to adjust so fast. 100%. 100%.

But yet it's also what we're using to adjust. So 100% it's a push and a pull. 100%.

Exactly. Crazy. It's Crazy.

Cut. It's causing the problem. And also giving you the tools to, to fight the problems that, that, you know, that that, that it caused very, but, But it's Very unusual situation.

Yeah. It's, and that's what I think, I think you just captured it. That's what makes it so exciting, right?

Exactly. The possibilities seem endless right now. We'll, we'll see what's real.

Like you said, 80% say they're using agents, probably less than 10% are getting any value out of it, because is it really working? But I think, I think every day it gets better. You know, we, we spoke about this open claw for all of its shortcomings and for all of the knocks, there's a reason it got 2 million starts.

Yeah. Right? Yeah.

And we'll have to see how it plays out. The, the reason, the the reason is value at the end of the day. Yeah.

Well, and that, and you know what, it, it's funny you said that because I'll bring a full circle to my parrots, right? This shimmy said thing that I'm doing on Thursday. We don't need it to be intelligent.

We just needed to do its job and deliver value. Because business runs on value, business runs on predictability, right? Mm-hmm.

Not, not debating, I think therefore I am right. It's not debating philosophy. It, it's, does it do the job I need it to do?

And if it does, it delivers value. And if it doesn't consistently, and its scale, 100%. That's it.

Omni guard. Omni guard AI or Yeah, Omni Guard ai. Yes.

Kobe, I wish you all the success in the world on this. Thank so much. Keep us posted.

I'm sure as, as you make progress and you know, 'cause you know how it is. It's ne it's not a linear road with a startup. It's, no, it's, it's, it's a little bit of a chacha dance, two steps forward, one step back, one step forward, you know?

Um, but come back and keep us posted, okay? Sure. Tha thank you so much for, for having me this You're welcome this morning.

Welcome. My pleasure. Thank you so much.

I'm the guard. I'm the guard. Ai Kobe's the guy who gets things done.

So keep your eye on that. We're gonna take a break here on text, on tv. We'll be back in a moment.