What’s Blocking Fully Autonomous Vibe Coding?

Hello, everyone. " My next guest is Asaf Zohar. io, and let's welcome Asaf to our show.

Asaf, it's great to have you on here. Hey, Yaron. Thank you for having me.

My pleasure. So Asaf, I mentioned you're the CTO founder of Stacktik, but people always want to know, okay, we got his title, but who is this guy? Where has he been?

What does he do? Yeah. How'd he get here?

Tell us. Yeah, sure. So, yeah, Asaf Zohar, based in Tsui, lived in Tsui for many years.

So basically, I used to be a Presale Architect for many years from a US-based vendor, and I used to have a nice income, stability, interesting work environments, and someday something terrible happens. Yeah. I don't know if I was bored or, I don't know, maybe vendor of myself somehow, but I did something very stupid.

So, like quitting a job, filling up patent, and opening a startup. And yeah. You got to be a little crazy.

Yeah. I've done it multiple times, but it's hard. Once you do it, though, it's in your blood.

Yeah. Definitely. So it's the journey way beyond the technology that's checking your skills, how persistent you are, how determined, how much are you willing to sacrifice.

And for me, for example, my biggest weaknesses as probably a founder or in general, I never chase people, I hate begging for money, I don't like to keep a long-term relationship for the sake of some business in the future. So basically, and this is a very important part of being a founder, at least I'm- Which is why you're the CTO and not the CEO or the CRO. Correct.

Yeah. But as a founder, you still need to have this kind of side, which I'm actually missing, at least I understand my weakness part and improving it, yeah. Absolutely.

Takes a big man to acknowledge their shortcoming and to work on it. io? So yeah, another little story, okay.

So, when AI became accessible for everyone, I was very excited. Yeah. And I always liked automation.

I used to write tons of code, automating things for customer, for myself. And when AI became accessible, I understood, oh, damn, everything I did now is actually, you can do it in few hours or days. Yeah.

All the years, all the code. But then I also identify a gap, and the gap is between what AI generates and what AI can automate. And this is where Stacktik became.

Yeah. I thought about this direction. What is missing for AI to be a real autonomous automation and operator for application?

And it all started from there. I'm talking about two and a half years ago. And we've been working very hard to bring the product.

It's a complicated framework, quite unique, based on five attempts already. Mm-hmm. First one will be approved in the US.

Basically, what we do is tackling the biggest AI problem in a very different direction. So if AI today is struggling to bring up the truth, yeah, the real truth, because if you ask AI about something, it'll give you so many different answers. Even for the same configuration, it will always be locked in trying to improve it, even though if it doesn't do anything good.

Mm-hmm. And everything we did is exactly solving these kind of problems of trustee, how AI can be automated, and this is how the story starts. Yeah.

You used the word trust. I think trust is one of the biggest roadblocks, one of the biggest issues around AI today. Correct.

It's interesting. The interview I did before this, Asaf, was with a security company to try to keep AI agents with guardrails, that they shouldn't break out, they shouldn't go do things they're not supposed to do. Every week we're hearing stories that this one's agent broke in here, this one's agent tried to break in there, that this agent exceeded its instructions because it was trying to fulfill its mission.

And it's important to remember, it's not that AI is evil or AI agents are evil, but they're very goal-focused. This is their goal, this is what their mission is, and basically, they're going to try everything they can think of to try to fulfill that mission within reasonable- Correct ... guess.

But that's probably, I think, the biggest issue that's stopping companies from saying, just turning it loose, so to speak, right? Yep. But then it's not the only one.

There are other reasons why we're not able to do fully autonomous, or it's not as much as we'd like to anyway. Right? Fully autonomous things like vibe coding and letting it run, letting AI do more of an ops role.

That's my impression from where I sit, right? What about from where you sit, Asaf, what do you think? First of all, I totally agree.

Yeah. And this is exactly the ambition. Yeah.

We want to create a dark factory or a single AI dark factory that people can actually vibe code production compliant ready. So how is it even possible, yeah? So what Stacktik provide, it's a framework that the AI will communicate.

" We will take it, and we will create something which is completely tuned up to a specific compliance, for example, like PCI DSS or HIPAA, like magic. Mm-hmm. Everything the AI will do will be governed, and everything will be calculated because the main problem is relationship.

Yeah. In one word, if I need to describe their problem. The thing is, if you have operational complexity or you need to create a application topology, there are tons of relationship, and this is a problem of AI.

AI doesn't understand all the relationship. It needs something to calculate it and to graph it like a metadata. Sound a bit confusing, I guess, but think about it like that.

Yeah? Yeah. If you think about application, and we take like a little piece of it, and we take a Kafka topic, connected to a sync database.

Yeah? Mm-hmm. So now you have the code, you have some infrastructure, and you have some little binary version of the connectors and another version of the database and another version of the Kafka, and then you have tons of configuration.

You have mTLS encryption, you have secrets, you have the Kubernetes configuration or whatever you use. You have layers of ARBA and enforcement, and then you have observability layers. You have logs and metrics.

Think about all these layers, yeah. So when you do a code, actually, there are so many layers going up on top of that. Yeah.

You change the code- You might impact the compliance. If you think about the security compliance on top of that, it's even more complicated. So we take all this relation, we map them, we database them, we make them, and when AI, whatever it does, it goes, and you understand, ah, okay, I'm changing now the version in the connector, we will have a two point of impact, and I need to automate also this XYZ.

This example, we do it for everything. We do it for the automation of compliance, for the automation of operation. Is it clear?

Does it make sense or I'm a little bit- Yeah, no, it's clear ... a little bit too technical? No, Assaf, I'm technical.

Our audience is very technical. Sure. You're in the right place.

Go technical. Yeah. So basically, this automation framework is exactly coming to solve this kind of trust problem, vibe coding.

So when you're vibe coding through our systems, everything will be production ready, everything will be calculated. And on top of that, we provide the validation because the main problem is the validation. It doesn't matter if I tell you everything is PCI DSS, and your code is amazing.

If you go to our UI or to the API of the platform, you will be able to see the full topology of your app, every layer, every validation, what is missing, every section of compliance. " This is, at least from my personal opinion, is what everyone looking for, or at least in the future, it will be the most important part. Sure.

So Assaf, let me ask you this, though. This market is evolving, changing, growing so quickly. Yeah.

Three months ago even, right? Yesterday, right? Basically, I don't know, I use Claude Code to vibe code.

Yep. I know. OpenClaude came out.

Maybe I started using OpenClaude or one of the variants because I secured it a little bit. But basically, I had an agent I used, right? Now we're evolving this whole stack of where we have a super harness, a harness, and we have multiple agents.

I use Claude Code for this. I might use ChatGPT Codex for that. I might use something local here, the homegrown thing there, and I switch, I'm pulling them in and out based on the jobs I'm doing.

Yep. Some of it is hosted in the cloud. Maybe I'm starting to host some open weight models at the enterprise level, right?

I'm hosting those locally. Is that something that Stacktik deals with as well, though? Because you're still looking at the code, no matter what agent turned it out, you're looking at the code, I understand.

But how does that figure into the equation? Oh, it's actually a great example, yeah? Because we are not fighting against AI here.

Right. We're actually, when AI get better, we are getting better, okay? And that's actually related to the first part, that everything is changing.

For us, it doesn't matter. When Claude is getting better, our platform actually is boosting up. And it doesn't matter for us what kind of client sit behind.

Yeah. So you can have multiple clients, you can have one client. All the session will go through the same MCP connected to our platform, and it's actually creating this kind of centralization and source of truth for all the AI.

Yeah, so it's actually a great example because this is more or less what the vision is because we understand that things changing, people will change AI client. It's very easy to change AI client. It's not like a vendor-lock cloud.

Sure. You can change it every day, you can change it tomorrow, if something better is coming, something cheaper is coming. So basically that's exactly the point, to take all this mess of AI agent to govern them, to create a real automation platform, which doesn't have anything to do with what kind of client you use at the end user spot.

Okay. Let me now turn to Stacktik specifically. Talk to me about kind of who your user is.

Who's the customer here? Enterprise? Individual coders?

Who? So we have a different offering for enterprise and a different offering for small startup. In general, everyone that have complex application and operation, this is a good fit.

And one thing which is very important to mention, but actually I skipped, I think, and it's related to what kind of customer. When we create the code and the operational and the syncing it to the cluster, to the production cluster, we warp everything as a completely agnostic sovereignty cloud stack. Which mean this stack is completely free of vendor lock, it's completely open source, and you can move it with a click from any cloud to another.

We think it's actually even better than the automation, and this is also related to what kind of customer will use it. This kind of portability, agnostic freedom, this is worth a lot of money. Because basically, you now bypass locations, regulations, cost of infrastructure.

You can put it on-prem, you can put it in air gapped, you can move it from one provider to another. You are now thinking vendor lock free. And I'm considering this a very strong offering that will be relevant for even someone that heavily using managed services today and interested in moving to a completely agnostic solution.

This is also a perfect match. In general, this is our vision. Our vision is to be completely open source.

And when people think about sovereignty, sometimes it sounds like European thing. Yeah, sovereignty, it's European, it's like hippie stuff, open source, but the hippies are actually the one that use managed services because ChatGPT and Antopic have 200,000 Kubernetes cluster. Yeah.

Maybe they use managed services for some 1% of their workload, but why? Because they are not hippies. They are smart, and they need to scale, and they need to create something completely innovative that's fully customized and cost-effective, and this is what we bring.

So anyone that have complexity, you want to agnostic it, you want to have a freedom, you want to automate it, this is where we're good. More complex, more value. Yeah.

And the solution is generic, which mean it can be a fintech application, machine learning compliant with the HIPAA, e-commerce like eBay. It doesn't matter for us. So any customer could be a nice potential.

Excellent. Yeah. Assaf, we're way over time here.

I got to end it, but I just want to end with this. People can get more information. It's Stacktik.

io, right? Correct. io, you can find all the information.

By the way, Stacktik is stack and tactic, just to explain, I forgot this part as well. Very good. Yeah.

All righty. Assaf, thanks for being on Techstrong TV. Good luck with Stacktik.

Come back and keep us posted. Thank you very much, Adam. All right.

Hey, we're going to take a break here on Techstrong TV. We'll be back in a bit. Thank you.