Ari Zilka on AI-Driven Observability, Big Data Challenges, and My Decisive’s Real-Time Approach

Hey everyone. We're back here at CubeCon. Uh, we got a little bit more to do here on our day two coverage.

My next guest is named Ari Zilker. I hope we got that right. And Ari's company name is My decisive, My decisive.

You may not have heard of my Decisive, my Decisive or Ari, but hopefully by the end of today or the end of this interview anyway, you will. Ari, we Ari, before we talk, uh, before we discuss my decisive mm-hmm. Let's, let's have a little discussion kind of, you know, give us your journey to being here, you know, at my Decisive.

Sure. So My Decisive is a proactive operations product. Uh, I got to my, to the ideas and founding this company 'cause I was one of the general managers at a little company called New Relic.

Sure. For five years before that, before that I was a general partner at Coastal Ventures. Sure.

Then, uh, working backwards found, helped found Hortonworks and then helped start terracotta, eh, cash before that. com. So I'm old enough to remember most of those.

Right. Okay. Look for those, for those kids out here.

Right. Horton Works kind of defined the big data, right. That was the big data.

Absolutely. Uh, age. Yeah.

With Horton Works, which was a spinoff from Yahoo. Yep. Correct.

Yeah. And, um, interesting stuff. Interesting stuff.

Indeed. New Relic has had a, an interesting also story of, you know, look it, it's rode the waves of this market from a PM to observability to, you know, a lot of different things. But, you know, it it, it affords one a real, uh, opportunity to, to yeah.

See what's in the market. Um, every founder I've ever spoken to, there's a passion for what they're doing. Somehow they think it's, it's gonna make the world better, even in a small way.

It may not cure cancer or, you know, bring world peace. Mm-hmm. But in some small way, it's gonna make someone's life better.

Yep. Talk to us about your passion. Um, easy.

I have a passion for what observability folks call root cause. Um, I'm also a scientist, a traditionally trained engineer, Uhhuh, and I'm frustrated with the observability vendors of the world saying this AI or that ai, and now I could do root cause, no one could do root cause. Uh, there's a difference between causal relationships between events and correlative relationships.

And I'm not gonna get into the math of it. I'll just say that I believe that we could do what the customers and the market needs, which is to bring in AI to help it run more smoothly at lower costs. Mm-hmm.

But I believe that the answer is not incident response through LLMs and through ai. And that root cause is the wrong question. It's when you ask a CTO what takes you down the most, humans take me down the most.

So instead of root cause, why not go after robotic change? Let the system change itself. We don't break it.

We bring in the genius chemical human brain to do root cause with analytics tools, but we try to get robots to fix systems. I love it. Let's, um, so you, I think you really frame the problem and the potential solution, no disrespect to you, but all of the booths around us here and they can't see, they only see this.

Yeah. A lot of boots here are promising a lot of solutions and, you know, they throw in a little AI snake oil, a little MCP here, you know, and, and, and agentic there and, and we're gonna solve your problem. Yep.

Why are you different? So look, I recognize Alan, we're swimming upstream against what you just described. Why we wanna be different is because we recognize from our insider seat in observability the, the signals not in the telemetry data.

There is no signal. Like you deploy an app in a database and the app starts crashing. There's nothing in that data.

And every observability vendor knows this that says it's the app or the database roll back. The one and the other will heal. That data's not there.

All we have is some CPU information, some URLs being requested and you know, a little bit more than that. Some Kubernetes and infrastructure logs. There's nothing that says, oh, fix the app and the database will calm down.

So why we're doing this, why we're swimming upstream is because we know it's techn technologically Correct. What we're doing. Okay.

It is principled approach. Mm-hmm. Uh, why we're willing to swim upstream against the AI claims is because we can introduce AI the right way.

We Can. 'cause ultimately you think it's the right thing to do and it's gonna be born out. Yeah.

Basically if you make the changes with a robot, then you can easily ask, was this a good change or a bad change? Whereas if you're looking at steady state operating data and say, did I just go unhealthy? That's an impossible question to answer.

Maybe this is normal. And I've heard too many customers say, I deployed an AI and my traffic doubled. And it started rebooting me because it thought this was an anomalous situation.

But I was handling that traffic and the AI took me down. And so, because I know that humans are gonna be in the loop for the next 10 to 20 years, because I know that change is an easier problem to tackle automatic change than automatic incident response, I'm willing to swim upstream. 'cause I think we'll win.

Fair enough. Let's talk about observability for a second. Ari, you've been in the observability sta space.

You've been in the big data space longer than that. Yeah. Quite frankly, the problem with observability is it was, is fundamentally the big data issue.

For a long time we had too much data to kinda really wrap our heads around, analyze and come back with actionable intelligence. Mm-hmm. With things like AI and, and other improvements.

It's not, you know, AI is just the latest Yeah. The greatest maybe. But you know, over the years with machine learning and what some says AI anyway, but Yep.

Machine learning, otherwise we've, we've been able to tackle that big data problem. But In tackling the big data problem, you know, it's the, the, the theory of constraints. Right.

We solved the big data problem only to find out bigger problems. Sit behind it. For many in observability, that biggest problem is my God, who, who can afford this?

Mm-hmm. Right. To to to really, now that we have the ability to analyze the data, we want to keep collecting more and more data.

The more data collect we collect, the more our costs go up and Yep. Storing the data, analyzing the data, reporting the data and it, and it's like, it's like an old Star Trek original series when you gotta keep feeding ball the monster. Yeah.

Right. Yeah. Otherwise he goes out and the whole planet goes to hell in a hand basket.

It's the same thing with observability. We gotta keep feeding the monster. Mm-hmm.

How do you help with that? So we come on pre with our solution. Okay.

First Of all, wonderful question. You're spot on. Um, but if I come OnPrem, then I could introduce you remember Michael Stone breaker?

Sure. Database genius. Uhhuh, he had stream base, right?

Mm-hmm. He started swimming upstream. He said there's a bunch of business problems that need to be solved in the stream in real time.

You don't wanna land the data to a spindle and then ask a question of it as an analytics query. You wanna pose that question of the data as it passes by observing the change in state. That's what we've built.

So our product is Kubernetes native Open Telemetry native. In fact it's going into the CNCF. We can talk about that a bit, but the, the idea is if no one's introduced an observability streaming analytics engine, and if I do that, then all of a sudden I'm bringing the power of, we do this in networking, we do this in databases.

There's stream base, there's streaming databases, there are network devices, software devices on the network. Palo Alto Networks F five big IP that run at wire speed. They're stateful and you can co-locate the application right there with the data as it originates next to the app.

We are one of those. So we are an observability appliance on-prem super fast, super lightweight distributes the cost of ownership of the solution to all the edge points instead of trying to be a SaaS that centralizes it and now needs to spend a billion dollars a year on AWS and needs to mark that AWS charge up to it and pass it through to its end users. Excellent.

Excellent. Quite excellent response to that Ari. Thank you.

Um, we gotta do a little housekeeping. Okay. Website URL.

Where can people go get more info? My decisive Do AI For the spelling challenge out there. M-Y-D-E-C, I don't know.

I can't think of You're One of the spelling challenge. I'm Gonna just the ISI You gotta use it in a sentence. com.

com Dot a I won The spelling voice a I dot a ai ai I still won the spelling. If you get rid of the dot, it works in sentences perfectly. I use my decisive AI to control my production system.

There you go. So it doesn't break And so it doesn't break. Yeah.

Excellent. Um, how can people get started? It's self-service.

It's completely designed like when a customer prospect shows up at our front door and we're lucky enough to engage people, we send them to the Slack channel, we send them to the website, we send them to the GitHub, it's all hanging there. Top level navigation off that website. One last thing you mentioned, uh, joining the CNCF and I'm gonna assume the, the working groups for o uh, opt Hotel and Yeah.

And so forth weren't that happening Next week Right after this conference. Really? Yeah.

So stay tuned for that. We may have to report that news though. We just reported it.

And besides joining, will you be donating some of the code you're using at my Decisive? So We're not ready to donate the core code. Um, it is open under a permissive license.

Mm-hmm. And we are part of the CNCF, so we will comply with all their tests. Open source.

Yeah. And harnesses and their open source policies. That said, we are contributing to open Telemetry itself.

So last week there was a missing part in Datadog. We, and it didn't work with hotel We fixed that. Gave it and it was upstreamed in under 24 hours into Core Hotel Distro.

We see them. Um, must have been a serious problem if they moved that fast on it. Yeah.

The next thing we're doing, we're working with hopefully gonna be able to work with other observability vendors and build what someone else calls enrichment for open telemetry. Okay. We're building on-prem storage for open telemetry so you can keep your data in S3 and not be scared to sample it, filter it or introduce our streaming technology.

'cause you could always grab the raw original and do something with it, whether it's in an open source database or a commercial database or you forward it back to your observability vendor. All of those modules, those are changes to tel itself and we're contributing those on the tel side of the fence, not on the My Decisive side. I love it.

Ari, thank you so much for coming here on Tech TV with us. Thank you. Continued success.

We'll be looking for the announcement about the CNCF and we'll continue the conversation. Appreciate Your time. Thank you.

Appreciate your time. Hey, we're live here at Q Con. What we want.

We've got one more today. Stay tuned. We'll be right back.

Act.