How AI-Driven Observability Is Transforming SRE: Insights from Mezmo CEO Tucker Callaway

We're back here at Cube Kahan. It's afternoon of day one of the main show yesterday. You know, they had all the satellite conferences, but today's the, the opening of the, uh, show floor.

And there's, uh, there's all kinds of things going on, uh, including, I think there's a, the cube crawl or whatever they call it tonight. Yeah, yeah. Tonight.

Yeah. I don't know if I'll make it to that, but we'll see. Anyway, let me introduce you to my friend here.

He's been on Text Drunk TV before. His name's Tucker Callaway. Tucker, welcome back.

I don't think you've ever done one of these live with us though, right? I think I have. I think I did two years ago at Q Con, or No, it was R-S-A-R-S-A.

Yes. Yeah, yeah, Yeah. At Con You're right at Q Con.

You're right. Yeah. I like doing RSA, you know, we just, we'll be at RSA this year.

All right. Going, we just got all my contracts done. Yeah, we're doing, so in addition to doing this there, I put on an event Monday at RSA, and this year it's usually DevSecOps, but this year it's gonna be on AI securing AI native dev.

Okay. So we're gonna talk about AI for That. We're gonna talk about ai, and that's a great segue.

No. Before we segue into ai. Seriously, Tucker is the CEO of me smo.

Yep. And I don't know if you all know me, SMO, and that's okay. If you haven't, don't be ashamed, but Tucker's gonna make you smart about mes o Tucker, go ahead man.

What's mes o about? Yeah, so Mes o is, uh, you know, we've, we've actually been handling telemetry data in the observability space for a while, and we are all about delivering a new observability experience that's fueled by agent operations. Now, obviously that's something we've started in the last 12 to 18 months.

Um, but we think there's a new way to go about doing observability and analyzing data. Uh, I think AI is going to disrupt this market. Yep.

Like crazy to tell you the truth. Yeah. com.

Okay. Um, you guys also play, you know, when we look at who does observability, who's using observability data, the s the uh, SRE segment was a big Yep. A big segment for mes.

And lately we've seen the rise of, let's call it ISRE. I don't know if I buy into, I dunno, it just sounds a lot of frigging I got opinions, You know, letters put together. Well, let, let's start with defining it.

What does AI SRE mean to you? So, I've said this a couple times, I don't really like the term, it's the term that we have. We don't like you either.

Uh, the reason I don't like the term is because an SRE is like a person in a job and a role, and it's also one that we've given a lot of of responsibility to over the years. And so just to suggest that you could AI a person or AI a role like that isn't right to me. So you think it's replacing the SRE with ai?

I, no, I think it's more I what I like. I think that's what people are referring To. That's wrong.

Yeah. And, and I think what I think we should be doing is, uh, to me it's AI driven observability, right? Right.

So, how do I take some of these mundane, routine tasks of, of, uh, like tactically managing the system off the SRE to allow them to go back to designing and scaling systems to do those things that require the big brains to do it. Got it. And not do the routine things.

And that's what really led us to, when I think about AI s or ai think about detection, diagnosis and remediation of incidents. But to me, that's really more AI driven observability. So I think this has a lot to do with what your view towards what role AI plays in all this.

Yeah. Right. I, I just wrote an article a couple days ago about the use of AI and journalism and media.

Mm-hmm. You know, and every media company is using ai. If they're saying they're not, they're full of crap.

They're lying. Yeah. They're lying.

I think it's pretty much like that in observability too. I think a lot of, you know, we see it in developers. 90% of developers are using ai.

Yep. Ops people, everyone PR people. We could replace them all.

No, we can't. But, you know, not PR people. PR people.

Right. But PR people are using a ai, we're all using ai. But what I wrote in the article is, you know, from the Billy Jolt, AI didn't start the fire.

AI can't start the fire. Right. It's the human.

That's the spark. You, you, you're out Almost anything you're doing with AI today, you must have a human in the loop. So to me, when I hear ai SRE I'm thinking about how is my SRE becoming better?

Here's how I think about it. And I like the spark analogy. Yeah.

You actually just re-branded around. You Got a spark around a spark. Very cool.

And you didn't know that. No, I didn because it was under There. That's amazing.

What are you having for dinner? No, I'm only kidding. So, but to me, that spark is actually, the data is the spark.

And what's changed a lot is how we can analyze data. So when we think about what observability fundamentally was, it was presenting complex data to humans for them to determine the root cause of a problem in the next steps. And so, like the, the solutions that are like out there today are basically taking people through an investigatory process and presenting them with the information in a very human consumable way that is no longer the best way to analyze data.

Yeah. Right. Now that's the best.

The problem is AI is a better way to do it. But the problem is at operational scale, the amount of data that telemetry or observability creates can't just be processed through ai. So we found that training the models, which is kinda the prevailing approach today and doing better prompts, was an insufficient way to actually derive operational outcomes with AI to do the ai SRE.

So we took a different approach, which is to optimize the input. And so we basically apply, it's called context engineering is the discipline versus prompt engineering. And what we do is we, uh, basically refine the data before it's supplied to the, the model.

And that allows us to give better, faster, and cheaper outcomes out of the model. Love it. So We can do it without training.

We can do it without, you know, using your data. We can, we can have you up and running in five minutes to do this. Really, it's, it happens at like 10% of the cost of what the traditional approaches are.

We have benchmarks to prove this too. It's pretty fascinating. Like, as you know, we've been, we've been focused for a while on the processing of real time data.

Well, it turns out that suited us really well in this transformation to an ai. I, I think, well, I was just gonna say that it 'cause that's kind of better lucky. Better, better be lucky than good.

Be best to be both. Yeah. We're both mine less than 25 years in venture back startups.

Um, But, you know, to me, so let, let's put the SRE to the cycle. Yeah. 'cause that's a human issue.

Mm-hmm. I, I don't think we're replacing humans with AI today. No.

I think we're replacing humans with other humans who maybe use AI better. Right. But the human's gonna be there.

But when we look at observability, observability, look, I've watched observability blow up here at CubeCon Yep. Over the last three, four years. Open telemetry, Prometheus, you know, have become open, tell more than any of 'em Yeah.

Giants. I mean, and almost the whole observability space has open tell kind of under the hood there and has a, a building block. Yeah.

I really believe that AI and, and ai, whether we're talking agentic ai, which is really, I think where we're going. Yeah. I think generative had its moment in the sun, but it's, it's agent, um, has the potential to blow it up.

Yes. Yet again, it was like observability was a, a fission bomb. And now we're looking at a hydrogen bomb potentially.

A couple ways that I look at that. One is I think the consumption of AI is gonna explode observability in a way that's far greater than what the cloud did. Yeah.

Right. That, that's one aspect of it. Uh, the other aspect is that, um, the way we think about analyzing that data is just gonna change dramatically.

Right? So, so an tel is a huge part of that. So it used to be that the incumbent vendors and observability own the creation of data.

Well, they don't own that anymore. Oell owns that. Right, sir.

And, and that was this massive disruption without any consequence because we were still fundamentally dependent upon those incumbents to analyze and process the data. Right. Because, Because they were the best way to process the data.

Well, they were, they were the face. They were the face. The Other stuff was kind of under the hood.

Yeah. So now the face is changing. 'cause now, now the best way to, as we were talking about, the best way to analyze that data is actually to do itally.

Mm-hmm. So now, if you're an incumbent, you no longer own the creation, nor do you own the analysis of the data. So now this is just a data management and an experience problem.

It's no longer this big giant problem that an incumbent can own. So the problem has like been shipped away at, but I think, or not, I think, I know AI has taken us to this tipping point where the analysis is now better done, you know, in the model, not by the human. And what we're proving, or have we believe we have proven is the models today are good enough.

They don't need to be trained. You just need to give them the right data. Right.

And if you refine the data in the right way, you will get the outcomes that you expect, that you want and hope for out of AI today. It can be done. And so to me, this begs the question then, what is the human's role in it?

So the human, like, so the human role is going to be, um, like for a period of time gonna be pretty big, right? Because like, 'cause if, again, if you get back into diagnosis, I'm sorry, uh, detection, diagnosis, and remediation, we're really focused on the diagnosis. 'cause I don't believe you can get to agent remediation until you have extreme trust in the diagnosis phase.

I agree. Right? Like, no one's gonna, no one's gonna let remediation happen genetically.

I mean, you might have a human in the loop trigger it, but that's just a human firing off an automation routine. We've been doing that for a long time, and it would get better with ai, but like, really tightening that loop requires the extreme trust in the diagnosis phase. That's where we're focused right now.

So I think until we have that extreme trust, we're always gonna have the human in the loop. But ultimately we need to be thinking about designing these systems for, uh, agents to process, not humans to process. And that doesn't replace humans.

That puts humans in control of agent processing. And it puts humans back into how do I design scalable systems and things like that. Which is what, which is, you know, not an SREI actually operated as one in a weird way for like in 99, but Story I thought you just stayed in the Holiday Inn Express.

Yeah. But, but You know, like, I, I believe that they got into the, into their discipline and profession because they wanted to build and design systems, not because they wanted to do observability in firefight incidents. Right.

Well, I'm sure none of the people 10 years ago raised their hand and said, I want to grow up to be an SRE. Right. We didn't have SRE until Google wrote the book.

Fair enough. Yeah. Um, but, but I, I do think you're right that I think the SREs come from the ops, what we used to call CIS ops.

Mm-hmm. Right? Yeah.

Some of them are now SREs, other pieces of the CIS op, you know, platform engineering. I think we're seeing a lot of overlap between SRE and platform engineer Agree engineering. But I think, um, at the end of the day, I'm old, they're all ops people to me.

Right. com Right. I'm a big believer in dev and ops Yeah.

And security and all these disciplines coming together. Um, I do think I'm a half glass, glass half full kind of guy. I do think AI gives us the ability to bring these together and to recognize some of the goals that you just said about why they got into it.

I've been, I was at Chef during the DevOps movement. I've been observability. So you go, I've been in observability forever.

We've been trying to do these things Yeah. For a really long time now. And I think we can actually deliver on the promise.

Yeah. Finally, I, I think having better observability and making the SREs job easier, if you will Yes. Is almost a byproduct of all these things.

Right. But in your case, it's the main thing. I mean, it's what mes mo's about to a lot of degrees.

But I, I look, I think it's a frigging great time to be in the, in the observability SRE business. Right? Now.

Here's an interesting thing to think about. I think I would argue that the only task that an SRE does in a visual ui, uhhuh is observability most, Most, most for what if they use everything else they do is not Everything else is infrastructure as code. They're all doing everything is code.

Like a UI is like not a, a traditional scaling of scaling a system. Now they do it in observability because they're forced to visualize and investigate and complex data. Right.

If we take that off their plate, they can now consume observability in their native tools. Like they can consume it. So What are their native tools, Command lines, slack, uh, cursor, things like that.

They don't have to go into this ui. They can actually operate, not have to switch contexts. They can go do the things that they do in the place that they do it and not have to leave their workspace to go to observability to then go back to where they operate, which is more on the command line.

Yeah. So I'm a Star Trek dude. Yeah.

I think with AI we have the capability to bust out from windows and command lines. The whole human computer interaction Yeah. Is gonna change.

Yeah. And, and so if I'm an SRE, I'm just going to tell the computer what I need to know right now. Yeah.

Right. Right. And it's gonna tell me, and whether it tells me, you know, uh, that I hear or it flashes something on a screen or whatever, the idea of being, having to be able to sit here and type in s**t.

Yeah. Excuse me. I know we're live.

Excuse me. The idea of being able to sit here and type in on a keyboard Yeah. Is history with like, our kids won't type on keyboards Yeah.

To talk to their, or to communicate with their computing devices. Yeah. Whatever they may be.

Yeah. No, I think it's true. And, and so that's going to create a whole new opportunity about, you know, and I don't know, is it something that, do we get wired in?

Is it holographic stuff? Is it just pure? Because some people don't comprehend what they hear they need to see.

Yeah. And I think, I think to that point though, I think there's obviously a, like I, you know, I've been on this kick of like, that's kind of anti visualization, anti dashboard. But, but you do need that for trust and you do need it for verification and you do need it for audits at times.

But I think you could ask, or not, you could ask the model for it. Just say, gimme a visualization. I don't need a, It gives it to you on the fly Persistent dashboard that's sitting around spinning, doing I agree with you.

I, again, and I think you just ask it Nick and it it conjures it up. Yeah. Kind of thing.

Right. You don't need to be spending time developing, you know, the UIs and, and I'm sorry for all my friends who are UI developers out there, but I, I do think we are on the verge of, you know what started, I guess in 1970s, right? Park, Xerox park with the whole window mouse.

Oh yeah. Way of, You know, computer human interaction. We might finally be outgrowing that.

We'll see. Look, I'll settle for, you know, high trust in the diagnosis of mes mo's. SRE agent.

Agent SRE cake. You wanna walk before? I'll settle.

I'll settle for that. I get it. It, I'll settle for That.

It, I get it. Let's Get there. First one Step at a time.

Me, I'm, look, I'm a dreamer, but, um, sorry. com. Did we mention that?

We Did. Check it out. O new brand with the spark and everything.

I love it. With the yellow. Sparks it up.

Gimme one of these again. Cheat it up perfectly. There you go.

Look at it right there, baby. Yeah. All right.

Hey, enjoy. Q Conn. Okay.

Cube Conn. Thanks. It's great seeing you again.

Thanks for having us. Come back. Contact truck TV soon.

Let's hear more. com with a spark here at Q Con. We're gonna take a break.

We got more coming your way. Stay tuned.