Mezmo Open Sources AI SRE Operations

Hey everyone, we're back at PlatformCon. I actually just came off the stage maybe a half hour ago, where I did my panel on Mythos and what it means for platform engineers. But I'm really happy to have our next guest with us.

He's been with us probably at least two or three times- At least ... I've interviewed. Yeah.

Tucker Calloway is the CEO at Mesmo. You've been at Mesmo, what, about six years? Six years, yeah.

Long time, man, right? Yeah. We've both been through this whole historic timeframe we're living behind.

Yeah. But Tucker, let's start with you. How did you come to Mesmo?

Give us a little bit of your background. Well, it's interesting. I grew up in the DevOps SRE space, through Chef was where a lot of my really formative experiences were, if you remember that, in the early days of DevOps.

And so it's kind of interesting, I started at LogDNA, now Mesmo, and as I was saying to some people earlier, I feel like this is my third job- At Mesmo ... at Mesmo, right? It's all been the same job basically, but it's been a third evolution.

And the last 12 months really have been like this opportunity to just reinvent. The world's rewriting itself, and so the last three months has been this opportunity to reinvent everything we do from scratch, like how we operate internally, how we think about providing value to our customers, and what the future of the world's even going to look like. So it's been fascinating to go through.

It is. Yeah. It is.

That's crazy. I think a lot of people out here are feeling the same thing though, Tucker. Maybe not three times in six years.

Sure. But this last year- But at least the last year ... year and a half- Yeah ...

we are all trying to figure out, who moved my cheese? Yeah. How are things changing?

What's my role? How do I excel? Right?

And you and I were talking off camera, it's not about losing your job to AI- Yeah ... but how am I leveraging AI to be more effective? Now, with Mesmo, as you mentioned, DevOps, SRE space.

The last time I saw you, and granted, it was probably 18 months ago- Mm-hmm ... maybe a little more than a year, we were talking about the AISRE. Right?

And that seemed pretty radical. Mm-hmm. Yeah.

I remember thinking to myself, "Wow. " Right? Yeah.

Here we are a year and a half later, and that's not enough. That's old school. It's old news now.

Yeah. That's old. Yeah.

We're way beyond that. But for purposes of our audience, let's start with that. What do we mean when we say an AISRE?

Well, there's probably two or three ways to take that question in my mind. There's one is, what is it? Right?

Then there's, what does it mean for the SRE? How does it change their role? And the third is, what's the future of that?

How's that going to look? Right? Mm-hmm.

And so to me, the first thing is, what is it? To me, it's a capability that's basically, the job of the SRE has always been about removing toil. That's the foundation of an SRE.

That's the thing, that's the ethos they believe in the most. Yep. And so to me, the AISRE capability, it's just doing more of what they've always done, which is, how do I remove toil through automation?

It's just a new opportunity and a new architecture to do that. So that's what, to me, AISRE is just the thing that they've always been trying to do, and giving them new ways and new architectures to do that. So the next is then that how does that change the role?

And there's all this fear, do we not need SREs anymore? And I think that's this crazy dystopian AI future- Yeah ... that I don't know.

I don't know what to say about that. No, some people buy into that. Some people buy into it.

I don't know. I don't buy into it, but of course- You wouldn't do it ... But I think the opportunity is for the SRE to become more of a reliability architect, so to be able to design more systems, design and build systems that leverage the new technologies and capabilities, which they've always done.

They were the ones that led us into DevOps. And I think the SREs are a role that will lead us into this new future. It's interesting to me that the majority of the attention in the market goes into building software when it comes to AI.

But actually, the biggest opportunity is actually running software. Because to me, running software includes supporting it, maintaining it, observing it, securing it, all that stuff. Amen.

So here we are with Cursor getting bought for $60 billion. $60 billion. All star.

Yeah. Yeah. Now 30.

Right. Or whatever it is. Maybe there was an old Still, yeah, still a great acquisition though.

Yep. Right? But the big opportunity is actually how do we run software, and that's what I think the AISRE is, one of the first vignettes into that space of running software.

So then I come to my third point, which is, how will people consume this? And I think right now, everything is kind of a bolt-on to an observability platform. And our mission is to create the operating environment for people to create agents, for the builders to create.

Right. Right? So instead of just consuming this outcome, and there's some great products out there that are delivered via SaaS that give you an AISRE outcome, our focus is on the segment of the market, which I think is a bigger segment of the market, that's going to go build AISRE.

They'll look at it as critical infrastructure. They want to own it. They don't want to buy it from a vendor.

It needs to be transparent. It can't be a black box. They want to control their LLM costs.

You get it all into kind of token economics and stuff like that. So we're super excited about that. Let's dig into that, though- Yeah ...

because I think you're onto something. Well, you probably think you are, too, because you've put the whole company on the line. Yeah.

But you're pretty invested in this. I'm all in. The hen and the hog, right?

Yeah. The hen wants to give an egg. Yeah.

There. Yeah. The hog has more of a commitment.

I'm making it. Right, exactly. But that being said, Tucker- First of all, look, couple of things.

I do believe that we're entering the age of the engineer. Yeah. Right?

Developers have to become software engineers. Mm-hmm. It's not enough to be a coder.

No. AI's going to code. You got to manage that.

Yeah. You're a builder. You're a software engineer.

You're a reliability engineer. I think with things like quantum engineers, fusion engine- Yeah ... the world, it's an engineer's world- It is, yeah ...

is what we're coming to. Yeah. Secondly, though, I think you're right.

There have been last gen tools that SREs have used, and I'm not naming names here, but they were good tools. They're great tools. Yeah.

They built good businesses. Yeah. Some of them have great exits in public and all of that.

However, this is a pattern I've seen in tech my whole life. The tendency is take the next shiny new trinket. First, it's a standalone product, but then it becomes a feature in the platform.

Mm-hmm. And it's very much bolted on. It's not organic.

And that might work for a certain segment of the market, right? Some people are happy with that because they get to stay in that tool that they're comfortable in. Mm-hmm.

But then you want more. You want more, you want native functionality. You don't want that bolt-on, because the bolt-on, at the end of the day, it's a bolt-on.

Look, I totally agree with you, and it is a pattern as old as time that new technology comes out, it emulates the form of old technology in its initial adoption, and then over a period of time, it evolves into what it should be. Right? And so the interesting thing now is we have effectively unlimited intelligence and unlimited coding capacity, right?

Like you could use the LLMs, and cloud code, and all that stuff. We have infinite dev capacity. We have infinite intelligence.

And so what do those two things... That used to be what we provided as SaaS vendors, right? Yep.

We coded intelligence into an offering so that you could consume that. So now the opportunity is that has become a utility, so we will build differently. There are opportunities to take those things and build them into embedded workflows inside an enterprise, and that needs to be open source.

That needs to be something that people own, and that's not consuming an outcome, that's not consuming SaaS. And when people talk about SaaS being dead, I believe SaaS is dead. I don't think vendors are dead.

I think- No ... SaaS is dead, and what happened? Right.

I think the way we consume SaaS and the way we think about SaaS and the way we, not monetize, but expense SaaS. Yeah. That's right.

Yeah. That whole CapEx, OpEx thing that drove SaaS, I think it changes in this world. But here's the other thing.

You talk about open source. There were open source tools for SREs- Yeah ... that really they became SaaSified because companies used that.

SaaS became a business model. Yeah. Not necessarily the best delivery method or something.

That's right. Yeah. But open source.

Mm-hmm. That seems to be the thing. Well, Mythos, Fable, don't worry, the open source models will be here in six months, right?

Yeah. Open source is like the great, what's the emancipator- Yeah ... if you will.

Equalizer. Right. It's going to set us all free- Yeah ...

to run and do what we want and not be under the thumb of that government, this government, this oligarch, that billionaire. Mm-hmm. Right?

Because once it's open- Mm-hmm ... you know? You can't stop it.

Yeah. Right. Yeah.

Is that part of Mesmo's thought process here is, hey, once we put it out in the open, people will recognize the power of it? I think so, yeah. I think there's also a big element for us, too, that is, I think a vendor now is a derivative of what the vendor used to be.

Mm-hmm. We used to provide intelligence. We used to provide a UI.

We used to provide a data store. We provide governance and data management, all that things. And we package that up into an opinionated way of delivering an outcome to everyone.

And so now the enterprise has an opportunity to take on some of those pieces on their own in a way they never have before. Yeah. So I still think that, for example, Aura, which is our open source AI SRE capability, is an opinionated harness, right?

So you take that opinionated open source harness, and you can pick whatever model you want, local model, pick your Mythos, Fable, whatever you want to do. And we built opinions in terms of how memory works, how the evaluation loops work, and we take all these things that you would have to build on your own and put them into the model, and that's what a vendor is now is actually providing that harness or providing the opinionated layer on top of AI instead of providing the whole thing. Because now companies can take that harness and build it into their workflows more effectively.

So to me, it is a democratization thing. Our expectation is that it takes off, but it's really more like being really thoughtful about what does a company need to own in the new world and what does a vendor need to provide in the new world, and it's different. It is.

Yeah. And I think that whole vendor-customer relationship- Yeah ... whether we're talking about open core, open source with premium models- Yeah ...

SaaS, what have you, that is radically changing. Like your world's changed. Yeah.

All of our world's changed- Yeah ... in the last year. Yeah.

That's changing there, too. Let's talk about Aura. Yeah.

So at its core, it's open source. Open source. Apache 2.

Yeah. If I didn't want to use Mesmo's premium features- Is there enough there for me to make something? That's one of the principles, is that it is a strategy for us as a company.

It is also something we're giving to the world. And for both those two things to be true, it has to be independently valuable. All right.

So it is independently valuable. We think we can make it even more better in some of the things we do. So for example, one of the highest value things we do is intelligence and signal engineering out of raw telemetry data to process that specifically for agents.

That's something we charge for. We have an infrastructure and architecture that does that, so we think it's a valid contract with the community. But if you don't want to do that, then it will do all the AI SRE tasks for you on its own, and you can go use it, and we hope you do.

Is the future of Aura to be something that is strictly like Mesmo managed, or do you foresee maybe another path where it becomes part of one of these open source foundations or something? I think if we're successful, it becomes part of the open source foundations. Yeah.

It's not our aim to hold it, control it. We're actually the most interested in the governance, memory, and signal intelligence processing layers, but we just know that in order to get to those problems, we've got to help the market advance. That makes...

Yeah. We need to help the builders get out of the SaaS consume solutions and observability vendors that they're with today and advance them into this new world. And so that's our aim with Aura.

We built it for ourselves for that reason and then decided to open source it, and we think that gets the market advanced faster, which suits us obviously, but it suits the market as a whole. As well. And so I can absolutely see us turning it over to a foundation.

But I think that's an important thing, right? You guys built this for yourselves, right? Yeah.

Then realized the right thing to do is put this in market and let it- Yeah ... find its level. You mentioned one big freemium or premium type of add-on.

What else? Is there anything else in there right now? Yeah.

There's three big categories we're looking at. So, the first is opinionated memory. Mm-hmm.

And so basically, taking the AI SRE task and creating specific memory of the memory format for that which we'll release in the coming weeks, maybe let's say the coming months, like in the next two, three months we'll release that as a commercial offering. What that does is it makes it so you don't have to use generic memory, but you can take... The generic memory basically as it gets context engineered is very lossy because it's just doing summarizations.

In our, we have an opinionated framework around investigations as a data primitive, and when you do that, you actually get a better long-running memory capability with less lossiness for the specific task of AI SRE. So that's one. The second is, Aura as a execution framework has human in the loop.

But over time you're going to need to manage how do I escalate trust, how do I elevate and escalate trust in those agents as they deliver outcomes for me? So how do I manage permissions on one side and then trust on the other? So that would be a capability we'll offer.

And then of course, the signal processing I mentioned- Right ... before. The first one.

Yeah. Those are the three. Yeah.

I want to come back to Aura. Mm-hmm. Obviously, probably on GitHub, I'm assuming.

Yep. You may not know this, but we'll try to put it in after the fact. Yeah.

The GitHub... That's Mesmo/Aura. Okay.

Yeah. Easy. Yeah.

com/mesmo/aura. That's right. And it's A-U-R-A?

A-U-R-A. com, and the first link- And it'll take you there anyway ... there is taking you to the GitHub repository.

Yeah. And you could get started right from there. Get started right from there.

We actually just last week released the init and onboarding function, so it'll get you going, discover your MCP servers, and you should be able to run AI SRE task in, like, 15 minutes. I love it. Yeah.

Me too. Tucker, man, I- Yeah. Thanks for having me ...

don't wait 18 months for me to catch up. God knows what you'll be doing. I know.

I know. com. Aura is the open source.

Check it out. We're here at Platform Con. Look, these are your people.

I'm sure you're having a great time. I'm thrilled to be here. Yeah.

Absolutely. Yeah. We'll be back with more.

You're watching Techstrong TV.