The Evolution of Akka and the Future of Agentic AI with Tyler Jewell

Hey, everyone. Welcome back here to Text Drunk tv. I'm really glad to have my, uh, next guest on here.

He's someone we've talked to before. I don't know if it was under this role, but let me introduce you to Tyler Jewell. Tyler is the CEO of aca.

Tyler, welcome to Tech Drunk tv. It's great to have you here, Alan. It's good to be back.

I think this is the third time we've talked. Absolutely. Absolutely.

Tyler, we're gonna get into who ACA is and we're gonna talk a little agentic ai, but first let's talk Tyler, right? Yeah. Because maybe you and I have spoken three times, but probably, you know, not everyone in the audience listens to every show.

I do imagine that. I Mean, that's gotta hurt your feelings a little bit there. It does, it does.

I have to tell you the truth. I was doing a webinar the other day with a live round table with live audience, and some guy got on there and said, uh, greetings from Boston. I've been a shimmy fanboy for years.

Love coming. You know, listened in on everything Alan does. And I was like, I didn't even pay him to say that.

So I, you know, I don't know who it was, but whoever it was. Thank you. Um, I don't know, does it count if it's anonymous?

Well, it, it, it said Rob CI felt like I was in Oh, I, you know, I was on one of those, call it radio shows. But anyway, Tyler, give us a little sense of your journeys out, you know, where, how you got here today. Oh, mine, uh, well, you know, I'm the CEO of aca.

This is the fourth tech company that I've run. Um, you might have been familiar with a couple of the others. Uh, WS oh two, which was an open source ESB vendor, got acquired by EQT last year for 650 million.

And before that, ran a company called Covy that I had started. It was a cloud IDE. It was a, uh, an early generation, um, cloud IDE, which was kind of competitive to Rept before Rept had come along.

And that was acquired by Red Hat. And I've had 15 years of venture capital and angel investing almost all exclusively in developer experience, developer platforms and developer integration technologies. Um, and I publish, uh, a blog where I keep track of all the DevOps companies on the planet.

There are 1700 of them. 17. You know what?

That's a little, we, well, when was the last time you updated that? It's been 18 months. Uh, but, you know, all things being equal, I think that the number of companies is about the same because, uh, I've seen just as many companies fold or go under as there have been new companies that have come along.

Uh, I, it's, it's a little bit tricky. I don't necessarily include all of the AI generated developer techs, platforms, open source projects that are out there. They've all kind of flooded the market a little bit.

And so it would be, I wouldn't call it, uh, polluting the landscape, but certainly, you know, uh, lots of threads to pull that could cause it to, you know, blow up in terms of total count. Absolutely. Absolutely.

Tyler, you know, it's been an amazing Jersey. Uh, when did, when did you come in here to aca? Uh, I made, uh, an investment into ACA in 2019.

Joined their board, then, uh, been on the board since that time. And, and ACA is an interesting company. It used to be known as Light Bend.

Uh, light Bend was a company that had, uh, raised a quite a bit of money from a lot of tier a venture capitalists. And, uh, they ran into some problems in 2021 and we're open about that. And they didn't have a viable business model.

Um, uh, they didn't have a good product plan. And, and the company faced a near bankruptcy event. And, uh, we actually had the investors step in, restructure the company, uh, change the product plan, change the business model, uh, and we've, uh, also updated the management team.

And it was all about a year and a half ago that I stepped in as the CEO, uh, because frankly, we thought there was a huge opportunity, and I was excited to get back in and operate again. Very cool. Very cool.

You know, just thinking back, Tyler, you said three times, I, Ima I, I, I imagine I've interviewed you many more than three times. Oh, well, I don't know. I mean, maybe that was over beers.

Did the beers and wine count? Well, Maybe you're right. You're okay.

Maybe, maybe. Um, so, you know, let's get this out of the way. AKA is spelled a KKA.

Yes. io. Okay.

Tyler, before we jump into the topic of discussion today, which is around, you know, uh, deployment models for Agen AI systems, audience out here says, I, I think I heard a light bin. I'm not a hundred percent sure, but who cares that that business changed anyway. How would you describe what ACA does for them?

Well, you know, AKA was originally an open source project invented by Jonas Bonnet, a, uh, Swede, who was also the founder of the company, and he's still our CTO. And Jonas is a bit of a savant in terms of distributed systems technologies. And originally, ACCA was a development framework for designing, building and implementing large distributed systems, distributed systems that could process, uh, uh, hundreds of millions of concurrent users terabytes of data in real time.

And to do this with very consistent SLAs, whether it's a latency SLA availability or recovery. And, um, over the years it's evolved and it's been deployed over a hundred thousand times. Uh, there's more than 2 billion people on the planet who use an application that's powered by ACA every day of the week.

Um, there are systems that are powered by Datadog, Netflix, Amazon, um, oh, let's see, Starbucks, uh, 50 different banks around the world. A lot of automotive groups like General Motors, Tesla, uh, apple, all make use of ok. And, uh, what we've evolved it into is it started off as a developer framework, and now it's an enterprise agentic AI platform.

And enterprise agentic AI for us is when your IT systems, uh, meet your reasoning systems, but you still need to maintain all the ilities, whether it's availability, scalability, performance, safety, security, pick your poison of that. Um, and, uh, ACA has a number of large scale, uh, enterprise AI deployments associated with it. And people use ACA to do, uh, model inference, rag based systems, uh, planning, uh, planning and execution based systems, uh, large scale, uh, a personalization where you're doing streaming from a bunch of different sources.

And you need to inference do a lot of inference off that in real time. Uh, and AKA's able to do this, uh, with systems that run up to three or 4 million, uh, transactions per second and offer global, uh, disaster recovery and failover at the same time. Wow.

So we're talking three, 4 million, uh, a second. Are we talking mainframes or is this still sort of distributed cloud sort of, or maybe in a data center? Uh, Uh, so it's, uh, it, it has to be truly distributed to get to that level of, uh, uh, transactions per second.

And effectively the way ACA works is, it's, uh, it's, it, it's runtime is an actor based runtime, and actors are some, uh, concepts from the 1970s in computer science. And effectively what they are is it's a concurrency mechanism, uh, that also allows for message passing between those, um, those systems. And so you can get high levels of concurrency, but also isolation with them.

And effectively what happens is when people build systems with aca, uh, that's generally gonna be some sort of stateful system. ACA nodes, uh, take responsibility for all the data. Uh, they scale themselves out on as much compute as you can give to them, and, and they, uh, manage the data in memory, um, as opposed to having to go to some sort of database or some sort of persistent store.

And when you do that, um, it does it in intelligent ways. Now, end users, as long as they can get to that right mesh, um, they get almost instantaneous responses. Uh, and it's also non-blocking with asynchronous communications.

So it not only scales horizontally, but you also get this very consistent instantaneous response for any type of transaction that you might want to do. Uh, and so ultimately people use ACA for when they have large volumes of concurrent users, like, uh, wiggy, which is, uh, I'm sorry, uh, um, I was gonna say Dream 11, dream 11, which is like the DraftKings of the Indian market. Uh, when they have a cricket match, they have 70 million users all gambling on every play simultaneously, and they have to offer a 17 millisecond response time for each of those users.

Otherwise, a user gets an advantage on that. And that access is not just static content, there's actual transactional content. 'cause you're looking at wallets, you're looking at player stats and whatnot.

Um, so that's an example of that. Other, other examples is when you need to process, um, uh, large volumes of data in a very consistent, uh, uh, way. So, uh, you know, uh, with Datadog, they do all their data ingest on aca, so they've got lots of streams of data coming from across all their different customers.

And, um, it's coming in a different volume, uh, volumetric rates, and they have to process it in a very consistent way, despite no matter how much data's coming in. So as long as they have the bandwidth to suck in that raw data, then they have to do the transactional processing on a inconsistent, inconsistent means. Got it.

Interesting stuff. Really interesting stuff. You know, one, one of the, I mean, I love doing what I do, Tyler, and part of it is because this isn't something you read about necessarily in, you know, we, we, everyone's talking AI today and we talk about AI and everyone's talking about agent ai, and you know, there's been so many conferences this week between Google and Red Hat and Microsoft and Dell, as, as you probably know, everyone's talking agentic ai, but this is yet another window, another facet of how AI, agentic AI is being used or can be used, right?

And, and in this case, these huge distributed systems and, and there ways maybe that we didn't hear a Google IO yesterday or something like that, right? Yeah, yeah. Um, so how is this, I'm gonna ask you the same question I ask a lot of the folks when it comes to AI though, where's the killer app and how, how real is this today?

Well, you know, the, uh, uh, you know, I think that we've seen the killer app already with chat, GPT in, in that people who have engaged with chat GPT is it, it is able to, um, uh, gather information, synthesize that, um, and reason on that information to some certain degree. And, and that that raw capability of an LLM, that ability to do some basic reasoning, which is, uh, just word matching that has broad, broad, um, uh, uh, you know, existential implications to every IT system on the planet. Uh, and, and so it's not just the killer use case, the kill the, if there is a killer use case, it's this raw reasoning ability.

And what I mean by raw reasoning is that, uh, what an LLM is capable to do for you is it can take a series of inputs and it can give reasonable outputs on that. And you can apply that to your IT ecosystem in a lot of different ways. Like, for example, you can say, here's a document, here's a template and I want you to fill it out given some context on that.

And so, you know, as long as you gave it the right structured input, it can generate a reasonable output on that. Um, you can sit there and ask it to, like, Hey, I need to categorize this data into three different buckets, and here's all the rules for that, and it can reason itself out and do that work for you. And so you can start extending this into really interesting ways, like, Hey, I think I wanna get some information, um, and there are four different systems I can call to get that information, choose which system I should call, and how do I call it?

And it will tell you, this is what you need to do. And so, you know, at that low level, those are the foundational building blocks of how you build a system that reasons at the end of the day, and now it can take those basic building blocks using LLMs and inject them into, um, their existing systems. And what ends up happening is that when you do that, you intelligently get, um, now your old school rule-based systems that have all this deterministic rules on how the business is gonna work alongside your non-deterministic reasoning based systems that are coming up with it dynamically on that.

And so with that, you can do things of completely change the personalization experience because you can have it reason about what kinds of things that needs to synthesize for the person that they're interacting with. You can automate things that had really complex rule sets that were too difficult to maintain. Now you can have the LLM automate those things for you, and you can also have systems that adapt to their environment.

Um, and that sounds all light and fluffy, but adapting to the environment, for example, would be, um, looking at a user's clickstream, monitoring that click stream and then telling an agent to change the goals that it's trying to do. If you're trying to personalize something like, like, this is what we do for Tubi. Um, so Tubi is one of our customers.

They take a live, a live, you know, activity of what the user is looking at, process that in a bunch of models, and then change, uh, the goals of the recommendation system and the recommendation agent that they have. So those, wow. You know, so I, I think AI is big.

I I think that nobody's really adopted it in the IT space yet. Early, early stages. Uh, but I think every system is gonna go through a transformation.

I like to say we're at the beginning of the beginning, not even the end of the beginning. Um, so I, I, I agree with you there. io.

Um, who would you recommend, like, you know, hey, this is a great website, if you are in this, you should come here and check out what we're doing. I, I think that, uh, for us, what we're saying is like if you are, um, experimenting with ai or you're at the point where you need to start bringing AI into your IT environment, come check us out. Um, we do 48 hour POCs for you.

Uh, bring us your use case. Uh, we'll educate you on what AI means and we'll actually implement it and show you what it's like. That's a killer app right there, my friend.

Good for you. Thank you. I love it.

Hey, Tyler, continued success. Good luck with aca. Keep us posted please.

Again, it's Akka io here on Tech Drunk tv, Tyler Jewell. We'll be back. Stay tuned.