Coding AIs in Large-Scale Software Engineering with Augment Code’s Scott Dietzen

This is Textron tv. Hey everyone, it's Alan Shimel. We're back here at techron tv.

Our next guest is Scott Deon. Scott is the CEO of a company called Augment Code. We're gonna find out all about augment code, but let's find out first all about Scott.

Hey, Scott, welcome to Techstrong tv. It's great to have you on here, Alan. Thank you for having me.

So, uh, The pleasure. Uh, prior, prior to running, uh, augment code, I've actually had an interesting journey. I did a PhD in machine learning a very long time ago, uh, when the technology was about symbolic reasoning rather than neural nets.

Uh, but I, you know, I got mm-hmm. Inspired by, uh, Jeff Hinton, who was one of my early AI professors. Been following his Yep.

Been following his career and his large language models emerged. Um, I, I saw this opportunity, uh, to maybe reduce some of the pain that I'd seen, you know, so, uh, multi-decade career of doing really hard system software at places like Pure Storage, uh, WebLogic, BEA systems, uh, and, you know, was always daunted by how hard it was to have crack open a brought a really complex, large repo and make changes to it, uh, and felt AI could help us do a lot better at maintaining these really large, hard software projects. That's amazing.

That's really cool, man. I, I, you know, I know you're not supposed to ask age and all of that stuff, but gimme an idea. When did you get your PhD in ml?

Uh, early nineties. Wow. Wow.

So, gen ai, a GI, all these things are not even on the radar at that point, right? Uh, it was very much about logic and symbolic reasoning. Uh, in fact, um, people like Jeff Hinton took grief from the symbolic reasoning powers that be, 'cause the, you know, they said, oh, if, you know, if you ever get those little toy neural networks working, we'll go in and find the symbols, uh, for you.

Uh, and, and so, I mean, he had the, uh, wherewithal to stick to that research agenda for 30 years until the technology, um, and the performance caught up, caught up, caught up with what, um, you could accomplish. And so, uh, I appreciate that persistence. I, I think we all do, we all do.

Right? In, in retrospect, um, you know, it's, it, it could wind up being, uh, kind of the, the, the, the project of the century, or at least the early part, first quarter of this century. Right.

Uh, um, excellent stuff, Scott. So, you know, I, I've interviewed so many founders and CEOs over the years, and, and the successful ones always have a passion for the mission, what they're doing now, obviously times like, uh, AI and this kinda stuff has been a passion of yours for 35 plus years. So, but I mean, it's ama I'm married 35 years, so I, I could appreciate the, the, the distance there, you know, um, that's a long time to be passionate.

What about augment code's? Mission kind of gets your juices flowing. I'd love to, we should define what is the mission, you know, maybe go there.

Well, let, let me, let me start with a passion and then I'll hit the mission. Um, you know, Okay. I loved programming when I, you know, when I got to do it.

And when, um, I first started in the industry and experienced software engineering at scale, I was struck. I was working with people that were better coders than I was. Um, but the job wasn't fun.

It was really hard. Uh, and it was, you know, um, as you, you get, um, many engineers trying to work on a multi-decade, uh, or at least, uh, you know, a significant scale, tens of millions of lines of code, um, uh, and it becomes really unwieldy. It can be fragile, um, extremely complex to understand what's going on, uh, require a tremendous amount of coordination, uh, to get those changes, right.

Um, and so as large language models emerged, it was like, wow, is there a way to attack this complexity? And the products that were in the market, uh, in those days, like the early, uh, GitHub copilot, for example, um, struck me as much more targeting simple programming than software engineering, where you really trying to, uh, deal with a large, complicated piece of, of software. So that's what we wanted to go after.

We wanted to, um, bring, uh, the full power of understanding of your code base, uh, to bear with ai. Uh, when you get started with most of these coding ais, they're like a new college grad, which I once was a long time ago, uh, where, you know a little bit about programming languages and algorithms, but you don't understand the software that you're trying to work on and the environment in which it runs. That's augments differentiation is that we come in with comprehensive knowledge of your software, uh, without ever training on your coat.

Love it. Again, you know, this is a, a mission a, uh, a company that couldn't exist until, until it was, the technology was there to support it, right. The, the, the horsepower and, and everything else.

Um, tell us a little more about augment code. I mean, you're the CEO, but how big, and don't, again, I'm not asking you to say anything, not kind of public, but you know, who's the customer base. Tell us a little bit more about the company.

Yeah, so, um, we only launched the product, uh, you know, within the last, uh, six, six months or so. Um, it's now available for anyone, uh, to use free of charge. Um, and, uh, it, it really impresses people pretty quickly, uh, in terms of, you know, giving them knowledge of what's happening inside their software.

So, for example, one of our customers Lemonade, uh, they've got, you know, uh, north of 10 million lines in a mono repo. Uh, and, you know, they're constantly moving engineers around that project, uh, bringing new engineers on board, uh, and augment helps them orient into the, into this new code base for them, or new area of the code base really quickly, uh, so they can come up to speed without having to put a load on the, on the senior, uh, employees. Uh, another, uh, company, uh, COUM that uses us, uh, they do migrations, uh, on upgrading of e-commerce AppSec.

So they're constantly jumping into new code bases and needing, uh, to find their way around in order to make changes, uh, that, you know, they're getting an order, uh, 50% acceleration, uh, for their workflows because of all this knowledge augment brings. So, you know, even though we're new, we've got hundreds of, uh, companies now using the product mostly in the tech industry, software and intensive companies, and, and once you try an AI that understands your software and it's environment, you never want to go back. So they say, um, that sounds great.

Um, what's the plan to commercialize here? Well, uh, you know, the, I I think the game is changing around AI for software. You know, the, the, as I said the earlier ais are o novice, uh, now everyone wants that contextual knowledge.

Um, and, uh, we're in a unique position to be able to deliver it. I, I think the hope is that we can pay down a huge amount of, uh, software debt. You know, if you look at, uh, just the United States market alone, we lost, uh, two and a half trillion dollars, uh, last year due to software failures.

I mean, there's not a piece of software I've ever encountered that doesn't have a long list of features nice to have, as well as refactoring in tech debt that people would like to fix and change. Now, I see this opportunity with, um, ais that understand your software, like augment, uh, that we can deliver the software of everyone's dreams. We can make it more reliable, easier to use, uh, secure, more fault tolerant.

I think all of these things can come out of, um, these much more proficient ai, like the one that augment is delivering. Hmm. You know, I'm trying to remind, I was trying to think.

There was a testing company, I, Gil Sw was the founder, CEO, and I'm blanking on the name, but they used to test for the, they'd look at a website and tell you how a person would see it, you know, based a person, not see a real person, didn't see it mind you, but they would tell you what a person would see, right? It was like optical something testing kind of reminds me of that, right? Where without a real human being in there, the AI is, is telling you what the human experience would be, so to speak.

And, you know, and, and that's an amazing thing. If we could do that, it's scale, you know, and we're talking large scale. Um, and that's one way, you know, that it's one way where we're seeing AI kind of change the game.

Yeah. But a, a great way to think about this is, you know, humans are really good at thinking over the long term about what they'd like to see in a piece of software, right? Do I wanna change to a microservices architecture?

Do I wanna move this application to the public cloud? Um, what, you know, is there a mobile interface that I wanna develop, uh, for it? Um, what we're less good at is all of the incremental, uh, steps that need to happen in order to accommodate those changes.

You know, especially in a large repository, you know, e even a simple, there's no simple change in a large repository, right? You wanna add a field to a data structure. I mean, you, you, you, if that data structure is persisted, you've gotta update, uh, the database schema.

You've gotta, um, change any of the command line or other APIs that use it. Um, SOIs can really help. Like if you open up augment code and you make that change, it will offer to pull that change through an entire large repository, potentially touching dozens of different files, um, and guiding you through that process.

And if you get interrupted in the middle of it, you don't have to go back and struggle to try to remember where you were, because the AI will help you pick up right where you left off. Yeah, absolutely. It, it, it's, it's a wonderful thing.

You know, we were talking, uh, on our Textron gang show last week that, you know, they, they say there's about 27 million developers in the world today, software engineers. You want to say, no, there's 30 million, all right, I'll give you 30 million. You want, tell me there's 25 million.

I'm not gonna argue with you, but somewhere in that range As a result of using ai, they say, look, by 2030, we may have a half a billion people who are, I don't know if you really wanna call 'em software engineers, but, but they're asking AI to develop code for them. So the amount of code that's out there that's going to be out there, you know, you're talking about exponentially more code and AppSec and stuff floating out there that really to the point where I don't think humans wrap their heads around it. It's like trying to say, oh, that's 30 million light years away and this is 40 million light years away.

What's 10 million light years to a human? Right? Uh, same thing with the, just the sheer amount of code that's gonna be generated by these AI and everything that really only another AI is gonna be able to sort of wrap their head around, you know, speaking that way.

Kinda wrap their head around it and, and make, you know, heads and tails of it all. Yeah, I would say, uh, two things here. One of the challenges with the, with this, these novice ais that I was mentioning earlier, is because they don't understand your code base, they're really happy to add new code.

Like if we were building something together and you had added something last week, um, if I start trying to add the same thing, most AI will be like, oh yeah, sure. Let's add, you know, yet another class into the system. What augment will try to do is say, Hey, why don't you reuse, uh, the code that Alan wrote last week, rather than try to add new code?

Let's adapt and reuse rather than proliferate code, uh, inside a repository. And one of the most exciting things I've seen is when, uh, augment wanted to delete code rather than add code into a repository, because that's how we can get to higher software quality. Um, and, and so most repositories I hope, can shrink as we make them better and more secure and remove dead code.

But I think you are right that, um, you know, the barriers to delivering software are, are getting reduced and the economic returns that can come from having all of the software that people want, software that can do any different task, uh, that we imagine, uh, I think that can unleash a lot of human productivity and, and wealth and make many more of us, uh, able to, uh, create, uh, the software of our dreams. Agreed. Agreed.

You wrote a blog article a little bit ago, Scott, about, you know, this kind of the, the impact this is gonna have. com or what's, what's the URL? com, And so they can get the blog from there, I assume?

Uh, yes, absolutely. It's, uh, on the top nav on the side. Excellent.

Tell us a little bit about maybe this blog post and some of the concepts there. So we've hit some of them already, right? That I, we really wanted to tackle the discipline of software engineering rather than just, uh, toy programming.

Uh, and that, you know, real software, you know, complicated software lives a long time, uh, you know, and it comes with documentation and, and testing and, um, you know, a lot of the development happens outside of, uh, integrated development environments. You know, um, engineers use, uh, tools like Slack often to communicate, you know, it augment, uh, slack can or augment, can join a Slack conversation. So if you're having a debate about how a piece of code is supposed to work, um, you can invite, augment, uh, into the Slack conversation.

It will render a verdict, and you can even assign a pull request right out of, uh, slack, uh, to, to augment, uh, and have work done. And so there's this opportunity to go beyond, um, just the development environment, uh, and, and tackle kind of wherever software engineering libs. Um, another thing that, uh, people often find is, you know, that, that the documentation is inconsistent, uh, with the software or somebody is coding something outside of security policies, um, uh, inside of a code base.

Augment could pick that up. Like we can identify when documentation is not current, uh, incorrect, when there is a secure security policy violation, uh, and then lead, uh, engineers in the right direction toward delivering a, a better product, more reliable software. Love it.

Very cool stuff, Scott. We're about outta time and these things go quick, I apologize. But first of all, congratulations.

You know, I lifelong programmer, right? Coder doing the CEO thing, right? It's, it's exciting, it's fun.

And, um, what, and it's also the fact is I, and I tell this to people all the time, what an exciting time to be in, in the software engineering world right now, right? With all, and not just software and in tech in general with what, what I is is, you know, the the promise of, of changes to come and, and new, new vistas to explore. Um, you know, keep us posted, keep up the great work and, and we're interested.

It's gonna be interesting to see how this all shakes out. Absolutely. I think we are in an extraordinary time.

A, we've gotten to see internet and mobile. Um, I'm convinced AI's gonna be bigger than the combination of them. Uh, there's so much potential to, uh, unleash human creativity, uh, that we are on the cusp of, so can't, can't wait to keep you posted on all of that.

Of course, you saw this all coming when you were doing your PhD back in the early nineties. Not even close, but I'm really glad to have not even Exactly. Glad be Is better to be lucky.

That better to be lucky than Smart Scott, right? No question. All right, man, we'll see you soon.

Thanks again for being here. Best of luck with augment code. Thanks Helen.

Appreciate you having us Alrightyy. We're gonna take a break here on Techstrong tv, Scott Deacon, CEO, of augment code here, and we'll be back with more. Bye-bye.