Developers, Take Note: The Open Source AI Coding Agent to Watch

Hey everyone. Welcome back here to Techstrong tv. My next guest is Scott Brighton.

Other, uh, Scott is the co-founder and CEO of a company called Kilo. I believe it's Kilo ai and we're gonna hear more about that in a second. But let's hear more about Scott.

Scott, welcome to Text Drunk TV man. How are you? I'm good.

How you doing, Alan? Good. Um, before we start, you know, before we talk about Kilo, let's talk a little bit about you, Scott.

You're the co-founder and CEO there. What, what's the story? What's your story?

Well, my story, um, so I am based in the tech hotspot of Washington DC mm-hmm. Um, I guess for former former life, um, my kind of first founding experience, I started a data and analytics consultancy called Brooklyn Data. Um, implementing, uh, data strategies and tools like Snowflake.

DBT kind of built that up from, Let, let me just ask a stupid question. You didn't do Brooklyn data in Washington DC I started in, in Brooklyn, so, alright. I, I am, I started in Brooklyn too.

Perfect. Oh, born there. I started my Brooklyn apartment where I was living at the time, pre COVID.

Uh, and then March, 2020 I scoodle down to, to Washington DC to get, um, free babysitting from my parents. Well, that, that's always, Hey, look at the cost of babysitting. It's, that's a good deal.

The, uh, yeah. So I started, um, Brooklyn data, um, essentially helping companies navigate, you know, the modern data stack, uh, build that company, uh, to about a hundred people and sold it. Um, very cool.

And it was, it was a fun experience. And, you know, after I sold and and left the company, I stepped back, I was like, shoot, was that like the biggest thing I'd ever be a part of? And not just building the company, but just like this, like transformational wave of the modern data stack where everybody, you know, who was working with data, who's kind of like life was transformed.

Um, and that's when I met my co-founder Sid, who, who kind of, we got connected and, um, started talking about Kilo and and agentic engineering. And I was like, aha, you know, I will be part of something bigger and it will be 10 a hundred times a thousand times bigger. And that's kind of, you know, AI in the agent engineering wave.

And so it's kinda, this is the, the second wave I'm riding. I love it. Um, you know, you're right though.

You, uh, no one wants to hear that they peaked at 32 and it's all downhill from there, you know? Uh, I, I get it. And then of course, you, you referenced your co-founder Sid.

That's Sid ndi. Mm-hmm. Right.

Uh, and our audience knows, well I haven't had Sid on the show in years, actually, I'll tell you the truth. But of course it Sid surround was the, uh, founder of GitLab. That's right.

And Sid and I go way back to when he founded GitLab, actually we used to. Wow. Yeah.

I used to go and we used, I forgot what they used to call it. Not GitLab days, whatever their user conferences were. Yeah.

They were small. We actually did one in Brooklyn together. Oh wow.

Okay. In Williamsburg at the Williamsburg Hotel. Cool.

Back when Williamsburg was just starting to be cool again. You know? I remember that.

Yeah, man. We did in San Francisco. I used to have sit on here every other month.

com. Yeah. Early in the DevOps revolution, if you will.

Uh, so give Sid my regards though. We'll Do, I'm, I'm chatting with him right after this, so, so I'll know you Said, always said hi. Um, so let's talk about Kilo, right?

Kilo ai. What's that about? Cool.

Well, kilo is, um, the most popular open source coding agent. Um, we're an agent engineering platform that helps, um, kind of engineers, uh, you know, move at, you know, what we effectively affectionately call kilo speed. Um, okay.

Which is, you know, I like to to describe kilo speed as you're driving to work. Uh, and you only hit green lights. And that's like the feeling that we feel like you should get while coding, especially with ai.

And that's kind of our mission is to enable kind of engineers to move at kilo speed. I love it. What a great way of explaining that.

Man. You know, I live down here in south Florida and it's seasoned down here. All those folks from Brooklyn and Greenwich and everywhere else behind, You're hitting a lot of Red lights.

I'm hitting a lot of red light. 'cause they stop, if the light looks like it's gonna turn yellow, they start breaking and um, yeah. It's crazy.

But anyway, how, you know, 2025 for many of us in the industry was going to be the year where agentic AI kind of takes the spotlight mm-hmm. From generative ai. And as we, as we sit here on the cusp, you know, well, it's not just the cusp, we're, we're in 2026, I think a lot of us realize that our agentic experience in 2025 left a lot to be desired to say Agreed.

Yeah. Why is 2026 gonna be better and how does Kilo make that happen? Yeah, I think that's fair because you see all these articles of, you know, you know, people actually moving slower with AI or AI projects failing.

Um, and I think what we've seen when we look around is that, um, the, the promise of AI isn't, isn't kind of clicking because, um, the tools are kind of working against us. And so, you know, if I look at, so say like the, I guess the guess the cursors of this world, um, that are essentially, you know, I would say selling, uh, subscriptions for a consumption-based product. And so the dynamics are the more you use cursor, the more a cursor loses money.

And so they're essentially throttling you when you hit a certain limit and then you kind of go to, you know, what I would say, you know, like the captive, um, ENT engineering platforms like Cloud Code, which again, great platform, but um, we kind of feel like it's, it's silly that you would be wanting to kind of anchor yourself to just a set group of models. Um, and you know, we're finding that people aren't having the right model for the job. Um, you know, Aquila, we've got 500 models.

And then the last thing is, you know, when you've been using VS Code for the last 10 years to develop and you know, your company's saying what and admit It And vs. Code's great. No, absolutely.

Sorry. And, and You know, I'm sorry. When your company says, Hey, you need to switch to this other platform for AI and you know, you know, of course you're gonna have to learn a new tool.

You've been using the same tool for the last year. And so where Kilo differentiates is, you know, pay as you go, no throttling ever, you're never gonna be working late at night, hit it, you know, pushing on a deadline and all of a sudden your context, you know, window compresses because you've, you know, used too many premium requests. You can use any model you want from the latest and greatest to kind of open weight models that are much more cost effective.

And then you can use any UI you want from VS code to JetBrains to CLI cloud agents App Builder. Our goal is to kind of meet the engineer where they are not put red lights in front of them. 'cause I, I think that's what the theme of 2025 was.

You know, companies kind of, you know, software tools waving, hey ai, but really just putting a lot of unnecessary friction. And so, I mean, that's our mission is just to reduce that friction through an all-in-one agent engineering platform. I love it.

You know, there's another aspect though, to Kilo and the Kilo mission and it, and it's very, um, very similar to what Sid did with with GitLab because when you look at, and I happen to know this 'cause I've interviewed Sid a lot of times, right. But when you look at Yeah. Sid's background, even before he had GitLab, he was all about open and open source, and that is really kind of woven into the DNA of kilo as well.

Exactly. Talk to us about that, Scott. Yeah, I mean, I think like, um, openness is just so core to who we are at Kilo.

And, and, and, and for me it comes down to I would say three, three main reasons. And I guess the first one isn't even unique to Kilo. Um, Sid and I are big believers that, you know, when you build in the open, you build better.

I mean, and, and frankly, you know, when you talk to kind of founders of, you know, open source, open core code available, um, products, um, they kind of talk about this flywheel of, you know, you know, we get great transparency, great relationship with the community. The community actually becomes our extended engineering team like Aquila. We're constantly getting contribution from individual engineers and large companies that one to kind of tweak and, and kind of get into Kilo.

And so I think, you know, first of all, it's a superpower for any open source company and, and, um, I've really seen the benefits at Kilo. Um, I think the second is maybe a little bit more philosophical is AI is is like this amazing transformational technology that, you know, I feel like will impact every single human on this planet. Uh, I'm a big believer that that shouldn't be locked behind like walled gardens or gates.

And so, like philosophically, Sid I Kilo we're big believers that, you know, we should be making AI accessible to people. And so that's a big mission. Um, and then third is, you know, if, if you look at the pace of development of models, um, you know, like I was saying before, it's like I don't want to commit to one or two or three models or, or labs like, you know, you know, Gemini, GPT, um, Opus Minimax, uh, Z DOIs, GLM, like all these great models have come out in that last three months.

The pace of innovation is amazing. And so I kinda, our world is, you know, we view that we're work, we're not kind of working towards a, a world of consolidation of fewer models. We're actually working towards a world where you're having lots and lots of models, specialist models, big models, small models, open weight models, um, and we're big believers that we need to be open to all those models.

Um, so that kind of, the engineer can have the, the, the kind of freedom to choose the, the best model for the job. I think a lot of sense. And you know, Scott, I I think that rides a couple of different waves.

Number one, no one, it's the anti-lock in model. Exactly. No one wants wants to be locked in.

Right. Number two, we are making tremendous strides like every day it seems. Yeah.

But when you look at like, you know, we call 'em frontier models when you look at today's frontier models. Mm-hmm. I think what a lot of us are learning is they're amazing.

I mean, think about if I took someone from 1972 and transported them to today, Marty McFly or someone Right? Yeah. And said, Hey, just like magic.

Right? It would, they would think Yeah, it was God, they, they would've no idea that there's not a real human in there they're talking to Yeah. Who's super smart or something.

That being said though, I think we're also realizing that these LLMs, these frontier models that are trained on the whole of the publicly available internet aren't necessarily the right tool for every job. And that sometimes we're better off using a tool that's trained on a subset of things or maybe some information that's not publicly available that sits behind a firewall. It's proprietary, but it's perfect for what I'm looking to do right here.

And so I think the future is, you'd say there's 500 models, kilo supports right now. Yeah. There's gonna be an infinite number of models or, or vectors or whatever you want to call them that a tool has to be able to plug into.

And if we have open standards that allow you to plug and unplug right into different models like that, that's, that's what the developer wants, not only the developer. I think that's what everyone wants. That's What every, everyone, every knowledge worker you, I mean, you, you want to just essentially go down the, the supermarket aisle and grab the right model.

And I think, I think what you're saying on a hundred percent aligned with our philosophy is, you know, we saw in 2025, people are trying to keep you in like a, I know a bar refrigerator style selection of like three models when really there's 500 now there might be 5,000 in a few years and everybody May have their own model Yeah. In a few years. And your tool should work for you.

You shouldn't have to re-platform move to a different development tool every single time you want to use a no new model. It's like asking the finance team to, you know, every year switch from Excel to a new product. I mean, you know, our philosophy is like, and we have a saying is, you know, models change.

Um, great workflows don't. And so that's what we're really focusing on is building a great workflow so that, you know, an engineer can go from conception to architecture to coding, to debugging, to deployment, to kind of monitoring all in one end-to-end platform using the best AI model for the job. Yeah.

I think that's the beauty of the agentic nature of this, right? Which is, look, if I've got a, an intelligent autonomous agent here, it's going to know enough to switch models as I need to. Right.

But, but you wanna make sure it's switching in your interest. And I think that's a philosophy, you know, we're really, you know, doubling down on is 'cause like what we've seen in the past is some tools when they're optimizing, they're optimizing so they lose less money kind of subsidizing your usage. And that's what we kind of see in the curses of this world.

But like, you should be able to, you should have an application that intelligently optimizes, but you choose kind of the trade offs that it's optimizing for. You know what I mean? I think you know best whether you are kind of wanting to, you know, this is an important complex task and I want to the top of the top or you know, hey, this is kind of pretty basic and, and I'm all game for, for for saving costs.

Yeah. And, and I think cost of model and token charges and all that is one aspect, but I think we're rapidly gonna transform to do I want the one size fits all model or do I want the specialist model? Yep.

And, and or I mean, I think you might have a one size fits all model that, you know, hits, you know, 60% of your use cases and you augment it by a lot of specialist models. I I, I don't think it's, and I think that's what you and I are probably saying and agree on, it's like the main theme is you shouldn't have to choose, you know what I mean? Right.

You, you're right. You have all your options in front of Exactly. Okay.

I think people get what kilo's about, Scott, let's transition here, pivot a little bit. Sure. How do people interact with Kilo?

Cool. Uh, yeah. ai, um, you know, register, give us a download.

Um, first of all, we got great resources on Kilo. Uh, we've got, um, a leaderboard that you can track trends. ai, which is, uh, resources for how you learn kind of a agent engineering.

But yeah, get into Kilo, download it, install in your VS code extension or even use App Builder, which is our kind of low-code, no-code kind of vibe coding type interface. Um, and I think the coolest thing about App Builder, and we launched it very recently, so I'm, I'm very excited about it, um, is you've got that low-code, no-code interface, but it's built on that same kilo backend engine. And so when you say, Hey, I want to, you know, create this app, um, even me as a non-technical user, uh, what is producing is not some throwaway garbage that I can say like, Hey Alan, look at this.

And Alan, you're like, that's great, let's put into production. Oh no, we gotta start from scratch. No, it's built at great production quality standards.

And so I can essentially share that code base with my colleague who's kind of a full stack engineer who can open it up in VS code and just keep going without having to, to kind of start from scratch. And I, I think like that's the whole vision is, you know, we've got a lot of bone ramps from kind of those professional engineers who can go to vs. Code jet brains, um, or CLI or kind of the, you know, engineers who wanna do it, or the less technical folks who, who want to kind of get into vibe code mode.

And at the end of the day, you'll always be producing something. Great. Fantastic.

ai. I want to thank you for coming on Text Strong tv. What a great conversation.

As I said, say hello to Sid. I know there's a few other people I know at Kilo too. Probably Brenda s hello as well.

Yes, that's who I was. I didn't wanna mention tell him I said hi. Perfect.

We'll do it. Um, but keep us posted here. Will do on Kilo Progress and I hope to see you soon.

Will Do. Thanks Alan. See ya.

All right. Hey Scott. ai here on Text, on tv.

Stay tuned. We got more.