Blitzy raises $200M at $1.4B valuation to autonomously refactor enterprise legacy codebases over weeks

begging Jensen for GPUs and it's hotly debated. Are people debate? They're not begging you.

We're working hard to serve them every day and uh and showing up with with as much humility as we can and working to meet this moment.

That's great. That's great. Well, thank you so much, Scott, for great to meet you. We appreciate it.

This is fantastic. Our next guest is already in the waiting room. We have Brian Elliott from Blitzy. He is the CEO with a huge fundraising announcement coming in to the TBPN Ultra Dome. Let's bring in Brian. How are you doing?

Here we go.

Hey guys, how are you?

Quick transition. Thank you so much for taking the time. Uh please, this is the first time on the show, introduce yourself and the company. Yeah. So, Blitzy, uh, we're announcing the $200 million financing at a $ 1.4 billion valuation today. So, we're Nice. Thanks, guys. Appreciate the love. Uh, so we're an autonomous software development platform specifically designed for complex enterprise use cases. So uh we serve banks, insurance companies, anyone with huge amounts of code and and we do uh we do autonomous work meaning the system will run for days to weeks autonomously recursively improving the code using all the foundational models together and you know OpenAI, Gemini and and Anthropics models.

Um yeah it's fascinating. So what is the go to market motion for you? I mean, you're focused on the biggest companies. Uh, like do you have like tell me about the shape of your sales force? Like what the process is like to actually deliver value because I imagine that a lot of times you're going up against like the build versus buy question almost like should we just build something like this internally? Should we just use the tools directly, the models directly? Uh, but you clearly have a fantastic value proposition because you've seen a lot of traction.

Yeah. So, we go in direct, right? It's very much like a palunteer like uh motion. And so we're going to go in direct. We're going to show you quickly. We're going to reverse engineer your codebase. Often times 30 50 million lines of code. And we're going to do all that within 48 hours of installing Blitzy or else we get

sounds like a threat.

That's no I mean obviously that's valuable. Uh I mean that's the case with all these coding agents is like the first thing is like understanding what actually is the business problem. What's actually going on? How how are you uh confronting like the diffusion question broadly? Like there are so many systems that feel like complete code. Oh, it's the beautiful you know endtoend engineered system and then you realize that oh wait actually like if this person doesn't fill out this form at this time like the whole system grinds to a halt because that's just the way this organization was designed years ago.

Yeah. So these are the like complex use cases that Blitzy is designed for. We we designed this for the messy brownfield legacy codebase where the system is going to come and it's going to create a knowledge graph and we're building and running the application, right? And so inside of a inside of a VM, we are building the application looking at what's going on and then we've invented a proprietary way to store and then have agents be able to traverse a graph, understand what's going on, and then build large amounts of work on top of it.

Uh uh how how did how did you talk about the category overall during your fund raise? It is, you know, every single day there's there's a new company in this space and everything we've seen so far is that there's just s such overwhelming demand that almost every company is is doing well and almost every company would have been seen as bestin-class if it was in another category, you know, based on their growth a few years ago. But how how are you talking about the category? How do you see it evolving when everyone kind of wants to do everything at least in the fullness of time? Um, so I'm curious how you pitched it.

Yeah, so we really focus on on autonomy, right? So if you look at what's out there today, there are things that run for a half hour, things that run for a day continuously without a human in the loop. Our system is is at its core defined to run for long periods of time. We're the most inference comput intensive platform out there driving up code quality. like we'll have our system it can run for weeks on end improving the quality of code specifically focused on those large large brownfield code bases. So that's where we found our our sweet spot inside of the enterprise. No one else is as laser focused from a technology perspective as we are in large brownfield large scale autonomy and as the models get better we continue to rise up and take advantage of that. So we are we are long transformer want as much capex uh investment into the category as possible.

Did you

from other people? Yeah,

we want everyone to keep throwing models at and and it's a like clear clearly um

clearly a compelling pitch because you raised the 200 million, but it is it is I I mean I think it's notable that I would say that would would you say that is quite a similar pitch to many of the other companies in the space or do you are they are they saying they're focused on autonomy but they're really not and it and it's it's more um point solutions or models.

The the world the world has looked at being able to have a system run continuously for a day as autonomy, right? And what we've done is we've redefined that to weeks at a time, right? Doing hundreds hundreds of thousands or millions of lines of code that have been endtoend tested ahead of it getting back to the human. So, we've set a new benchmark for what's possible against these large scale code bases. Uh like we're basically setting the frontier for for autonomous software development.

Okay. I I I can't find it, but I saw Meta put out a new benchmark today and the benchmark was basically take the frontier model and try and recreate an entire repo, an entire piece of software from scratch. And uh all the all the frontier LLMs were were were performing very poorly like they were 0% 3% maybe it was very rough. Uh you see a similar dynamic with Arc AGI V3 the Frontier models the most amazing models are at like 0.2% 2% 0.5%. And and then at the same time, you have people with like, you know, AI agent psychosis and vibe coding tons of things and like incredible results from the models and like incredible revenue ramps from all the labs. Uh and I'm wondering how you square these like like what what are the models still bad at? How are you processing the question of like spiky intelligence? Where are you getting the most value? And where are you seeing okay, it's not quite there yet? So you see a rapid depreciation of intelligence after you exceed like 100k context window, right? And so they're advertising a million, 10 million, there's someone subcited out 12 million, you know, context window.

Yeah, I wanted that's not real ask you about that, but but continue.

Yeah, that like they're using sparse attention beyond that point, right? And so you can kind of get it right. Sometimes the academics will call this context pressure. And so what's important is that you you limit the LLM's ability to look within its effective context window just in time every time to do the right amount of work. So as the code bases scale like you see results depreciate uh pretty dramatically, right? And so you have to be able to have a system level approach, not just throw a model at it with a slightly larger advertised context window to solve these large scale problems. And so that means like orchestrators and like teams of agents like there's one agent on this 100k token chunk of code and it's very good at that and then it's coordinating with another agent. Is that the solution?

Yeah, but code is relational not serial, right? And so like like you you can't just say the serial 100k token which is like uh 10,000 lines of code but you have to cut it in half to account for prompts and tools, right? Now we're talking about a minuscule view but the entire relational aspect of well it's something way over here in this service that's actually relationally relevant to