The Exchanges

Q Are there any opportunities that you see within the gap? Like, what would you need to see breakthrough in order to reach the next level?

A I think two things. The opportunities we see are the man-machine symbiosis. I think in many ways, LLMs are good enough to automate 80% of the work, and that's a lot, actually. And so rather than trying to do moonshot systems where you have an agent that does everything, You know, I think you could have application software that looks a little more like Devon type systems where you have the human supervising the agents and correcting them where they're wrong, helping them direct into the right direction if they're on like the wrong path of a, of a decision tree, uh, helping correct errors directly. Like that's what software could look like. Now we're, we're so like, um, Bias towards this AI does everything. Uh, it's a totally autonomous system, but the opportunity could just be man machine symbiosis. The AI does some, and the human does a lot too still. And so I think that's, that's likely what, what, where we're going to go, at least in the near term. And what could change is you could just have changes in fundamental research. Um, or one, OpenAI, for people who don't know, GPT-FOR-O-ONE, they launched a new model, does seem different, and I don't know if everybody knows the techniques they're using, and it could be a step function increase, like a research level change, that it's not just like some base transformer model that they're throwing more compute at.

AI assessment note: “I think two things. The opportunities we see are the man-machine symbiosis.”

Q Shifting to the defense wave. You've mentioned that this is one of the most exciting and profitable areas of venture investment and now competing with primes on historic contracts. So I'd love for you to break this down. What do you, why is this so exciting? Why is this most profitable and maybe overlooked?

A Yeah. We tend to think that the defense wave will be very important. We're seeing, as we started the conversation, um, we said that there's a lot of change happening now, particularly due to AI, right? And, and driving a lot of change in industries and, and questions of where value capture is going to be. This is happening in defense too. And if you look at, we track obviously the war in Ukraine quite closely. If you look at the war in Ukraine, warfare has completely changed. I mean, It's not even World War II type blitzkrieg tactics. You're back to, like, small group infantry assaults, um, extremely high battlefield, uh, awareness with drones everywhere, highly, ability to have highly precision strikes, literal drones that can fly and track cars or people and explode into them. That's just completely changing the way Warfare works, and that usually means that there's going to be a change in the way defense procurement works. We find generally, uh, the defense primes are quite good at some things. It's not that everything is bad. Defense primes are good at aeronautics. They're good at explosives. You know, I tend to, I always say, like, they're good at boom. Uh, don't compete on boom. Um, what they're bad at generally is software, is AI, is autonomy. And that's usually where we see a lot of the gaps. So I think Those technological changes and the changes in warfare are driving …

AI assessment note: “Those technological changes and the changes in warfare are driving new openings”

Q you reflect on this moment in time versus even our conversation in Q one, there has been a lot of change and a lot of exciting moments, a lot of surprising moments, uh, the Devon, Devon launched. Ilya left open AI. There's very large aqua hires. There's shutdowns. What has most surprised you over the course of H one, 20, 24? And maybe even was there anything you got wrong?

A My views on AI keep changing. So, um, because everything's so turbulent, you know, I initially thought that Application layer companies wouldn't do well. I think incumbents actually, well, I initially thought incumbents would capture all the value. And then when I saw Devon, actually, my mind changed where I thought there is room. Like it, if you take the thought experiment of saying, now we have this super intelligent layer that we can enter into any software system that exists in the world. If that existed 20 years ago, would software look different than it does today? Like the answer is probably yes. So It's very possible then that we will create new application layer software companies today that could disrupt the incumbents. Uh, and I think Devon really opened my eyes to that where I said, okay, this could be actually just what software looks like going forward. This sort of like supervisor worker type paradigm where it's not workflow software, it's supervisor software where you supervise the agents. So that, that was probably the biggest, you know, mindset shift that I've had. I still don't think it's like going to be completely Uh, monolithic where all application software would die, because I spent the 1:10 minutes of this call saying I was bullish on, uh, application software. I think most cases incumbents will stay, but there will be areas like software development, m…

AI assessment note: “I initially thought that Application layer companies wouldn't do well. I think incumbents actually”

Q For those that might not have context, what is magic.dev?

A I think that was it, and they call it, uh, we're trying to create, I think a fully autonomous model to create a fully autonomous software engineer, but it just raised a hundred million dollars. And I'll tell you a lot of focus on very specific code generation models. Um, and those to just. Summarize, make it a lot easier for developers to create software per co-pilot operates Much more like an auto-complete. So if you're writing an article on email or in Word, you automatically notice a sentence. It's pretty good. It does save you time, but it's not fundamental. In software now, we want to be able to create greater levels of abstraction from like function level to module to even program level of automatic writing. And so they're all elements of that path.

AI assessment note: “we're trying to create, I think a fully autonomous model to create a fully autonomous software engineer”

Q cheap. What are the public market multiples like? Are they still very, uh, volatile? We just saw an incredibly large, uh, funding round with OpenAI just so off the back of the heels of XAI, which raised six billion. OpenAI just raised 6.6 billion. At a hundred and fifty seven billion dollar valuation. Are these startups anymore? What is this class of AI and where are we with VC funding?

A Yeah, I think in the market, we're still seeing a tale of haves and have nots, I'd say. If anything that, well, we could first start with the public markets. Uh, software companies are cheap. I publish our internal software benchmark and you see forward revenue on a software company six and a half times. That's about where it was for the last I don't know before five years ago, the last decade. So, you know, five years back, it's the 10 year average. So we're back to a normal market there, which I think is curious because my particular view is that incumbent software companies will capture a lot of the value of AI. You know, if we're understanding that my views change, maybe every quarter as people see in my updates, you know, my current thinking is you probably won't capture a lot of the value on the Base model layer. That doesn't mean no value. Um, obviously there's value to providing a commodity service and most commodity businesses in the world are big businesses. Uh, you know, there are people who make memory and who make hard drives and they're quite big businesses. Uh, there's some returns to scale. So all those caveats aside, I tend to think the base model layer will be quite competitive and probably not as profitable as people think. You see the dynamics today where You know, there's huge convergence, I'd say, in the performance of these models. Oh, one just came out. …

AI assessment note: “software companies are cheap. I publish our internal software benchmark and you see forward revenue”

Q Um, okay, so to close out, I just would love to know, what are you looking forward to most this, this year, and what advice do you have for founders in the C to Series B stages going out to raise this market?

A I'd say the thing I'm most excited about now is, you know, there's been a lot of talk of is SF dead? Is it back? Um, it has, it has, you know, the startup environment been completely democratized across, you know, every place and you can just be remote. It doesn't matter where you are. And I think the most exciting thing I'm seeing is that I've been in Silicon Valley for maybe 15 years. And it's really, I think, in a special place now in the Bay Area, where it feels much more like 20 13, 20 14, 2015. Where, you know, when I first got here, I got here in 2008. That was like a current time. But over the last five years past that, it became really exciting. It was a lot of young people who were really excited about building, who were just here because they love technology. It wasn't about the money. It wasn't necessarily about getting rich or anything like that. It was about building really cool things. And that's, I mean, that's a lot of who I found or succeeded. It was not a group of people who wanted to make a lot of money. It was a lot of people who were on the same mission and wanted to see what we can accomplish on that mission. And I'm starting to see that Energy back here yet where it is a lot of. Young people coming out of school, not just young, but a lot of young people doing it who are in awe of like. The new capabilities we're seeing and just trying to push the bounda…

AI assessment note: “I'd say the thing I'm most excited about now is”