00:00Bain Capital Ventures has raised a $1.6 billion fund to back early stage AI companies for building
00:06on a playbook it established with its previous Fund 10. VCB partner Slater Stitch is with us
00:11and this fund is for life after AGI. Explain. That's right. Well, first of all, Ed, great to
00:19see you. So good to be here. Thanks for having me on the show. You're right. So one of the
00:24big
00:24ideas that we have for this fund is that we have stored up this incredible ability with
00:31artificial intelligence and we've seen some applications for it. In some cases, there
00:35may be the more obvious applications. But as the technology gets even better, we should
00:40expect an explosion of new applications. And we wanted to approach this fund with the philosophy
00:46of really thinking about that hard from first principles. What does it mean like when we
00:51go through a technological revolution in intelligence just as profound as electrification or mechanization
00:57earlier? What would it mean to sort of have a post-AGI economy? There are three different
01:03sort of, well, four probably flavors of this. Infrastructure, physical AI, which can mean
01:07robotics, security, which, you know, given the events of the last seven days is massively
01:13important, and then also services. Of those four, to your mind, which is most present in the
01:20post-AGI life that you're envisaging? I think that all of those will have important roles
01:26to play. And it's just a question of sequencing, you know, sort of in what order will we see
01:30these different things? Do you see an order? I think that, you know, things that seem sort
01:35of like knowledge work on a computer flavored are going to be the form factor where it's easiest
01:39to make progress fastest. We sort of know how we would ascend up the scaling curves there.
01:44The models, the base models get more intelligent each generation. The harnesses get better. The
01:50sort of task horizon for agents gets better. The tool usage. It's just easier to do things in a
01:55purely digital domain. So that's where we see it happening first. But of course, we're incredibly
02:00optimistic about applications and robotics and other physical world applications over time.
02:06You guys talk about life after AGI. The difficulty is AGI is not universally defined. Some would say
02:17we're at AGI. You know, I'm thinking more recently about OpenAI's release of Astra. And Greg Brockman,
02:26as an example, saying, you know, this is the first step to true AGI. How are you defining it and
02:32measuring it? Well, you know, one thing I think about is if I just gave my past self information,
02:38what I have called this AGI, if I went back to how I thought in 2016 or something like that.
02:43And by
02:43that measure, absolutely. I mean, I think if you had told me that we would be solving millennium
02:47problems or being able to run these incredibly useful long running agents, that would have surprised
02:53me at the time. So I would say that from a economic and also even from a scientific perspective,
03:00a lot of the promise we're already seeing there. You know, I do think AGI is one of these terms
03:05that
03:06can never be fully defined. No matter how good the systems are, there will always be sort of some
03:10next thing. But I've, you know, I personally am most interested in the ways that will affect the
03:15economy. And I feel like the technology is already there to do so much. This is an early stage fund,
03:21right? So it's $1.6 billion to invest at that early stage for companies where you feel they have an
03:28opportunity in that future economy. So if we take physical AI, robotics, for example,
03:33where would you apply that to the economy today? Where are the gaps in this US economy where you
03:37think, okay, this is ripe for robotics to be the solution? Yeah, I think that the nature of,
03:46you know, again, I think lots of things will happen in the long run. In the short run,
03:50you know, application areas with predictable environments with sort of relatively set tasks
03:55are the ones that will, of course, be easiest. But I, you know, in some ways, in the long run,
04:00that might be the least important. So, you know, for example, we're investors in a company called
04:04Sunday Robotics that's working on home robots, sort of fully integrated to that use case. And I do think
04:12that when you take an application like that really seriously, you can make progress very quickly.
04:16Um, you know, I do think that there's also, um, there's also this question of, you know, um, the
04:25industrial applications, um, will probably have some of the biggest impacts. We have, you know,
04:30sort of good industrial robots for relatively predictable manufacturing tasks for manufacturing
04:36tasks that are less predictable. That's where intelligence can actually play a role.
04:40Can we talk a little bit about investing? So this is a specific fund. It is fund 11. Technically,
04:46it comes off to fund 10. What's realistic in a calendar year? Do you go after high volume or do
04:53you just say, okay, like this is high conviction in this environment. We have a thesis for this fund.
04:57So we'll do a handful of investments and kind of cap it. How does that work?
05:01We don't think about it in, in that way. Like the way that we think about it is that we
05:05know what it
05:06looks like when we see a incredible founder pursuing an outlier opportunity that we believe in, you know,
05:12we might get the conviction on that opportunity in different ways. In some cases,
05:15it could be thesis driven, like you're describing, you know, I would say that our investment in
05:20fleet, for example, the RL environments company, um, happened that way. You know, in other cases,
05:25um, it's really through conversations with the founder and learning the opportunity that we
05:29get the conviction. But I don't think about it in terms of, you know, capital deployment over some
05:35period of time or some sort of strict rules about how we would deploy it. It's yeah.
05:39Take us inside Bain Capital Ventures, how it's working as a firm at the moment. So if you have
05:43an opportunity or an idea, you're the sponsor of it, how did the partners kind of get around it and
05:49decide like, this is the bet we want to make, particularly when you're saying, okay, we want
05:52to invest in an opportunity for life after AGI? Yeah, I think, um, it's, um, so one way that I
06:02think
06:02about this is, um, opportunities that we think could be tremendously valuable and that we understand the
06:08technological path, um, to achieving them. So for example, um, you know, in companies like
06:16Decagon or, um, cognition, you know, we understand the promise of what those companies are in
06:22automating software engineering or, um, customer support. And it's sort of, um, clear that the plan,
06:28if the, if the sort of founders are able to do the things that they want to do, that we
06:31should be
06:31able to achieve that. There's other cases, you know, company like periodic labs, for example,
06:35just building an AI scientist initially focused on material science applications, where we spend,
06:41you know, sort of more time on the technology and sort of why, um, we think the founders are going
06:47to be uniquely able to achieve that. I'd like to end the conversation talking a little bit about
06:51cognition. You know, they've just closed around, but the thing that's changed for me in the last
06:56calendar year or so is the, the ads in San Francisco there. I'm paraphrasing, but it says Devin is
07:03actually good now, which is sort of an admission that it's taken time, but that's amazing. I mean,
07:08you see billboards over SF all the time, but just in cognition's case, what you saw in the
07:13development arc. Totally. So, um, I really love that campaign. And, um, what I like about it is to
07:18me, it's cognition calling their shot really early on. So, you know, at the time that they sort of put
07:24a flag out and said, we're going to be the autonomous software agent, um, that works. Um, you know,
07:29I would say that the base models were not quite there yet. Um, at that time, but that was an
07:33easy
07:33thing to do, but they, they understood exactly where things were going and sort of understood
07:38that that was the valuable place to be in the long run. And so I think it was like a
07:42courageous
07:43thing to do. And I think they called it perfectly. Is there a next cognition then with this fund that
07:49you think you can find? We do. I think that there'll be, you know, other applications that look
07:54like this. And then I also think that as you just sort of go up levels of intelligence automation,
07:59there will be all sorts of things that we can't even anticipate yet.
Comments