At this moment in time, there is a gap in the technology ecosystem. With access to large-scale general AI models, you can be up and running with something that users want - that changes people's daily lives - in weeks. These models are only getting cheaper and more powerful. The right comparison in terms of opportunity size isn't "existing software markets," it's the human work being done. Founders should be ambitious to change entire industries and invent new experiences. The capabilities are there.
This is not a comfortable environment for big companies. The pace of change in AI will blow your hair back. You can't read the research papers quickly enough. Big companies are slow to make bets. They need to worry about the financial quarter and do "change management." They struggle to reason about non-determinism. They have too few hackers and too many bureaucrats. Those with strong leadership are in the biggest race of their economic lives to adapt to AI. The cosmic scale has suddenly tipped in favor of speed and unruliness, the only advantages startups have ever really had.
There has never been a better time to start a startup. But we, very consciously, have a small, early-stage focused fund, because we want to be beloved partners to founders, and we don't want to be in the business of "deploying capital." The asset-gathering train is hard to get off of -- many venture funds raise ever-larger funds in order to be paid ever-larger management fees, and then they need to invest that capital in order to...raise more capital. We want to bet on great founders when they are really early, actively help them build really important companies, take every phone call and be as aligned as possible. I take home no cash compensation at Conviction (but do make carry when companies succeed).
So what can we do when we believe the opportunity is much bigger than we can invest in, with our traditional fund model? Embed is an experiment to see which of our advantages we can share with a broader founder community.
What do startups need? They need access to talent, credibility and customers, advice, money. Some of these things can be distributed in programmatic ways, valuable to a lot of people with just a little lift from us.
We have a deep and broad network, and want to help cofounders and early hires find one another. We get a huge amount of inbound, and already do matching manually, conversation by conversation. We're creating a Schelling Point, a way for people who want to be on the frontier of AI to find one another easily. It's as if you put a sign in New York City that said, "Oregon Trail Caravans, This Way." Embed also creates a channel for founders to work with our recruiters, sourcing the best researchers, engineers, designers, product people, business folks who explicitly want to work in early-stage AI.
We regularly talk to customers, have credibility with them on AI, and can make introductions. We could do it manually. But here, scale actually helps: more startups, more likelihood of matches. Our friends, early adopters who want to find solutions, want efficient ways to broadcast their problems to high-quality founders.
At the very beginning, startups are dead simple but insanely hard. Recruit, talk to users, build the product, fiercely repeat the latter two steps until it works. No one can do that for you. But advice can come in the form of a framework for attack, historical and market context, good but unintuitive defaults, a foil or check or expansion of your thinking. We can introduce you to what "truly great" looks like, with intimate, small group dinners with legendary founders.
Money is perhaps the very simplest thing to hack. Let's split it up into people and compute. We'll give people enough to quit their jobs and get started, with no-brainer terms. GPs from every top venture firm and super-angels are coming to our demo day, and they trust us to filter for interesting opportunities. We can teach founders to tell their stories and fundraise.
In one very specific and weird way, AI has made it expensive again to start a startup. Cloud CPUs/storage made it so a set of dropouts could build a vanilla SaaS application in a weekend. But as more companies suddenly need supercomputers, we've now traveled backwards in time to some previous era of hardware complexity and poor accessibility. We have dedicated H100s that we parcel out at cost to founders and friends, and we have partners for generous API and GPU credits. We can share those resources with a broader set of founders, so that experimentation budgets and access are not barriers.
This feels like a set of infrastructure and advantages that is too valuable to not share. But one thing we think people will really care about is a little more subtle. Good founders like being around other good founders. As a now-legendary founder friend described to me, "Starting a company sucks. Being in your parent's basement, not having a real job, feeling embarrassed like everything is always going too slow, that sucks. The pressure of having to show other people I'm making progress is the only thing that got me through." We want to create that pressure -- weekly dinners are a mechanism for accountability (and energy).
Uniquely, the wild west of AI products and companies means there's much to be learned from a set of smart and creative peers. Founders are wrestling with shared questions: how to charge for work replacement, how to think about inference cost, how to deal with CUDA memory errors, how to introduce long-term memory. Our office slack has a #Llama2 fine-tuning channel.
The other reason good founders like being around good founders is that founders believe. They're open-minded. They understand the technology. They have their own weird versions of the future and they want to try and see yours. Great, world-changing technology ideas are not birthed in conservative settings. It helps if everyone is a nerd, everyone is exploring how we can talk to computers, everyone thinks about the upload, everyone moves quickly and has ideas. Things you might be reluctant to say in more polite company, you can say amongst the converted.
What's in this for us?
We do not expect to make a lot of money directly off the accelerator. An uncapped note is not a good investing instrument. While we do expect to write larger checks to some companies coming out of Embed, we expect more of the companies will be backed by other investors.
If Conviction is about, well, conviction, then Embed is about reach and curiosity. We want to see and back a broad range of interesting projects. We're excited to take more people on our caravan to the frontier. We think our advisors and dinner party guests are the best in the world if you want to build a technology company, so we think the caravan will be, itself, fun. We want to make our network, already great, more tightly knit and more powerful, and that will benefit the founders we work with.
Frontiers like this are exciting, weird and lonely places. Hitch your wagon, and let's go explore it.