Master Plan - Justin Glibert (Foundation)
The co-founder of Foundation discusses self-replicating machines, formal verification of software, and how to build a hard tech company.
Welcome to the 3rd issue of Master Plan, a series of conversations with deep tech founders doing really hard things.
Justin Glibert is the co-founder of Foundation, an advanced manufacturing company focused on self-replication. Foundation’s initial vertical was space manufacturing, and you’ll see some of that reflected in our conversation. Justin takes us through his thoughts on bringing Moore’s law to the physical world through self-replication, startup advice for working on hard companies, why it’s important to spend lots of time learning and internalizing the fundamental principles of a field.
Justin introduces himself
I describe myself as someone who likes doing things that do not have a playbook. If a task or activity doesn’t have a playbook, that means there aren’t any readily available resources online, like books, that tell you how to do it.
Things that have a playbook
Let's give some examples of things that have a playbook. Basically, you can do some sort of pattern matching under action.
Today, becoming an AI researcher has a playbook, even if it's a difficult job. If I meet someone who tells me, “I went from doing a Bachelors in Art to working at DeepMind,” I’d be deeply impressed. But deep down inside, I know there’s a playbook for that. I know it’s perfectly doable. There’s a clear path from artist to DeepMind.
Side note: Justin wrote his own playbook for teaching yourself enough AI to get to the research level.
Starting a tech startup has a playbook because even if it's extremely difficult, plenty of people have done it before. There is so much information out there that explains to you how to do it.
If today you wanted to get in the top 1% of the strongest people in the world, you know how to do it. It could take 10 years, but there’s a clear path from where you are now to the top 1% of the strongest people in the world. There is a clear training program with specific workouts.
Things that don't have a playbook
Things that don't have playbooks are activities where you can't pattern match what you should do compared to what people have already done because you can’t find anyone who’s done the thing before.
Something that has a playbook doesn't mean easy, but I'm more interested in stuff where the path has not been laid down. It can be anything from starting very hard companies to writing genre-defying fiction.
There’s this part of science that’s kind of dead, called molecular assembly. It’s the idea of being able to put molecules and ions wherever you want in space. Because if you can arrange molecules however you want, you’re able to do tons of things, from copying the best Wagyu beef in the world to building laptops with a billion cores in their processor.
It’s very trivial in its description—we want to put molecules wherever we want. The science was really hyped up in the 1980s. Tons of people were working on it. But if you unroll history from then to now, not much has happened recently. People stopped working on this for various reasons.
An example of doing something without a playbook would be if someone who took a look at molecular assembly, read all the papers, contacted the researchers that are still alive, and tried to figure out a way to bring back some of that usefulness into a product or research paper.
There’s no way to Google for a Medium article that tells you how to learn molecular manufacturing or start your own side project around it.
When I meet people, one kind of quick filter I have to see if we’ll get along is whether, in their life story, they’ve done things that are not just hard, but also without a playbook.
Foundation’s initial plan was to build in space
The plan we announced for Foundation at the Pioneer livestream was we were going to build in space.
And the reason we want to do that is because we think it's a great first market for a company whose final goal is to build self-replicating machines—a machine that builds all the machines. If you can build a robot that can build more robots, you bring Moore’s law to the physical world.
I’ve spoken to so many people at NASA, ESA, JPL, and co. It might be a bit too early for factories on the moon. However, today is definitely the right time to start a space company. Surprisingly, people don't understand how fast things are moving.
If you want to build a constellation company or do things in orbit, today's the best time to start such a company because SpaceX literally became the bus service of space. It's actually getting quite cheap to do stuff there. So it's a great time to start a space company, but we think Foundation might be a bit too early.
I can't really talk much about our current plan. Not because we have a culture of secrecy, but mostly because it's not mature yet. The main thing that stays the same is you want to be able to build faster. And the key to that is being able to build machines that can build other machines.
Rates of progress in the physical and information worlds
Take a look at the economics of the information word.
The interesting thing is that computational capabilities have been multiplied by 100,000 in the last 25 years. Moore's law is at work and we get 40% cheaper compute every year. And that's completely normal to us. You are expected to wake up on Christmas Eve every year and get a computer or a phone that is 50% faster than the one you had last year.
On the other side, in the physical world, you don't expect to take a flight every summer that goes 40% faster than the one you took last year. People take Moore’s law for granted. But in the physical world, we have nothing like that. If you work out the numbers, you realize that the physical world improves at approximately 4-5% a year, as opposed to 40% a year.
And one of the most important reasons why it’s not going faster is because we have still not figured out a way to build faster. We have factories today. If you have upfront capital, you could build things for really cheap, but the problem is you need $100M in upfront capital.
The reason why it’s so expensive is that we don't have machines that build machines. We have machines that build stuff (aka factories). But we don't have factories that build factories.
So you want to be able to bring Moore’s law—your 40% rate of improvement a year—to the physical world. And how do you do that?
Self-replication brings distribution costs to zero
If a 3D printer could print its cousin or a sibling, that would make things much faster because you have exponential laws and your distribution costs become zero.
Why do you think software is so good? It's because the distribution cost is zero. If you write something that is good, you only have the upfront capital of development. And then if you consider the servers are free, you can distribute that software to the world and everybody can enjoy it.
In terms of like physical products, that's not the case. You build something once—great for you, but there’s no way you can distribute it to a hundred million people.
That’s 80% of the rest of the work. The prototyping is actually the easy part.
What if this changes? What if for every single physical product out there, the only cost is the upfront development costs and distribution becomes free just like software? This would change so much stuff. Your life will literally become 1000x better in a span of weeks.
Anybody could come up with something that is better than what exists right now and can distribute it for free. You could have kids in their basement writing cars instead of writing apps—that would be so nice and full of memes—and you will be able to get one in your own garage.
How to make manufacturing 1000x better
For the information world, the reason we have no distribution costs is because a lot of unknown people built the internet. They built the software that is used everywhere in terms of infrastructure. But we don’t care about those people. We only care about the kids that built the apps that run on top of the infrastructure.
For the information world, we cheated a bit. A lot of investment has been put into developing the infrastructure itself. But I think if we can bring Moore’s law to the physical world, we’re going to see the same set of extremely cool companies and life-changing products we have seen in the information world.
And the key to doing that is making manufacturing so much better that we won't even call it manufacturing anymore. We will need a new word. That's what we're working on. We think that maybe the whole space thing might not be the best way to do it, but we're definitely still working on making manufacturing 1000x better.
The main path to this is building self-replicating machines. Self-replication is the equivalent of AGI for the physical world. AGI gives you complete control over information, and self-replicating machines will give you complete control over the physical world.
If you look at humanity's tech tree, you have two sub-branches— information and physical. And the last node in each of those sub-trees is AGI and self-replicating machines (or molecular manufacturing, since molecular manufacturing leads to self-replication by definition).
Self-replication in biology
We already have self-replication. Biology does it. We don't really control it.
You can do this from the bottom-up. That is, you start with cells that are already self-replicating. You think it's really hard to control them. So you spend billions of dollars and a lot of research time to figure out how to control them.
Let's go biotech, right? Gingko Bioworks and similar companies get self-replication for free, but then they have to fight with a poorly written O.S. and really obfuscated code. That's what they do.
So you basically have those two different approaches. Either you start from biology and you learn to control it, or you start with something you already know how to control but learn how to make it self replicating.
How to ask for advice
Something that is very important is you need to find the right people. You can’t reach out to a random JPL guy and be like, “Hey, can you become part of our technical board of advisors?”
The issue is if the guy actually doesn't care about the problem in the first place, he's not going to be involved. Just like pitching is much more about pitching to the right people than pitching the right way, asking for advice is much more about asking for advice from the right people than asking for advice correctly.
I think that if you reach out to people that are already extremely interested in what you are working on and you’re also bringing some sort of fresh approach and naivete to the problem as well, this is refreshing for people.
If you just reach out to people who have spent their entire life on the problem and you're like, “Hey, I did my homework. I know the prerequisites. We'll be able to have a good conversation together. Can we talk?” There is no reason to say no, and this is the way we've approached it.
You need to not make a fool of yourself in the first five minutes. After that, you can sound stupid. It's just that early on, you want to show that you’ve done your homework because there is nothing worse than someone reaching out to you who doesn't have any context and just thinks that they will get some free advice from you. It's the equivalent of someone reaching out on LinkedIn for a job. Even if they know you don't want to be hired right down, those bots just send a billion LinkedIn emails.
You don't want to be that person. When you ask for advice, you need to be extremely well-prepared. And if you’re also asking the right people, it's really easy to get advice.
Just like everything, there is always some networking involved. It's very sad, but even in today's world where you can literally talk with anyone anywhere, you still need to know some people and you still need to build some relationships, but it's definitely easier today than it used to be (for younger people at least).
How Justin met his co-founder
My co-founder is one of my best friends from university.
I think it was in my last year of high school (or the one just before) I was on a video games forum in French because I'm from the French-speaking part of Belgium. And there's this guy who says, “Hey dudes, I'm studying in the South of UK and it’s actually fucking lit. You guys should not go study in France or Belgium because it's super lame, come study in the UK.”
And this struck me as a very great idea and it's something I hadn't considered. I thought that I had to study in Belgium like it was a law of nature—I thought this was something that I was forced to do.
And I was like, okay, this is interesting. And then somehow a year and a half later, I ended up getting acceptance letters to all the universities I applied to in the UK. And I decided to go to the University of Edinburgh because it's super old, it looks like Hogwarts, and it's fucking amazing. So I went there and in the first year of university, I met a lot of really, really smart people.
And one of them was my co-founder, Leopold. We lived together in our second year, and I traveled with him to a bunch of different places.
Why co-founders need a strong existing relationship
I’m a bit biased right now, but I don't think you should start a company with people you don't know very well and with whom you have not spent a significant amount of time as friends.
When people do business, they tend to be different and you need to be able to go through all those awkward moments and all those moments where you should have been fighting and stuff, because then when you actually do the hard stuff, things are all right.
So Leopold asked me “you know so many smart people, why me?”
And the thing I told him is that I know that if shit happens, it will be fine. We won't actually fight because we're both so stoic. And we know each other so well that we have arguments anymore, it's kind of funny. As opposed to other people who are really good friends on the superficial level, but as soon as like shit hits the fan, you don't want to trust each other anymore.
So I think you really have to build up that trust and that stoicism. You need to be able to understand each other very well and go through that phase where arguments are still possible, because then when you actually do things that are very hard together, like starting a company, you won't have problems.
I tried starting small products with people that didn't know well, and even if they were brilliant, as soon as we had problems, things would become very awkward. And it was hard to communicate and be upfront about things and be like, “Hey, you suck, you did something wrong, you should change what you've done.”
It was hard for me to give that kind of feedback. With Leopold, I remember we were preparing for the Pioneer livestream and one of our objectives was that we wanted to have the best pitch deck out of every single pitch that has been given on every single Pioneer livestream.
We really wanted something really good. I was the one giving the pitch. Leopold was the one who made the pitch deck and he did all those fancy 3D renders. And we were together in Spain with a bunch of friends in a hacker house. And one day I wake up and I'm like, “dude, your 3D models are shit. This looks so bad.”
I would not have been able to say that to someone that I wasn't really good friends with. Think about it: you have someone who spends five days working almost 24/7 on one single render. You can't just come to them one day and tell them it’s bad and they have to start from scratch. He was a bit mad at first, but he accepted it pretty quickly. And afterward, we got something that was way better because he saw the truth through my criticism.
So this is the kind of relationship you need to have if you want to start really hard stuff together, whether it's an art project or a company.
I have a lot of uncertainty about what I'm doing.
I can call my software friends over the weekend and say “okay guys, here’s a quick AI model for doing X. Let’s build a business out of it.”
And I know we can do it. I don’t know whether we’ll succeed, but I’ll be able to do every single technical task around it. There are no fundamental roadblocks in terms of physics that stand in our way.
With Foundation, we were working on this space thing, and even today where we’re doing something radically different, there’s some stuff that—even if they don’t go against the fundamental laws of physics—is still fricking hard.
It's very hard as an outsider in a field to decide whether something is barely doable or completely impossible in a reasonable amount of time because this is where the frontier is.
You don't really care about things that are surely doable. You don't care about things that are so trivially simple that even a child could do it.
You care about the things that are just standing between doable and impossible.
For Foundation, we had to basically draw the line and be like, okay, can we do that? Can this be part of a plan or is this too hard? And there's no way to figure it out in a reasonable amount of time. Anytime you work on something that is on a different tier of science and is really low in the technology of its scale, you will have those problems.
The benefits of building hard startups
The best part about Foundation is that it's cool. I can wake up one day and be like, “Okay, today I'm going to read 40 papers about turning a specific mineral and lunar dust into pure titanium.”
And I'm going to find it super interesting because it's a different tier of science. I can speak with incredibly cool people. We have kids (and people of all ages) send us emails every week about wanting to work with us and do cool stuff with us.
When you work on something that is hard, it becomes much easier to recruit. We have seen it as well. It becomes much easier to raise money. It becomes much easier to just get people to talk with you at cocktail parties when you tell them what you're doing.
And this is very important. It might sound trivial, but the attractiveness of an idea is something that will open a lot of doors. Space itself is very cinematic, so people can watch rocket launches and they feel excited and they feel the thrill when the Falcon 9 thruster turns off because the rocket landed on the ground.
When we started with the idea of doing Foundation in space, the theatricality of the plan was really helpful. We were able to make some renders in 2-3 days of a moon base, and sending that into people’s inboxes would almost always lead to a call. This is something people can’t say no to.
Talking to customers as a hard tech company
The first step is to define who your customers are. There are different steps, and I think people have been so spoiled with the fact that with software, you can get customers from day one that they think it’s still the right thing to do.
And it is the right thing to do—if you can do it. For most companies, yeah. But try to picture the early days of Tesla. You have some people in a garage that just bought a Lotus, and they tried to put a battery inside it and it doesn't even work yet.
Imagine them trying to go around and being like “Hey dude, you look like a customer for us, how much will you pay for this hacked car that runs on electric motors instead of fuel?”
This doesn’t really work. It’s not a problem the customers have right now, it's just really hard to go through the classical playbook of trying to speak with customers early like you do when you build a SaaS product. When you work on things that are very hard and a differentiator, it just doesn't work.
For Foundation, what we did is we tried to figure out whether people like SpaceX and NASA would eventually pay for this stuff.
So I guess you can say that it's talking to customers. I think you can say that whatever you do is always talking with customers, right? Because anybody could be your customer. But I would say that in our case, we were trying to figure out how big the market was. And you can't do that by just Googling stuff online.
Leopold was working at SpaceX and he asked SpaceX managers. “What's the policy at SpaceX?” SpaceX is famous for never buying stuff from outside contractors. They sometimes do, but they don't publicize it because they want their image to be of a company that does everything.
But we asked them, “Let's say you go to Mars and you need a fancy exoskeleton to protect your habitat. Would you buy it from other people?” And they were like, yeah, sure. It depends on how expensive it is. What SpaceX actually does is if it's too expensive to do it in-house, they will buy it from other people.
How to build for markets that don’t exist yet
I think that when you're building for a market that doesn't exist yet, the most important question is when will it be ready? When will the market actually open?
Imagine if tomorrow I can snap my fingers and be like, “Hey, I have the AWS of the moon and you can rent some excavators from us and get power and internet and whatever.” Nobody would pay for that yet. Maybe NASA would give us some money because they'll be there soon, but the market is nonexistent.
So when you build for a non-existent market, you really need to get your timeline and your Gantt charts right. You need to know when everything will be ready. It was fine for us if the market’s not open now, because it would take us at least five years to achieve the things we promised doing. We had to time it very well because otherwise you just run out of money.
If you build pickaxes when miners aren’t there yet, it’s dangerous. You have to time it with the gold rush. If you don’t time it well with the gold rush, you die.
On the lean startup
I think the lean startup is a plague.
I really do think it's a plague because everybody does the same thing. I mean, it's great because it reduces uncertainty. It's like you're playing a video game, right? The video game is hard, but you know what the next level is, and you know what the final bosses and you know what the middle bosses are.
When there is no playbook, there is way more uncertainty because you don't even know where you are. It's like when you're playing Dark Souls and you think you've killed the boss, but it's really only the first phase.
And with the lean startup, you can actually quantify where you are because you have your DAU, you have your amount of money raised, your ARR, and you have all those numbers that depend on precisely where you are in the game and how to progress. And so it's all about reducing uncertainty.
But when you do things that have uncertainty, it's really hard to actually know what the best path ahead is.
Formal verification - an underrated space
There are industries that people are paying attention to, but not enough. I would say one that is very dear to my heart is formal verification of software. I really think that programming as it is right now is completely fucked up. It makes no sense that you spend more time fixing your programs than actually writing them.
And you have those massive monoliths as a result of everything from Facebook to some simple components of bigger machines being quick-written. And it's fine because we just don't see it. Things go wrong so often that we’ve accepted it as a consequence of nature, but it shouldn't be the case.
Some problems in computability are undecidable (like the halting problem)—you can’t actually check if a computer is bug-free. But there has been a lot of theory and research in the field of formal verification of software.
Formal verification is being able to prove that a piece of code will respect the specifications we give it.
For example, take Amazon S3. It’s the Simple Storage Service of AWS. You can basically put files there and it's available everywhere on the Earth. They went through a lot of trouble to prove that the core functions of S3 respect the spec.
Basically, they formally proved that the core part of S3 will never delete files it’s not supposed to delete.
Formal verification is extremely difficult and we have to make it easier. The interesting thing is the whole blockchain thing injected a lot of money into the field because people wanted to formally verify contracts because you only write a contract once and then you put it on the blockchain and you can’t change it anymore. I mean, now you can, because there are upgradeable contracts, but let's say you can’t anymore for the sake of the argument. You want to make sure that people can’t just steal the money out of the contract, right?
Picture a bunch of people in a spaceship going to the Sun for experiments or whatever, and then the whole ship just crashes. You have a blue screen of death because the spaceship runs on Windows embedded and nothing works. And this will happen. Linux is buggy. A lot of core pieces of software are buggy.
There’s a funny joke in the industry of software verification, where you have a bunch of dudes working on a missile. And one of the new engineers on the team talks with the manager and says “there is a memory leak in the missiles.”
And the manager says “Hey, it doesn't matter. The lifetime of the missile is 180 seconds. We added extra RAM so that the missile explodes before everything locks up.”
So it's actually quite funny because the way people deal with that right now is they just make software that reboots, or they just add more memory, or they make computers go faster.
I really think that the more ambitious we are in terms of controlling the physical world, whether it be through deploying robots or building massive spaceships, the more important formal verification will become.
How to get more people into hard tech
We're going to approach this question from the eyes of entrepreneurs. We could approach it from so many different sets of eyes—it's both a problem of regulation, a problem of funding, and just a failure of the nerves from people to think that stuff is not possible.
But let’s approach it from the perspective of young people.
For someone who's like 20 years old, really smart, and wants to work on those problems, something you should not be doing (in my opinion) is focusing too much on short-term gains over long-term gains.
Today, for smart people, it is very hard to sit down and go through the three books from The Feynman Lectures on Physics.
Let's say you want to become good at physics.
It's almost impossible to do it because every day on Twitter you have friends who’ve raised 6 million to do crazy stuff. And so every single day, you open your books, and you take your notes and you start writing stuff, and you have to solve those equations.
And every single day you tell yourself, why am I doing this?
I could just go out and bullshit investors and build a company. And I think too many people actually do that. Myself included. I managed to resist for a while and I spent a lot of time learning different, difficult things, but it's very hard not to have ADD in this world. It's very hard to stay focused on important things that take a while to be learned.
I think the reason why we see so many young people building short-term software companies is because the requirements are actually pretty low. Anybody can learn programming in six months and go out and build something that is big enough. And when it becomes very hard to scale, you just have to hire people.
The problem is that this works for software, but when it comes to any other industries, like hardware or space or biology, I really don't think people can go and take five courses at a university and read 5 million blog posts and then build a company around that.
They must be bullshitting themselves. And the problem is intellectual. Honesty is extremely important, both for yourself and for your sanity, but also when you work with others. And so I think that people should spend more time learning very hard stuff and be intellectually honest with themselves. And that means learning to improve.
Spend more time learning difficult stuff
There's this thing called TKS in Canada—The Knowledge Society. And it's a bunch of kids who are 16 years old and they take courses in quantum computing and brain/neural interfaces.
And then those kids put it in their Twitter bio, “I'm a quantum computer algorithmic designer at 16 years old.” And that’s cool. Those kids are super ambitious and they probably work hard, but there's no way they can actually do the things that they pretend to be doing. A friend I know mentioned a TKS guy who has blockchain developer on his Twitter bio. And the TKS guy asked him how to learn HTML and CSS.
Don't try to go too fast for those things. Learning mathematics and physics properly is extremely hard. I'm not saying you need to do it for every single life path, but I'm saying that if you want to do something very hard in the physical world, there are prerequisites you can’t ditch—you have to go through them.
Too many people these days are just trying to do things too fast and not waiting for two or three years. Waiting doesn’t mean you’re slacking off, you’re actually doing really hard things. You’re basically fitting a bachelor’s and a master’s in two years. This is the kind of stuff you need to do if you want to be able to build those super hard products you dream about.
My one advice I would say is that just spend more time learning difficult stuff and don't try to take shortcuts too much because if your foundations are not good enough you will have problems along the way.
I know other people would disagree with that, of course, but this is one of my stronger opinions.
Why more people should learn physics or math
Go into physics or math. The thing is, whatever you do, being able to have very quick estimates, and having some rough rules when it comes to the dimensionality of things and stuff is super helpful.
I would learn physics because I actually do think it's important. You're literally learning about the rules of the universe.
So whatever you do, even if you just work with computers, physics is useful. Even if you want to tell yourself “I wanna be a software guy, I don’t have to care about the hardware.” Well, learning how the universe works will always give you an edge over people who don’t know how it works.
You can follow Justin on Twitter at @JustinGlibert if you haven’t already. If you want to be notified every time we send out a new interview (we have a few more coming up soon), you can subscribe here. For any questions, feedback, or requests, send them to me via Twitter.