BUILD

SUCCEED

David DeRemer:

Hi, I'm David. This is Build to Succeed from Very Good Ventures. Today we speak with Pratap Ranade, co-founder of Arena AI. Pratap shares his incredible journey from studying advanced physics through Y Combinator, McKinsey, Palantir, and ultimately co-founding one of the most interesting and innovative AI startups that are out there. It's packed with insights and I think you're going to really enjoy it. All right. Pratap, welcome. Thanks so much for coming.

Pratap Ranade:

David, thanks so much for having me. Great to see you as always.

David DeRemer:

Likewise. I was thinking the other day about how you and I ... You're one of the ... Not oldest in terms of age, but longest tenured people I've stayed in contact with because all the way back to early 2000s our kickball league, but I think you were still at Columbia at the time and it's just been really cool to see your journey and all these years later still have an opportunity to connect with you.

Pratap Ranade:

No. Thanks man. Yeah. It's awesome having longtime friends who've seen so many chapters and you've seen so many chapters of their lives too, so this is a privilege.

David DeRemer:

Yeah. Likewise. I think it's a great place to start because of all the people I know you have such an incredible set of chapters that got you to where you are today, and so I was wondering if you could just to get everybody in the audience caught up with who you are and the things you've seen and environments you've been a part of, take us through that journey from physics major to now running an AI startup.

‍

Pratap Ranade:

No. Totally. Happy to. So I had a meandering walk, like a random walk a little bit through this as you alluded to. I started out thinking ... I was the kid who didn't grow out of the, I want to be an astronaut phase, and so that just continued through college, Stanford blinders on, just do physics. And then I think it was my junior going into my senior year, the second space shuttle at Columbia exploded on re-entry and you had no more manned spaceflight programme. This is pre SpaceX. And I'm like, well, I only really have a physics skill set. Everyone goes into grad school. So continued on to grad school to be a researcher. Which is actually where I met Ryan who connected us in the first place. He was my grad school roommate. And so continued in grad school for a long time and that was actually in a strange way now today Arena is finally solving problems that I had as an experimental physicist.

David DeRemer:

Wow. That's cool.

Pratap Ranade:

I spent most of my time building physical hardware and I was bad at it. And if you think about the way experimental physics works, you have theoretical physics where you think about a theory and then do the math and you're like, "Great. This is my idea." And then what the experimental physicist does is they build the equipment in the actual experiment to go and sample the universe and see if that's real. You basically, if you think about in my case, you're like a shitty hardware engineer, a shitty electrical engineer, a shitty mechanical engineer and a shitty plumber, all to try and create this little experiment that can tell you about quantum mechanics. And if you think about where this goes to is you publish a paper and hopefully the results are good in telling you about physics and 99% of the time the problem is that you screwed up your experiment.

And so I was a bad hardware engineer basically trying to build advanced hardware. That was chapter one, which will become relevant when we get to today. But then of course after grad school, I was in New York and I was inspired by all of the business world. It was moving at a million miles an hour around us. And so actually I went to McKinsey to do some consulting. I'd never done econ at all and I figured this is a good way to learn about the business world. Was there for a few years, got to work in a few really exciting countries in Africa and China and Europe and then missed technology and I left and actually with Ryan, our mutual friend and my old roommate, we started our first company. So that was out on the West Coast. Technically the first few weeks were in Vietnam, but then we went and worked out of his parents' basement for a while in Portola Valley. I still remember we interviewed ADA users, they all said they didn't want it. We panicked right before our YC interviewer were like, "Well, nobody wants this." We pitched it anyway. We got in. And then I remember Sam Altman pushed us pretty hard to launch.

I still remember the first meeting lasted seven minutes with him and he was like, "Have you guys launched yet?" We're like, "No, not yet. We have bugs." And he said, "Well, fix the bugs." We're like, "Look, there's a lot of bugs. We're trying to build a general purpose, no code web scraper for any website on the internet. We're not going to fix the bugs any time soon." And he was like, "Well, I want you launched in two weeks." And then we seven minutes into the meeting, he was making an expression at the time. He was like, "Why are you guys still here? I thought you had bugs to fix." And so that was our first lesson of, okay, ship, launch. That was a really momentous experience.

We went back. We launched in two weeks. It went viral on launch. We grew over the course of the company, well over a hundred thousand users. Sam and Peter Thiel and Max Levchin all invested. So we had a very crazy a chapter and then a few years later we're acquired by Palantir. So that was the next interval commonalities between Palantir and McKinsey now is we saw again another lens into selling into large complex organisations. And so fast-forward to Arena, we started in late 2019, 2020. And the principle is very simple. My co-founder Engin who I met at Palantir and I basically said one is we want to work on a frontier technology. And broadly speaking, I feel like in any 10, 20 year window of human civilization, there's a couple of technologies that define that era. You can think about the mobile phone, you can think about the internet, the early web. And for us, AI was one of those.

I'd taken a bunch of machine learning in grad school. He was an AI scientist. We're like, "Look, there's something interesting cooking here." And this was before LLMs. This was a deep mind era, AlphaGo and early open AI with emergent tool use and things. This is super cool. And then we're like, "What are we good at?" We're like, "We're weirdly good at selling to enterprise." And we said, "What would it look like to do a company than frontier AI for enterprise?"

So chapter one of Arena was very industry agnostic and let's go solve hard problems with advanced deep learning and reinforcement learning. As you mentioned, a lot of that early problems were very different from the problems we solved today. They were still very much at scale having a real impact, but there were things about sales and pricing for CPG companies. That was a lot of the early traction. And then we did one pilot in the hardware engineering industry and it went insanely well. It was smaller initially in revenue, but we were like, "Wow. This is really exciting." And we just personally were drawn to it. Fast-forward a couple of years later, post ChatGPT, I think a lot of things changed obviously. Two major-

David DeRemer:

The whole world changed.

Pratap Ranade:

The whole world changed. Thank god. Two major implications became clear. One is ... And then if you go a year or two years after ChatGPT now you see agents emerge. We're like, look, this is a new paradigm. So basically by default, I think in an agentic world, if you aren't agent first, you probably are on a legacy stack. So we're like, "Well, embrace it or call ourselves a young legacy companies. So screw it, we're going to embrace it. We're entrepreneurs. We like risk." And so built agent first. That was the first thing. And the second thing was on a business level, we were agnostic. We had the hardware engineering product, we had a CPG product and we were like, look, in our opinion, to win, we needed to pick a vertical. And if we looked at it honestly, we were deep in two verticals and the question became pick one. And I think now in a world where the cost of developing software is rapidly declining, who knows where the software engineering agents go and these models become more and more general and they can take on more and more ground.

I think the question of differentiation that tended to be a question that startups can kick can down the road for quite a while because you build user base and deal with it later. I think that question we encountered earlier. And so we basically said, look, the hardware engineering problem is a very difficult problem. It requires a lot of applied physics. It requires systems. The agents help a lot, but by no means solve the problem. You require additional machine learning systems and the value is very fundamental. It's like can you speed up the work of a hardware engineer? It's not like a blog where you're going to author something and like, oh, it's subjectively good. It's like this is a flight computer for a rocket. It needs to work and you're not going to lie. It's either going to work or it's not, and it's going to be very spectacular if it doesn't work, so you got to get it right. So there was that determinism so we doubled down and we doubled down on hardware engineering. It's been going incredibly well. But yeah. That's a little bit of the story arc until here.

David DeRemer:

Yeah. I love it. I definitely want to dig into that focus, that thing you did. But before we get into some of those details, given that you've been at this for a long time, you've been exposed to these things even in your history, you spent time with Sam Altman, you worked at Palantir, very data-driven. And you were involved in AI and machine learning from their early days and were building foundational models before anyone had heard that term or LLMs were a commonly understood concept. I'm curious from your point of view, where do you think we are in this new life cycle? You said it's a new paradigm. There's a lot of hype out there, and then there's also a lot of scepticism and there's a lot of people like the legacy companies where it's like there's all this crazy stuff where the whole world seems to be changing and then you have tonnes of big companies that are legacy or are slow, are not really ready to make this leap yet, but even their core, basic technology needs to catch up. From your point of view, when you think about the hype cycle or general life cycle of a technology, where are we right now?

Pratap Ranade:

Yeah. It's a profound question. I'll be the first to admit, I evaluated my own projections and decisions over the last two years of AI evolution, I've been wrong a lot. So I was like, okay, wow. Updating the principle. I see a few things that I think point us in a direction of where we're going. So one which I think is very important is what we're looking at here is not so much an invention as a discovery. If you think about what we've built, we've built systems now that are exhibiting emergent properties and emergent behaviour. It was not very unlike computer science of the past where you coded a system, a determinate system, and a deterministic system and executed what you want. You've now built a large neural net put in a tonne of data and it's now finally exhibiting emergent intelligence. This was always the hope with AI is now what kind of emergence is it? It's not an exact copy of how our brains work. It's like an artificial neural net. It's a very simplified approximation and you have different types of data, so you have something very different.

Emergence is happening and it's different. And I think that's meant this thing which even the people architecting the systems themselves don't know. And I think there's a belief here that there's this huge ... I think as Sam Altman mentioned in a recent interview where there's a huge technology overhang, which is the capabilities of these LMS are far beyond the imagination of the products we can build. I think that's certainly true. I think what we see right now is you have incredibly capable, almost like ... I see the foundation model industry evolving a little bit like the chip industry. Very high CapEx. Very good for us as consumers because price competition, it means the cost per token is dropping exponentially and the capabilities are going up. But you almost have a new, well almost like X86 tile processor. You have a new kind of computer that can do general purpose processing for you.

And so I think us as programmers and engineers, you want to be using it, you want to be testing it, you want to be seeing what it can do. Did you have to write the algorithm or could you actually put this arbitrary brain in to solve it? And you know that brain is on a glide path where it's getting better and better. And then what we've seen really, really interesting is now we're understanding the limits of current agents, but now you're having multi-agent systems and agent swarms and agents using tools. And that's starting to become a really interesting ... If you think about these as almost artificial people, but they're aliens, they're not people like us. They think slightly differently. What we're seeing now as building little Petri dishes, like a multi-agent agent swarm systems which are some of the systems we use. They're like Petri dishes where you have a small society. You have a society of different agents interacting with each other using tools. And so you have these emergent systems and you're putting some structure around it and boundary conditions to see how it evolves.

What's interesting is it's a new field R&D that I think is happening. And so the short answer is I think we have to play with them. We have to push them, see where their limits are. But I think we are in this thing where something ... If you think about the internet or even the mobile phone, there were people that were like, "Why do I need a computer in my pocket?" I remember Ryan, we're going back to our old roommate, he launched a tip calculator. That was amazing. He launched a game. And you had these things that became popular. Mobile gaming is an industry. Who would've thought you're going to sit with your thumbs and play games on the subway, but that's a huge industry. So I think we don't know yet how we're going to use this new system. I think a lot of what I read I think is a little myopic because what we're thinking about is a very zero-sum paradigm. We're saying, "Oh, these are the jobs that we can do today. This is what humanity can do today. This is going to eat our jobs."

Now I think what we are seeing is this huge capability uplift. And so what does that mean? A company 10 years ago, what a company a startup would've taken on is probably now going to be what is a feature of a new company? So companies are going to be more ambitious. If one developer is 10 times more productive, you can actually do more. If we look at it at a very scifi kid human level, no, we don't have space elevators, we don't have jetpacks, we don't have flying cars. Why? Economically those problems are too hard to solve? I'll actually ship a small app that helps you ... I don't know. With a productivity tool for planning your meetings every day. Okay, fine. Did you really need a company to do that? That's finally easy enough that each company can quickly wire up its own and that's not a product anymore, which okay, on the one hand maybe scary, that's the products you're building, but on the flip side, the human potential, those humans can take on much more ambitious problems.

So I think what we're seeing is I actually don't believe we're going to be very zero-sum in seeing a lot of this loss. I think we're going to see things that we couldn't do start to become possible for the first time. Even we're seeing this. There are problems that we're like, "Oh, this would've been a 10-year problem. This is too hard. There's no way we can take it on." We're like, "No. Actually we can take it on now. We can ship that in the next six months." And so that is actually I think really powerful and you can get really excited and creative about where your product does. But where this goes, I'm not sure what we've built is really artificially general people, approximate intelligence. I don't see that at least in the current systems that we have. I think we're building a new kind of intelligence that's going to be available on tap and that we're going to use as tooling. I think people are very much going to be critical. I think I'm very excited about embodied intelligence and getting robots into the workforce. I think that's also going to be really amazing. You think about elder care, how awesome would it be to have robots for that? I think that would be great.

So I do see a tonne of these. I think I'm very bullish on robotics. I'm very bullish on this. Personally coming as a physicist, I think what I'm most excited about is I think we can push the frontier of applied science actually significantly forward. Both in terms of materials research, in terms of aerospace work, in terms of drug discovery, and I think that's going to be incredible. But I don't think what we've built here is just like ... This is not a thing that's going to replace David. If anything, I am placing already a huge premium on the analogue interaction. We're in person every day. I value more my time with my friends and my family so much more. I value that analogue handcrafted experience. And so I think we're going to see just this new world. I would be hesitant to predict more than that, but that's-

David DeRemer:

Yeah. No, I get that. That makes sense. I love how you took us back to the ... What was it? Tipstar I think it was that he created?

Pratap Ranade:

Yeah.

David DeRemer:

It was right after the app store came out, and I think that's maybe a good frame for us to think about where we are. People forget when the iPhone came out, a couple of things. Didn't have an app store. It was only Apple apps, and it used that horrible AT&T Edge network with very ... You had to connect to wifi. Do you remember this? The bandwidth was so terrible, right?

Pratap Ranade:

Yes.

David DeRemer:

I want to test a thing I've been thinking about. How I've been trying to contextualise where we are in this AI boom is actually thinking about mobile networks and bandwidth. I think AI is a new kind of bandwidth. We don't quite know how to talk about or measure yet, which is in the early days, you had Edge. It was good for basic things. You couldn't really build businesses around it. But now we all have ultra wideband 5G in our pocket where we all expect at any moment to be picking up our phone and instantly streaming high definition video. And you think about the types of experiences that has created and businesses that has spawned and the types of things we just expect and take as normal now. And I feel like AI is in that just starting to launch the app store mode. And we haven't even contemplated what it's going to mean when those app stores really take root and the supporting bandwidth and other supporting technologies around it fuel that innovation. I feel like there's a bandwidth story here that there's something to just the speed and the processing, the amount of things that you can deliver to someone quickly has gone totally crazy exponential here.

Pratap Ranade:

No. I think that's super interesting. It's this thing that we now have intelligence on tap. It's literally electric utility. It means that there are things that we were not very thoughtful about before that everyone can apply. You can apply that. Or things that didn't make sense for a human to do because it was dangerous or it was too expensive, you can think about anything on tap. And then I think at the advanced level, it's an powerful assistant. But I think in this general human replacement, I don't really see that. I see it as this ... I love your bandwidth point, which is, yeah, I expect my LTE to be on and me to be able to stream. It's crazy. I expect I should be able to stream video just walking around. What an absurd expectation. And that's normal now, and it became normal very quickly. And you think about the number of like, "Hey, I want you to think about something. I want you to use simple tools to do something." It's still actually quite a pain for me to book a dentist appointment.

David DeRemer:

Right. Yeah.

Pratap Ranade:

Why. That should be easy. There's a lot of things that just aren't happening. Yeah. I think it's going to be one of those things where if you have a ... I don't know, a data centre goes down or something or intelligence brown out or black out, I think you're going to have a big ... We're just going to be like, "What the hell? Why can't I book my appointments?" We're going to all get ... I think Yuval Noah Harari talks about this in his book Sapiens, where if you look in the long arc, civilization has become wealthier. What any common person can have access to now is more than what a king had access to a hundred years ago. And so in a weird way, everyone's going to have basically a bunch of sidekicks doing work for them, and that's going to be the basic expectation, which is cool. And especially if some of them are physically instantiated with robots. Yeah. There's actually a tonne. Go get my groceries. That level's probably already solved by Amazon, but you know what I mean.

David DeRemer:

Yeah. Yeah. Yeah. Well, I think we could probably keep talking about this for six hours. I do want to get to your story with Arena and I want to make sure we tackle that and then we can circle back to this. I love it. Thinking about the future is so crazy, especially right now. It's such an interesting time and also scary for a lot of people. I think there's a lot of really hardcore things. I do have some questions that I want to get to around the dystopian futures. Let's go back to 2020, 2019. You had done Palantir, you had done Enigma, I think right after that.

Pratap Ranade:

Yes.

David DeRemer:

And you met your co-founder. I'm curious when you guys thought about starting that, what was the spark behind that idea of getting started? And then obviously you've been through some pivots, there's been a lot of change, both what you guys have focused on as well as the market and the industry around you. And what are the consistent things that you've pulled through that to today?

Pratap Ranade:

Yeah. The starting impetus ... Now since it's the second startup, I think there were some learnings we had. We had a couple of principles that guided us on this, and we actually realised that a lot of startups will iterate and pivot. And so weirdly, we were less sharply opinionated on exactly the product. We were more opinionated on what the company did, the market, the general technology direction and the business model. So actually we had fixed constraints in places where we didn't before and we relaxed a constraint around weirdly, the core product. That was the evolution between my-

David DeRemer:

I feel like a lot of start-up companies don't start that way. They focus with a product they're obsessed with or a user need. Was that your McKinsey or Palantir background?

Pratap Ranade:

Coming out of McKinsey, the first thing we did was build a very product centred company where this is the problem that we have personally, this is the product that will solve it and we'll just launch it. I think that there's a tonne of really positive stuff with that. But one of the things I think I learned at Palantir is ... I think Palantir taught me a lot of things. Incredible company. I have so much respect for it. I think that there's a Maslow's hierarchy of problems worth solving. And I think at this base level of the Maslow's hierarchy ... So if you think about this and you take that and you say, "Look, where are you in your career path?" So you're more likely to take a risk and start a company generally when you're younger. You have less financial constraints. So in that case, what happens is you're probably what, four years, five years into your career? So when you're four or five years in, there's certain problems in ... If you think about B2B, the technology SAP that you've seen, you've seen payroll, you've seen onboarding, you've seen HR, you've probably seen outbound sales, you've seen some of those things. And that's where the companies are. That's what the problems other people go solve.

Then there's two other layers. There's a layer above it where it's like, "Hey, I'm 50 years old. I'm running GPUs at one of the giant semiconductor companies and I have this bigger holistic view and I'm seeing problems that actually aren't just problems shared by an individual, but problems that happen with teams." Okay. I've got a design team, a test team, a validation team, a production team , a fab, and I'm actually getting all these teams to work together. I've chosen ... And actually usually chosen as I've inherited from 50 years ago a job structure and job definitions below me. So there's the designers as the test centre, there's the simulation guys, and that's the construct in which I operate.

Now very few people, especially few people who are going to be entrepreneurs, have witnessed that problem set. And so what I think Palantir taught me is if you're willing to be patient and do something that looks ugly and unsustainable, flying to customers forward deploying, which I think is also often poorly understood to mean solutions engineering, it's absolutely not. It's actually just deeply understanding a set of problems that most people don't have access to, to then find products that solve them. That's really more what it is. And it's like that problem set is relatively under attack. And then at the apex of this pyramid, you have problems that are highly valuable, that are hidden, but they're so bespoke that you shouldn't do them as a product company that should be consulting. And so that's like the stack.

So we said we have built a skillset now for myself, McKinsey and Palantir at getting deep in and solving enterprise problems. This allows us to access the second tier of the Maslow's hierarchy, and that's good because there's less competition and we're more differentiated. So that was principle one. We wanted to take that motion, that forward deployed motion and do it with enterprise. So that was a principle that has remained true even through our pivot. We just change the industry, but the motion and the muscle and the culture around that is the same. The second piece is if you zoom out any startup which is successful, you've got this beautiful hockey stick, but most of the time when you zoom in, it is like this. And why do you and your team go when it's in a dip? Because inevitably the dips will happen. And when you zoom out, like Steve Jobs I think said, you can connect the dots looking backwards. But in that moment you can also, if your mind is having a dark day, you could draw the line and it points downward and you're like, "Well, why am I here?"

So you need something that genuinely excites you. And I think for us, from both of our backgrounds intellectual interest that frontier technology was AI. And again, we started before the LLM boom. So it was broadly speaking, deep learning in deep RL and we were like, "This is really exciting. This concept of emergent intelligence is incredibly exciting. What could you do with it? What could you build with it?" And so that principle guided it. So we were like, "Look, that's how we want it to work."

And the last piece was in the new constraint set was monetization, which is ... At kimono we had a tonne of usage and then we thought about monetization much later. We wanted to think about monetization earlier to help us qualify the problems that we're chasing. So okay, yeah, how much is this worth? Can we get revenue early? But I think for us, revenue's a bit different. We're not looking at revenue specifically as this is the thing that's going to ... That's not the indicate. That's not actually not our primary metric. We're still developing a really powerful product that I think will command crazy revenue. We're looking at revenue really as validation of the problem. This is a hard problem. We're going to spend a lot of time solving it. In an AI world where the marginal value of software is decaying so rapidly, I think it's even more important to think about what is a very hard differentiated problem. But you don't want to go and solve one that's not valuable. And so that was the other constraint. So if you think about the update between Kimono and now, it was revenue early to validate problem, go deep and access that problem and this deep Maslow's hierarchy with enterprise and pick a frontier technology you're rationally excited about. In our case, it was AI. So that became the founding kind of problem statement led to Arena.

David DeRemer:

Awesome. Great recipe there. I think that's something that maybe a lot of people, if they're entrepreneurial and they're thinking about starting a startup that hierarchy. A lot of times I think people who are just starting out, they're thinking about there's a problem I face doing something. How do I book a better dentist appointment like you said before, or something like that. And it's a problem that I've solved. You're like, "Okay. Lots of other people like me have that." But I really love that thing of taking it up to the next level of teams and companies and industries, having that bigger picture perspective and then putting those real constraints and guardrails. I love the idea of revenue as validator too. Totally agree. I started a consulting professional services business where it's very, very clear if you're successful or not because either clients sign up with you or they don't. You get feedback pretty fast. \But I think you're right that revenue is a validator and sometimes people lose sight of that. It's the if we build it, they will come mentality. And I feel like if you're building it in a product experience and you don't have something where you feel confident someone would actually pay you for it, probably not a very good business.

So let's talk about then ... I know you guys have been through a bit of a journey here and you mentioned earlier that you had focus on CPG and hardware. How did those emerge as ICPs or targets or industries you're going after? And then you guys made recently this real clear focus onto hardware, and I think that's just a really interesting thing of that exploration phase of pursuing a bunch of different fronts and figuring out where to go after it. And then identifying the one that you're like, "We're going to go all in on this and really own this category." Take us through that sequence and how you made some of those decisions.

Pratap Ranade:

Yeah. Great question. The first thing, which is funny is ... Again, connecting the dots, looking backwards, this makes a lot of sense. I'm finally solving a problem I personally had when I was a physicist. And two is I looked back at our very first website that I made on a Saturday right when we were starting the company. And actually the website basically said, "Hey, we're going to go and build out these large AI systems initially for markets to prove out that they work and can make money on real data in order to then do them for robots and spacecraft." And so literally hilariously, we're actually doing exactly what we said on the website because that was where our passion lay. But to your point, actually, the journey was interesting. And so the early traction, again opportunistically, the first question we had ... Again, this was before language models, was, well, it's all well and good to say you're going to have revenue, you're going to go build with advanced AI that's incredibly data hungry and you're going to solve this class of problems, but now suddenly you're a couple of guys with some MacBook Airs trying to convince a enterprise that has that data to trust you to deploy an active learning system.

In the wild, we were like, "Okay. This is just a showstopper if we can't convince someone to do it." And so we were lucky enough to have some great early partners. They were in CPG that said, "Hey, that's an interesting enough idea. Let's let you try." And I think as we've evolved on the market, I think we were incredibly lucky that we had some specific executives who were quite visionary, very deeply thoughtful, who took that bet on us. And I think in that industry, that's actually a very rare thing. You commonly will have that more in an advanced technology industry where you have to think about that. But you think about the CPG business writ large, you're preserving large brands and you're optimising on something that's existed for a long time. And so it's rare to have that kind of leadership. And so we were very lucky to have that. And that actually got our business off the ground. We were able to develop a lot of these systems, a lot of real world reinforcement learning for full automation, prove out value, AB tested in real markets, prove lift, build robust systems that had hundreds of millions of dollars flowing through them. And so that built a tonne of the muscle in the DNA. I think that is still true today.

Post LLMs, I think a few things changed. So one was just I believed from a business perspective, it would be very hard to be vertical agnostic because I think if you're already a very large company, so you're a Databricks or a Palantir or a large info provider. You can be very vertical agnostic or you're open source. But in a world where that much capital is going into AI systems, how do you just win a vertical. Sorry. Win horizontally. And so we were now in two industries.

‍

And so then it begs the question, okay, if we believe that and we believe startups need focus to win, then we shouldn't be in two places. If we believe we need to pick a vertical and we're in more than one, that led to an immediate intellectual conflict where now you have to choose. And then the hard thing is, well, there's a very simplistic version where you're like, oh, choose the bigger business today. And I actually worked through this and wrestled with this a lot because there were two things that felt off with that. One is the nature of impact you're having. We're not actually concerned about today. We're concerned about how big this company can be. And whether you're talking about single digit, double-digit million dollars, no one's actually investing that a company as a double-digit million dollar revenue. No one actually cares. You're talking about something that could be incredibly large.

And so you care more about the projection going forward. And so then the question becomes how much real value are you creating? And if you look at where the industry, your earlier question about is going, how is that value creation going to increase with time? And when we saw what we were doing early days with hardware we were performance optimising the most cutting edge semiconductors on the planet and making them better and actually beating humans at it. And on the other hand, we were getting paid a lot more on the CPG side. If you ripped out all of the technology from a CPG company you still have a pretty great CPG company. If you're a big beer company, probably the number one concern right now is not technology. It's you should probably buy a non-alcoholic beer company. It doesn't matter that much in that industry.

Whereas if you're in an advanced hardware engineering industry and you're allowing someone who's shipping advanced technology to ship faster or better, you're directly solving a very material problem for them. And also going back to what we said about zero-sum versus abundant worlds, if you're a hardware company, you are rate limited by how difficult the fundamental physics of the problem is. If you could do more in that time, you wouldn't be like, "Oh, my jobs are going away." You're like, "Oh gosh, I'm going to ship a 10X better product and I'm going to win." And so it's a very strong incentive alignment. If you can make them faster. There's not a shortage of stuff to do. They're like, "Oh my god, great. My next ship is going to be better."

And then what was really interesting and what we found to be the case is there's conventional, let's say old B2B SaaS thinking, which was great, let me sell to my 100 other friendly startups and I'll mature my product and then I'm going to go and convince the enterprise. Now, let's think about it. AI systems have a reputation for being unreliable. If you're trying to go and convince ... I don't know. Whatever you think is the most advanced hardware company in the world. You're picking a giant space company or a giant aircraft company or a giant car company, you're trying to convince one of them to buy your product. Are they ever going to look at your and hundred other startups and say, yeah, because so many startups use you, I think it must be ... No. It's actually probably going to look a lot more like Nike, where you need Michael Jordan to wear your shoe and then everyone else wants the shoe. No one actually is a Michael Jordan level basketball player. They still would like to wear the shoe that he wore.

And so it's a very different mode and that's a much harder problem. And so we realised, we're like, "Wow. We have a few things." Smaller revenue, which made it hard, but organically growing. While we were paying attention to the other business, this thing quintupled, and we're like, "Wow. We're doing something else, this thing quintupled." Macro forces are driving adoption. We're highly relevant to them. And then we were earning the trust of probably the most demanding ... The Michael Jordans of the world had started working with us and we said, okay, there's less money today. But these guys are just solving ... We're at the early innings, especially with what AI can do. It's only going to get better every six months. The percent of the problem we're going to be able to solve is only going to increase. And these industries ... Just look what's happened with Nvidia. Look at the AI data centres that are constructed. Look at the new space race, the new geopolitical world with defence. Those industries are going to get propelled forward through massive macro forces.

And then the final thing I'll say is if you look at where everyone went as individual labourers, we had a 90% uptick in the last 50 years in software engineers, but a similar downtick in hardware engineers. So for every dollar going into one of these hardware companies, there is an overworked 55-year-old hardware engineer with a tonne of expertise that is trying to ship really fast, really advanced things that can literally explode with not enough time, not enough team, and is using tools from literally the '80s. We are using Copilot and Claude code and advanced IDEs and linters and debuggers. Most programmers, they're a programming language that's abstracted away so much of the hard stuff, they're not even dealing with how the computer works. On the electrical engineering side and the hardware side, you literally have people wrestling with electrons and a pretty raw level with old tools. In a weird way, we were just like, this opportunity is so big, very difficult problem. But again, a lot of our team is weirdly, somehow organically, even though we were in CPG, ex-physicists, ex-space flight engineers, and we're like, this is what gets us up in the morning. So that was ultimately what led to the decision.

David DeRemer:

I love it. I love it. And this is Atlas.

Pratap Ranade:

This is Atlas.

David DeRemer:

Maybe for people who haven't come across Arena, do you want to do the quick pitch of what Atlas does, specifically how it works and how it actually helps these electrical and hardware engineers you're describing?

Pratap Ranade:

Very simply, Atlas is J.A.R.V.I.S. What we're trying to do is to make hardware engineering easier. Hardware engineering is much more difficult than software engineering. The software, we have a REPL loop in software that's very tight, it's very easy to write your code, run it, works, breaks. In hardware that REPL loop is slow and it's expensive. And especially if you think about modern hardware. Compare a modern fighter jet to a World War II or a fighter jet. The main change is the electronics. You had hydraulics and metal, and that part is, yeah, it's evolved, but it's the same. Now you have advanced electronics, avionics, flight computers, sensors, and it's perfect if it all works. And all of that is dictated by basically electrons. So you've got your sensor is sending a signal, which is an analogue signal that will then be converted to a digital one that then will go to a flight computer that will then run your code. That is probably subject to a lot of stray electromagnetic fields, especially if you're going into space and through the Van Allen belts. So you have a lot of physics is happening here and you just need that to work.

So ideally our code runs, but whatever your input to your flight computer is wrong data because the electronics of your sensor are screwed up or uncalibrated, your flight computer made a beautiful decision, but it thinks your plane is nosing up, but actually your plane was level and it puts it in a nosedive. These are real problems. And so you think about those are the problems we're solving and what does it look like today, Atlas looks like today? Today basically our focus is largely electrical engineering at three levels of the stack. One is you're working at the chip or the PCB level and you're trying to functionally understand the thing that you're building. So you're checking a design, you're actually understanding its functionality, you're interacting with Atlas to ask questions about the design, to understand it, to traverse and find connections. In a complex design, you're trying to actually calculate tolerances. You're trying to run simulations like SPICE to understand functional behaviour. You're trying to generate a test plan that can take you days. We generate that automatically. And then you want to execute the test. You want to interface with your oscilloscope, execute the test, interpret the data. So that's really where we start right now is a lot of it is verification, validation and testing on the electrical stack.

What's interesting is what you're building is this EE partner expert. So you're helping these EEs in expert mode, but I think a lot of you have played with language models now. And what's one of the cool things is you could give it a book and say, "Explain this to me like I'm a literary professor of literature," or, "Explain this to me I'm an ESL student," and you're going to get two different answers. And so the same paradigm is true here. So for example, at the systems level in automotive, we have one of the top three automotive companies in the world is using Atlas and they've assembled an amazing, a beautiful car. It's coming off the line and there's a small light that's not working. 40% of the time that error is electrical in nature. And now you have a high schooler, a high school graduate who's sitting in a plant ripping out carpet in seat to try and trace kilometres of cabling and figuring out where it's coming from. And again, the same brain now operates here and is saying, okay, it's most likely here. It's like laparoscopic surgery for a car or an aircraft.

And then in the case of a fully autonomous system, you can now combine the fact that you have the deep understanding of all the cabling, the electronics, how things should flow, the simulation. And now you just add in a little bit of sensor data to make sense of the flight physics. You're like, "Oh, my drone fell out of the sky." Okay. Now your autonomous system. So a lot of it is basically, okay, assuming the mechanics are fine, you could probably visually check the mechanics, is the rotors intact and stuff? But then probably electronics are a firmware failure. And now you can actually root cause it. But what's interesting is this isn't some backwards facing anomaly detection or causal ML that's comparing statistics. This is actually you know what CAN bus you're using for communication, you know what wires are connected to what wires, where the sensor is, you know what code's running it. You actually can intelligently debug. And this is where our long-term vision is basically self-repair, self-healing.

The drone ideally should tell you how to fix it. I would love my AC unit to tell me how to fix it in simple language versus me having to call a tech. With AI, we can put that expertise into the AC unit. And so that's actually long-term. Where this goes is from Star Wars R2D2. So Jarvis today, R2D2 tomorrow is how I would simplify Arena.

David DeRemer:

I love it. It's mind-blowing, man. It's really cool. I remember seeing when you guys announced Atlas and where you guys were going, you had some videos and things on your website and social media that you put out. I had this moment where I was like, "Oh, great. Wow." There's something crazy happening that you guys are doing and helping to drive a vector of AI that I think most people not seeing because they're focused on ChatGPT. And I read somewhere recently, somebody said, "We have figured out how to shock minerals with electricity to think like a human brain before we even understood how the human brain actually works." Because a lot of these elements simulate a lot of the things about human cognition, how we work, memory prediction, all these other things. But when I saw your stuff, it was like, whoa, now we're actually coming up with AI that actually helps us to create the very machinery, silicon chip sets, all the hardware that actually those things use and are built on top of.

I had a little bit of a Terminator moment there where I was like, oh shit. But I think we all have that, right? And I like to think how you said before of no, what this really going to do is unlock a lot of potential and new experiences, new opportunity. I think it's incredibly cool you guys are applying AI to the physical hardware side of this, which I think if you think about it, is a necessary condition or a loop for these things to really get these scientific breakthroughs, to get this technology breakthroughs. As much as we're doing with user experience and how gen AI and things are changing software and how people interact with information and data, applying this stuff to the underlying hardware systems, electrical architecture, all these things seems actually like something we don't talk about a lot as an industry, but seems a very important critical essential step to keeping pace with the rate of change. So what do you think has been the hardest part of bridging this software and AI with the physical systems? When you think about getting into that substrate and that focus versus focusing on how do I build AI to build websites? It's a very different outcome. What's been the most surprising or difficult challenge?

Pratap Ranade:

The physical world is really hard because it's unforgiving and it doesn't lie. There's no, oh, this is a good blog post. I'm going to start. There's no subjectivity in the physical world. All of us, our human bodies. Our world is still governed by the laws of physics and the laws of physics still supersede what mankind has built. That's still our limiter. Why don't we have space elevators? Why we haven't solved the applied physics problem. Well, we haven't. We know it's solvable actually. Physics tells us it's solvable. We just haven't figured it out yet. And so if you think about that, there's a couple of things that are very hard with hardware. One is the behaviour of these systems is actually rooted in applied physics, which I was actually surprised by this. But I would say on average, the average software engineer is slightly scared of advanced math and physics.

There will be many, many words for it, but there is a problem avoidance that happens. And so what you have is like, oh, here's Maxwell's equations. These are four PDEs. Okay. A you have to engage with a very difficult problem and a very difficult field that is just ... It's not easy. You have to use complex numbers. You can't just live with real numbers. It just starts very, very basic where no, you have to use partial differential equations. You have to use complex math if this is actually just the way the world actually works. If you don't find that fun or aren't curious about it, not even a starter. So I think the first thing, which is that is a hard thorny beast.

I think the second thing is there's the toy problem. The joke about a physicist is you basically are like ask a physicist to describe a cow, and the physicist will say, "Assume the cow is a sphere." And so we love to simplify these problems into toy problems that are solvable, but the real world is messy. So if you look at this as you've got ... For example, we have a small hardware lab here where we actually field test Atlas to dog food it. We have a small flight computer, it's connected to a small IMU. The goal here is to be able to enable anyone with a no hardware or EE experience to build a drone and actually field test ... And not just a drone with your Arduino but make the flight computer, make the sensor, make sure the whole system is working. And so here's a very simple case where, okay, you have an IMU, it's telling you your position in space and acceleration. So what you expected is your IMU sitting on the table, no movement in X, no movement in Y, and you should feel acceleration of free gravity in Z. And that should be 9.8 metres per second squared.

And your very simple example is, I've got my flight computer, I put some wires, I connected to the IMU, and I should be reading 9.8 metres per second square. Actually, both of these things look fine. I should connect them, it should work. And weirdly, that doesn't always happen. And this is not because those two deterministic parts of the system are necessarily broken. You have wires that connect them. Now every wire is an antenna and the wires are picking up stray electromagnetic fields, and that's creating interaction. And now you're seeing these weird parasitics that are interfering with your communications protocol. And so you might not be reading 9.8 metres per second square. And if you want to put this on a drone, you pretty much need the sensor to tell the flight computer what's actually happening. And so just something that simple. Accelerometer is working right, the flight computer is working. But if you don't pay attention of how you wire it together and you don't know, you can't measure all of this.

And so that's where you also have this additional complexity of you need to build systems that work, that are highly context aware of what's happening in that physical environment. And the physical environment is all of the electromagnetic radiation around us, which is significant matters. And so I think these are some of the problems that actually make, I think, operating in the physical world hard. One is you have this applied physics that you have to wrestle with, and then two is just so many things that can go on. And so your root cause debug processes very wide. The surface area of that is extremely wide. You're like, "What could it be?" It could be my fluorescent light is operating at a wrong frequency in my test lab. It could be anything.

David DeRemer:

Totally crazy. Well, that's wild. And that's a problem set that a lot of people are not at all considering or thinking or dealing with in their day to day. We get a bug, oh, reboot the computer, or something like that. Crash, restart. It's like, no, we're not figuring out. If I turn off the lights is in my flight computer going to work a little better? Totally crazy. It's super cool. It's very, very exciting thing. You mentioned something before about just this math thing. You have to find those certain things interesting. And these are really advanced problems. This isn't just a smart person with a CS degree can come in. There's a certain degree of focus and passion and energy for these things. As you've pursued this as a leader of a high performing team of very, very smart, capable people where you've had to navigate a bunch of changes, 2019, 2020 when you're starting, COVID happens. The guy who told you to fix the bugs is all of a sudden launching ChatGPT. And was probably both awesome because it unlocks this category, but also probably terrifying. And in this pivot, all these focuses. What are some of the management or leadership approaches you've taken to keep everybody marching in their same direction, keep everybody engaged and motivated and focused on tackling such a hard problem?

Pratap Ranade:

Yeah. I wish I had a silver bullet on this one.

David DeRemer:

You and me both.

Pratap Ranade:

I'm on a learning journey myself. And I think part of that is I feel like as a team that's investing a lot together in a startup, I've evolved my own leadership style. And I think a few things that I've learned, I feel like, and that I think hopefully served us well here is there's a degree of authenticity and transparency that you have to provide. You have to be an honest leader. And sometimes things are hard. Sometimes things didn't pan out the way you want. Fundamentally, startups are risky things. It statistically doesn't make sense to do a startup. You shouldn't. It's not an index fund. It's unlikely to succeed. So you are betting on yourself. And I mean that for every team member is betting on themselves and their teammates to be the outlier.

‍

And so what's interesting is you have to be incredibly direct and honest because you shouldn't selling your team on the vision. You should be presenting the opportunity that we have, why it might work, why it might not work, and then have folks opt into what the kind of hard looks like. I think the view I've had is look, if you're going to do anything well, it's hard. There's no such thing as their free lunch. You get to choose the kind of hard and you get to choose who you share the hard with. And that's the thing which has I think been a key element over here.

‍

Another principle we've had is very much, we just had an interview for senior engineer. One of the things we believe is it's the leadership muscle at Arena is a field commander mindset. There's nobody who's a pure manager. If you're going to send your team into battle, you're going into battle too. We had a team that basically for the last month was in an epic surge pulling late nights and weekends and it's like the team leads were on the ground with them. And as many weeks as I could be on that specific deployment, I was on the ground with them. And the idea is basically, look, we're not in this alone. You're never going to be alone. Our job as leaders is to be in it with our people, asking people to do things that we ourselves are willing to do. So I think that's been actually really ... Because especially you pivot, it's hard. There's a lot of new things to learn. There's a lot of intensity you need to bring. So those two have I think really been important.

‍

And the last one is just this principle that we have. It's one of our values on the website is kindness over-niceness, which is, and it took me a while to learn this, but it's like we're nice people. I think people at Arena are very nice, but what we want to do is you really want to be kind. And so the idea is, for example, let's say someone didn't do something particularly well. Now you might be, oh, I don't want to tell them about that. It's not a big deal. It'll hurt their feelings. But they don't hear that feedback, they don't get better. That continues to be a thing. And then later on some harsh external force happens, some big thing drops and you're like, now it's a very severe conversation. So something that my co-founder, Angin loves to say is he is like, "Look, it's all about giving feedback and doing things when the stakes are low."

So instead the kind thing to do is like, "Hey, here's all the awesome stuff you've done. Here's something that didn't go well." And just do it quickly and do it right away and give that person the opportunity to grow into that. I think sometimes in our quest to be nice, we rob folks of that opportunity. So that's been another principle. I think that supersedes the pivot even. For us, that's just been a real ... And it requires continuous work. I think we can also often default into ... Because it takes a lot of work to do that because you have to think about it, you have to write it down, you have to pay attention. It comes from a place of caring about the person, but truly caring about the person is not always telling them what they want to hear. This also goes for customers, which is customers might want X. You actually are living inhabiting two worlds. You're inhabiting your world on the frontier of AI and the customer world. You might have a better idea, but it also might require you having a little bit of a debate with a customer that can be uncomfortable. So I think pushing into those uncomfortable zones is something that we've been doing and I think it's a muscle we want to continue to practise and grow.

David DeRemer:

Yeah, really, really wise advice. I love that framing to the commander. I think for different businesses you need different metaphors for leadership. Certainly some industries, some positions and roles you need a boss, you need somebody who really tells people what to do. Some roles, I think you need a manager, someone who's helping them, really managing, directing their work. And then some industries I think you really need a leader. Someone who's like, "Follow me because I'm running in first." And I think it's important for a business to find the right framework. The other thing, the CEO thing, it's like you're the journey. There's somebody I follow online, he's the CEO of this company called Stitch. And he posted a thing recently that the job of the CEO is to figure out who to disappoint, to be mindful of who you're going to disappoint, but then make that tough decision. And so when you're making these tough choices you just can't please everybody, and I think you have to have that conviction and I love that thought of that transparency and giving people like, "Hey, here's the vision. You don't have to agree. Who's going to opt in? I'm not compelling you to do this. You have a choice." I think that helps to drive empowerment. So really I love those lessons. You should write those down and it'll be the framework for your book one day.

So just to wrap this up, I think first of all, this has been incredible. I feel like you're one of the most brilliant people I know and just your ability to parse through all these complex ideas and deliver them in a way that's both engaging and exciting is awesome. You've had such an incredible experience weaving through this thing and actually just unbelievably awesome that you started in physics and here you back are after all that journey and all those things that you did led up to making you better at solving those initial problems. Is there a principle or something that when you look back through that course of history that has been a core principle that you feel like has really truly evolved and grown for you over time, that has been a consistent thread or principle that's helped you to lead and grow and tackle these problems with enthusiasm?

Pratap Ranade:

I think yeah. It's very interesting when you ask it that way or what has actually been true consistently in all those situations. It's a very good question. I have to feel like it's the following, which is I think this is definitely true for me. I think it's true for many people. It's like when you don't want to go work out, but then you work out, you still feel better after. You're tired, but then you go and somehow have more energy. It's this thing about giving the importance of giving and giving, generosity for sure, but I'm not speaking about that. I'm speaking about giving yourself to a thing. I think this idea that it almost matters less what it is. We talked about this with changing the direction. But is making sure that whatever you are working on right now ... It doesn't matter if you're working as a barista right now, whether you're learning a programme, whether you're a student, whether you're an athlete, whether you're a founder, you have to be all in. And weirdly, that honesty with yourself about going all into it.

And in some ways it's hard to say, this is what I'm doing for the rest of my life, but relax that criteria. No. Go all in for a period of I'm going to go all in for a few years. And I think what happens there is this beautiful transaction where you give so much of yourself to a thing and nothing happens realistically at 40 hours a week. We're talking about just that becomes who you are and then that gives back to you and you grow. And this magic happens where your deep investment in a thing creates this cycle where you grow and it gives more to you than you wound up giving to it. And I thing can be working with a group, it can be working with a team. But I found that the one principle that's been true is when I was doing triathlon, I was like, I want to go to the Olympics. When I was doing physics, I was like, I want to be an astronaut. Whatever the thing is, that should be the primary main obsession. And I think whether it works out or not in the end doesn't even matter. It's like that transaction is like the core essence of life and growth.

Even with people giving to others creates deep friendships and value back in new ways. It's the opposite of short-term transactualism. It's like, no, it's long-term commitment, but repeated active choices that create long-term commitment and it's better for you and everyone around you. So I think that's been a theme that I feel like I've started to internalise myself as a principle.

David DeRemer:

That's awesome. Love it. Super cool. This was great. We could keep going for another three hours. I feel like if people want to find out about Arena, obviously you guys are doing really incredible work. If someone stumbles upon this and wants to learn more about how they could use Atlas or your team or if they're interested in joining your mission and proceeding, where can they find you? Where can they find you personally? Where can they find Arena? How could they get involved?

Pratap Ranade:

Yeah. Perfect. So you can find us at arena-ai.com. So you can check out the website over there. I have a substack, so it's again Pratap Ranade. I'm on Twitter, same name. You find me on LinkedIn. So again, feel free to message me, feel free to message the company. If you're interested in what we do careers at arena.com. You're going to get our fantastic recruiting team is going to respond to you right away. And yeah. We're I think a really fun crew. If you want to come build Jarvis, please come join us.

David DeRemer:

Yeah. Who wouldn't want to get on board with that mission? Sounds pretty cool. Well, are you hiring? No, I'm just kidding. Cool, Pratap. This was awesome. Just packed with insight. It's going to be very difficult for our marketing team to go back and figure out what are the key things to highlight. But can't thank you enough for being on here and for just being a good friend over the years. Really appreciate it.

Pratap Ranade:

David, thank you so much for having me. It was a real pleasure.

David DeRemer:

Awesome.

Pratap Ranade:

All right. So we'll hang out.

‍