Latest / Pirates Only / Smack and the Fight for Decision Dominance

Transcript

Smack and the Fight for Decision Dominance · Pirates Only

• Andrew Couillard: Welcome back to Pirates Only, the Black Flag podcast from Harpoon Ventures. I'm Andrew Couillyard, and 14 months ago, I actually sat down with today's guest, Andy Markoff, for an episode with Stanford Defense Tech Underground called Tinder for Bombs. But at that time, his company was only 19 people. He had two prototype contracts and a founder who told me he refused to do public appearances, which has changed since then. And since then, the Pentagon has p kicked anthropic off its classified networks, ran an 38-day air campaign against Iran AI-generated target lists, and started Asking small companies to move faster than they've ever asked before. His company since then has raised $90 million in just five months, tripled its head count, and is now building hardware that straps to a Marines' forearm, like you're looking at in this camera right now. So I wanted to get him back on to actually unpack what has changed since. And Andy Markov is a co founder and CEO of Smack Technologies, which calls itself the first frontier AI lab built for national security. SMAC trains its own models using reinforcement learning inside of a synthetic war environment to make those military decisions. Omega for the commanders and the Planning campaigns and fires, and alpha for the small team at the edge with no comms and a tablet. Andy spent more than a decade in the Marine Corps, became a Marine Raider in Marsock and finished as a special operations officer during the Battle of Mosul, where he watched the kill chain run on whiteboards, PowerPoint, and a SharePoint server that crashed nearly every morning. After the Corps, he did business operations at Palantir, ran operations for Origin Materials, and started a sports performance company with his wife before founding SMAC in 2024. With this first boss from Marsock, Clint Alanis, and a former Tinder VP of technology, Dan Gould. Alright, so there's a lot to unpack here, Andy. But last time we talked and you had a podcast, like seriously, you only had 19 people, and you told me you didn't want to do any public appearances and you just started. Andy: Yeah. Andrew Couillard: Now you've done Reuters twice, you've done three other podcasts, and you're at 51 people. So what's changed, man? Or 50, like I think 85 people too. Andy: My marketing team defeated me and I just gave up and I just do whatever they say. yeah. That's Andrew Couillard: That's what happened, dude. They finally got through to you, huh? Yeah. Andy: my my endurance isn't what it used to be. no, we've I I think when was the last time we talked it was what like a almost a full year ago, right? Yeah, it feels like Andrew Couillard: That was fourteen months ago. Yeah, it was well over a year. Yep. Andy: it feels like like a decade ago. Andrew Couillard: Yeah. Andy: yes, I mean I guess, you know, since then we've really we obviously closed our series A. went Andrew Couillard: Yeah. Andy: on a higher end spree after that. That was when Anthropic set itself on fire. and there was, I think, you know, that that really opened up an interesting debate, which which in some ways was an opportunity for us around like, you know, that that took a couple different paths. But like one aspect of it was, you know, well, like if anthropic thinks that you know it gets to tell the government how to use its models, like, you know, are Andrew Couillard: Yeah. Andy: there companies out there that don't operate that way? So I mean that was kind of like one strand that opened up opportunities. It opened up like a a a separate and what I think is an actually far more interesting debate on just, is a large language model, large vision model, general purpose model even like the right tool? Is a monolithic model in the cloud actually gonna work in a peer level conflict? You know, and and I think, you know, our our opinion is obviously no, it won't. But I think it c the kind of like break in, hey, anthropic is no longer working with the Department of War, you know. that opened up this aperture to look at is it even the right technology anyway and or how are other people approaching the problem, which I think is a far more interesting question. And that that's really that's opened up a lot of opportunities for us really in the last year. Andrew Couillard: And how did you guys come to that question, right? Because what I first spoke to is Tinder for Bombs, helping make decisions. Obviously LLMs have dominated the rhetoric within the AI landscape, but what do you mean by d not LLMs may not be the right tool for this? And how are you approaching that differently? Andy: Yeah. So so I so I think there's I think there's a couple, there's a lot of interesting threads to this. The first one is more like, you know, type of model specific, like large language model. You have to think about what can that model learn, like that what it's trained on, like what like what can be learned in the data that's trained on. And so it can only learn things that are encoded in the semantics of text. There's a lot of really interesting stuff and a lot of useful stuff that is encoded in the semantics of semantics of text for military decision making. But but most military decision making is actually geospatial and temporal. And you're trying to understand the effects and platforms that can deliver them on both sides and like the signatures that would be coming off them if they were trying to do a certain thing in time and space over time, and looking at all these little tactical actions and how they roll up into like larger effects. And that just is a, you know, the training data set for that, I mean, A doesn't exist, but like even if it did. Isn't text. And so it's just not even the right training data set or algorithm like to really to do that. So I think I mean that that's one thread. There's another thread where, you know, when access claude, GPT-5, you know, whatever, like you're accessing it from the cloud. And if we think about a peer-level conflict, like we like cloud access is going to be limited, denied in many areas. And if you think about a peer-level conflict, like we're used to being able to talk to each other whenever we want. maybe, maybe too much so in most of my time where it would have been nice. Yeah, Andrew Couillard: Yeah, probably too much, honestly. Yeah. Andy: it'd be awesome if actually my boss couldn't have got a hold of Colleen couldn't have got a hold of me like all the time, but but Andrew Couillard: Yeah. Andy: that was not it's not the world I lived in. But I we're not gonna be able to do that. And so we're gonna have to really be decentralized. And what I would say is that, you know, most of the decisions today that are made in an operation center that has cloud access. that will have to be made before the data that those decisions are even made on even get back called to the operation center in the first place. And so this idea that, in a DDL environment, you're gonna have this monolithic cloud-based model that at the edge may helping make decisions where they're all going to need to be made is just ludicrous. and so I think that that's a second piece of it is like big monolithic cloud-based models are irrelevant in a peer-level Andrew Couillard: Yeah. Andy: conflict. And two, There are differences in the class of algorithm and type of training data that you need to learn the right things. You need to solve military decision making. And language semantics is part of it. Like image classification is part of it, but it's Andrew Couillard: Yeah. Andy: it is a s actually relatively small subset of the space, and and therefore like insufficient in and of themselves. Andrew Couillard: No, I think that's so well said. And I think for a lot of people that were former warfighters or wore uniform of some form, know the nuances of what a peer-level conflict will entail and how it's going to be communications denied, degraded environment. We're likely not going to have satellite connectivity. So how have you approached, and again, can you explain a little bit more about like the Alpha and Omega product, Alpha being more on Andy: Yeah. Andrew Couillard: the ground at the edge, Omega helping the C2, the planters itself? How have you approached that? How's that philosophy that you've lived and experienced kind of shape that? Andy: So I think when we I'm gonna frame this from, you know, when we talk about what are we building, and like we are building a campaigning tool. And and and campaigning is a part of you know, military kind of space. It really isn't talked about a lot. Andrew Couillard: Yeah. Andy: I mean, people talk about strategy. I don't know that people understand strategy, but they talk Andrew Couillard: At a high level, yeah. Andy: about strategy a lot, they talk about tactics a lot. No one talks about this middle layer where you're I mean, if you think of the roll-up as like We execute tasks. Tasks are rolled up to tactical objectives. Those are rolled up to operational objectives that link to strategic objectives. No one talks about this middle layer, but like Andrew Couillard: Yeah. Andy: it's the it's one of the areas where we fail over and over again because it's hard and we don't buy technology for it. There's not a program office for campaigning technology. We buy tactical platform technology. And as a country, we do tactics really, really, really well. We don't, we have trouble campaigning where you can understand how these tactical actions actually roll up into a lasting effect in the theater that's linked to your strategic objectives. And so we we come at it with like that's what we actually need to solve. And and one of the the fundamental problems is we are so if you even look at the war like in Iran, at a tactical level for the first couple of weeks, we just crushed it, right? But there's still obviously a conflict going on. no one is better than us tactically. it's like not even close. But the issue becomes, yeah. It's Andrew Couillard: For sure. And the Maduro rate even, you proved that at mu the most elite unit level, right? Yeah. Andy: exactly. But but the we and we don't buy technology to roll that stuff up. And I get you back when I was when I was a team commander, right? I could go out operation, we could execute it tactically perfectly. I had nearly unlimited assets. And what we would do is I would call a greedy solution, right? But something happened that we weren't expecting. And you just use more force than you need precisely to get the job done. And if you think about doing that at a theater-wide scale against China, it gets really scary really quickly how fast we run out of everything. and so these. Andrew Couillard: does man. But even before China, right? when we last spoke, Operation Epic Fury hadn't even kicked off. So what did that campaign prove and what did it really expose from a decision making standpoint? Andy: I think we we showed again how much better we are than everyone else tactically, and then we showed again how difficult it is to predict the points at which we culminate and become so low on certain types of munitions, fuel readiness, understand like the larger impacts, understand how to move things between different theaters, you know, to like optimize for like we need the extra force here now, but like what are we giving up elsewhere? Those are really hard decisions. It's not cause our like military decision makers aren't super sharp. It's just an intractably difficult problem. and so I think it just Andrew Couillard: Yeah. Andy: showed us again what the gap is in a really, really stark way at this campaign level and the need for like we can be super excellent excellent tactically in little bubbles of success that can fail at the theater level very, very quickly. And when you do that and you run out of stuff, like you can't recover up from it on a timeline to actually win. one of the examples I guess I'd say is I'm gonna I'm gonna make this like hypothetical, but it'll be like related to that like things that people have heard about and seen in that theater, right? So like imagine we plan a wave of strikes, and we you know hundreds of targets, hundreds of assets, manned and unmanned, coming together. And we go out and then a couple assets fall out for maintenance. Weather changes, which makes certain assets like less capable than that we'd originally planned. You know, the enemy does some things that we weren't planning on. Maybe like their air defense in a certain area was like better or like than we thought it was gonna be. We ended up spending more munitions. You you we the the operational planners like in an op center are having to like look at this unfold live and like on the back of a napkin figure out, okay, like how do I get back on plan? I can't compute it, nothing's giving me an answer, I'm gonna overkill it. And they just like Andrew Couillard: Yeah. Andy: spin f take this take the forces that are on strip alert, put them in the air, send them, blow through all their munitions, do it again, do it again. And and then in the course without looking at at the end of the day, you may have gone through five, six, seven times the amount of ammunition that you were planning on. And then you still don't know at that point that that actually means in two weeks, you're so short on these three SKUs of ordnance that you can no longer run strikes with this platform set against this class of targets. And now your op tempo slows down and you weren't planning for that. And because you weren't planning for it, you haven't repositioned Andrew Couillard: Right. Andy: forces from other theaters. And now you have this lull in the fight that was totally knowable based on your rates. If there was actually a decision-making tool that was both recommending more optimal decisions and letting you know the implications a couple weeks out of like, if you keep doing this, this is what it's gonna mean in a couple weeks. That's just too much for someone on a watch floor to do in their head live. And so we just make these decisions live, live, live, live. And you don't actually know the implications of it until they're staring at you in the face. And with the right decision-making tools, they should be knowable at that moment. Andrew Couillard: Now I think this is a great example for how does Alpha and Omega play to help the warfighters in this exact example you just described. Andy: Yep. And and so the way we think about Omega is Omega helps command the like commanders really so think from like the kind of echelon one four star combatant command, like down to the regiment level, Andrew Couillard: Yep. Andy: build operational objectives and tactical objectives that roll up to those and then allocate forces against them. So it's got the kind of like ins, ways, means planning and commanders intent. For for a set of operations, right? So, that's in in the way we're using it is you people overuse the word harness all the time, and it means a different thing to to everyone. So I'm what I mean by a harness is like it's like a contract of these are the command's priorities. These are the things the command wants to optimize for at this point in the conflict. These are the contracts with your adjacent units that are not breakable. Pass that as like a constraint or kind of a harness, sandbox, whatever you want to call it. To a bunch of proliferated small models at the edge who are going to deal with dynamic execution. And so they have that guidance as a harness on board. And this is at alpha. And then those edge units are sensing with their local sensor networks to understand to understand here's the red threat picture. And then they're doing, they're solving for their plan and adapting on their plan with. with a model that's helping them replan, but that model has to not break the constraints passed by Omega. And so that's how you get in a way where you can basically make theater optimized local decisions without talking is by this type concept of like passing a commander's attent and adjacent unit harness or constraint harness to the edge and then solving locally at the edge. Andrew Couillard: Got it. So the actual compute the solving is happening physically at the edge with these, let's Andy: Yeah. Andrew Couillard: call it hardware agnostic, you know, alpha enabled sensors that are out there. And why do you think that's better for more of like a peer adversary for the broader aud audience? I think I understand why. Andy: Yep. So so I think, yeah. So so s a lot of it is is starts with centralized plan repair requires backhauling like all of the data to like a central node, analyzing it and then deciding Andrew Couillard: Yeah. Yep. Andy: how to repair it and pushing it back out. That's a massive data backhaul, a massive push back out. In a digital environment, it's just like not realistic. So you you won't Andrew Couillard: Yeah, totally. Andy: be able so some of it's just physically like You won't be able to do it. And so your options are either. So your options are you will end up having to make most of those decisions in a decentralized way at the edge because you have no other choice. And so if you do what Andy did when Andy was a team commander and did the you know P for plenty rule, where it's just like add another stick of C4, you run out of C4. Andrew Couillard: Yeah. Yeah. Andy: what you need to do is make the right use of the limited munitions. resources that we have. And here's, a specific example would be let's say there's I don't know, 10 red tracks, you know, just 10 enemy like platforms. And in old world, like Andy and his team would be like, we're gonna go after all of them. But the reality may be that like the theater actually only cares about like two of them and you should leave the others as long as they're not a threat to you because there's another thing that you need to save your munitions for that's way more important. That type of guidance can be passed in this harness where I'm solving locally and now I know there's two new tracks. There's ten new tracks, but two of them I actually need to do something about here's how I'm gonna do that. And I don't need to go ask, hey, like how should I use my limited munitions that you gave me. Andrew Couillard: Right. And I wanna actually run pack like you made a comment earlier about reinforcement learning is a better use case. So why would it not be better for let's just say a local, maybe open source, LLM based model to do what Alpha Andy: Mm-hmm. Andrew Couillard: is doing? Is it because large language models inherently are stochastic, you don't get those deterministic outcomes? Is it Andy: Mm. Andrew Couillard: or a different reason of why you guys have decided like reinforcement learning is is the right path? Andy: The the I'd say I guess I'd say and maybe some of this may have changed since last time I'm talking. I I would say my my true belief is that there's an actual hybrid algorithm approach, and there are advantages to Andrew Couillard: Hybrid. Okay. Yeah. Andy: different algorithms and they all have a role to play. So I would start with that. Large language models, you know, at the edge, you may not actually be able to deploy an LLM. So like you might use a small language model just from like a compute perspective. That may be the only thing you can do. But they are it is an easier interface for humans. To interact with a language model and you know, summarizing data and kind of like helping run like an agentically orchestrated workflow, they are absolutely the right tool and we employ them in our products. The area where there are other algorithm classes that are better suited to the problem is if you think about like a lot of military decision making, it's actually more like a robotics problem, in that we're like, you know, RL of some flavor is an appropriate. Methodology. And there are different strains of RL, right? There's your old school robotics RL, there's your deep RL, we do a mixture of both. And and look, when I talk about this like harness concept, I mean, we are also using traditional like decision science algorithms like operations research. There's kind of like Andrew Couillard: Yeah. Yep. Andy: my the my bigger point is use the right algorithm for the right thing because a lot of what we're talking about is understanding. A the geospatial relationship and time relationship between things over terrain that isn't in text and it's in kind of more like scored scenario data. And so that's where like kind of your deep RL algorithms are are useful, is because like that's what they're actually trained on and that's what they learn from, and they learn to reason that way. That's why like they're particularly useful for a lot of these problems. So are language models, so are so are certain like operations research models. And the way we look at it is. This is actually about deploying a bunch of really small models that are best in class for a certain thing that get rolled Andrew Couillard: Yeah, yeah. Andy: up into a plan rather than just being like we're an LLM company or an L VM company or an RL company. Andrew Couillard: And you guys are just picking the right tool for the job, right? I think it's a good way to approach it. So Andy: Exactly. Andrew Couillard: now that you're going series B, you also incorporated this hardware component. So can you talk a little bit about how this has evolved now to be in the hardware space as well? Andy: Sure. So I if we think about Alpha, the the concept of Alpha is get the small models deployed forward where the decisions need to be made. The thesis Andrew Couillard: Yeah. Yeah. Andy: for the hardware is whatever hardware isn't being made that needs to be made to get this stuff deployed to the edge. That Andrew Couillard: Yeah. Andy: will come in two kind of like main strains, at least is how we can see it right now. One, which is what we've been working on, is the you know, the the Clint keeps I feel like calling it like the Avenger Gauntlet, but it's like the quarterback sleeve. I used to have a quarter. Andrew Couillard: The Avenger Gun, I like that. It does look like a quarterback slee. Yeah. Andy: I used I used to have yeah, it's it is. And the the embarrassing thing is I used to have one on my forearm and it literally a whiteboard that I would write with grease pencils on, like the knee pads that pilots have, but it's the forearm sleeve. Yeah. Yep. Andrew Couillard: Do we have the same thing the team leaders had it, right? When you're running drills, when you were actually doing operations and it had everything you need. It was your playbook, right? You were the quarterback. Yep. Yep. Andy: Exactly. It's like and so some some of its form factor, right? Where it's like that is I can have that on my forearm. Andrew Couillard: Yeah. Andy: I'm all the decision makers are already used to it. Like I can hold my weapon and actually look at Andrew Couillard: Yeah, yeah. Andy: it as opposed to having just go like face down in a tablet like like a teenager at a mall. And and so that gets the, you know, it's to think like model and you know, kind of cots compute the like rugged eyes can fit in a radio pouch and this like on the forearm. for local sensor autonomous orchestration off of small teams is like the initial use case for that. The other one is how do we get the model not just to an edge node, right? It's gonna we could certainly partner with other companies and we we are on like pushing the model onto edge nodes. How do Andrew Couillard: Yeah. Andy: actually get it on the drones? It's like we're not trying to fly the drone. We're not even trying to do, homogeneous swarming because like a lot of the drone providers like do that on their own. Like, how do you do heterogeneous swarming? How do you integrate a bunch of heterogeneous swarms with a bunch of manned platforms with a bunch of, you know, teams on the ground? Yep. And Andrew Couillard: Yeah, I swarm technology is one of the major priorities coming out at the at the DOW right now. Yep. Andy: and so, you know, I think, you know, next year is going to be also looking at like how do you get some type of payload that's like somewhat, you know, vendor agnostic. I don't know that purely vendor agnostic will be possible, but like we're starting to look at like how do you actually get it onboard the system? Andrew Couillard: Yeah. Makes sense. And then you're effectively the orchestration layer, right? You're the integration orchestration layer for unmanned systems. Now you're incorporating this AI native technology. Yeah, yeah. Andy: Yep. It and that's I think that that's yeah. That and I think that's as far down into autonomy as we go. Like we we are not building drones. I'm not trying to fly them, but like that orchestration layer and time Yeah. I Andrew Couillard: Yeah. Yep. you don't want to build another drone company? You don't want to be the four hundred and thirty second drone company in the United States? Yeah. Andy: yeah, I it doesn't doesn't feel like a unique advantage for us at this point. the Andrew Couillard: Yeah, i that's fair. Andy: but but you know trying and I think that you know where we are really focused is how do you take autonomy And then make it relevant to the campaign. And and and I think that's where we think we're in a unique position for the orchestration because we actually have the ability to roll that back up to the operational level objectives and do the assessment against it. Andrew Couillard: Hundred percent. Yeah. So I want to go back a little bit only because I think it was a unique catalyst for you and your fundraising when the anthropic episode happened, right? And they effectively got Andy: Yep. Andrew Couillard: thrown under the Pentagon and you had the supply chain risk, you know, rhetoric going on. What did that look like at SMAC? And how did you guys either like look at that as a warning or did you look at that as a catalyst for your business and like just walk us through like what was that week like over at SMAC? Andy: Yeah. I mean, it was it was crazy. I mean, it was actually the week we announced the series A, even though we'd like actually closed it earlier, just like it was the finally like the time we Andrew Couillard: she had no idea the series A was gonna yeah, wow. She already planned the announcement and then boom. Yeah. Andy: No, it l honestly, the the crazy thing is I had the press like briefing for the A Andrew Couillard: Yeah. Andy: the Friday that this happened. And like no one wanted at that point they like, Hey, it's like cute that you're having an A, but like what do you think about Anthropic and like how does that impact you? So I mean it like just like happened. And and yeah, and I was like Andrew Couillard: So it just drowned out. Yeah. Yeah. Andy: I was like, I guess I guess I guess I'm gonna talk about this in public and hopefully I don't say anything Andrew Couillard: Yeah. Andy: that kills our company on on national TV. So the Andrew Couillard: Yeah. Andy: So that that happened and there are kind of two. So I mean, interestingly, the media thought that it was actually a catalyst for us in some ways. Cause like there's this like idea that like, you know, Silicon Valley knows better than the military on what they need and how to build it. Like that feels a thing you sh then what's your take on that? Right. Which I mean, obviously we have a take on that where you know, I I mean to go back to it. I mean, I think our opinion on that is that, you know, we have a duty to our end users. To make Andrew Couillard: Yeah. Andy: sure that they know what the model is trained on, what it's capable of, what it's not capable of, the same way that Lockheed has a duty for, its pilots to understand the capabilities in F 35. But like Lockheed's not in the cockpit, like shutting down controls if the pilots fly it outside its parameters, right? So I think it's that's kind of Andrew Couillard: Yes. If they disagree with it. Yeah, yeah. Andy: the same thing as it's like there are people in uniform that took an oath to serve honorably and faithfully, and like that's like it's their choice. Like we need to inform them, we need to be responsible. If we don't tell them what it's capable of, that's absolutely our fault. But anyway, that that's our take on it. So that was a story that carried through the press. And then you started to have venture look at this is an opportunity, Yeah, we've got the big labs, but are the big labs gonna solve this problem? We've got an example of one getting booted out of the department of war. What if the actual opportunity in the department of war was smaller models? By people that have actually served that are domain specific. that's actually a really interesting Andrew Couillard: Yeah. Yep. Andy: story. And so the venture side started taking off. And in the Department of War, it's not like they've given up on large language models, large vision models. And they would they, you know, I don't think that they should. They play a role, but like they started looking at, okay, like what are other vendors and providers that have different risk profiles than this kind of general purpose Andrew Couillard: Right. Andy: lab that's like blown up in our face? And so it just it opened up a different interest from just a de how they de-risk their AI strategy is we're a very different risk category, right? We're we're we're newer. We certainly don't have the compute of one Andrew Couillard: Yeah. Andy: of the labs, but you know, the other difference is like we've actually all I mean, there's a disturbing number of Marines in the company. We're we were trying to like, you know, bring other services in too. But like there's a lot. Yeah, we we we do. Andrew Couillard: You need more navy guys. Come on, yeah, dude. You can't too many Marines. Andy: We do yeah, we we need more day we need we need more there are too many Marines that need to stop. Like we actually have Andrew Couillard: Yeah. Yeah, yeah. Andy: Tide pod gummies now is like how many marine I'm not even kidding. Andrew Couillard: Do you have favorite f crayon colors too? Does each marine have a favorite crayon color? You got one right now? here we go. You see it live. Andy: I have a whole this is like when when we have when we have guests over this is like what goes in the conference room and I also have t I have Andrew Couillard: Yeah. Andy: gummy Tide pods that come in like Andrew Couillard: Yeah. Andy: a box that say like don't don't eat soap and don't wash clothes with gummies but anyway I Andrew Couillard: Yeah. Alright, I need a gummy tie pod next time I'm up there. Yeah. Andy: I will you you and you will have one. The where's I go with that? But but like the profiles like you we've all we know the problem. We know the problem as well as the end user Andrew Couillard: Yeah. Yeah. Andy: because like Half the company was the end user like a year ago. And we underst and I, you know, I think there's an element of like, we're here for the mission. Like you don't have to convince us to work in defense tech, right? Like we're not doing it like because it's an interesting problem and we can live with it. Like this is like the only thing any of us ever wanted to do. Yeah. Andrew Couillard: Or because it's in vogue or hot or profitable finally. and I actually want to pull that thread because I think what's so unique about the culture you've been able to build is that the mission is in service to the warfare, the end user. You've been in their shoes, you understand the mission. I think from a more visceral stance than likely some of the executive leadership at OpenAI are anthropic, right? Not a knock on them, they just came Andy: Yeah. Andrew Couillard: from different worlds. So how have you used that to create a unique culture within SMAC that serves that mission? Because I would imagine, Andy: Yeah. Andrew Couillard: let's be honest, Marines aren't like leading AI researchers. I have a medicine one of them if you have you probably took all of them but like how are you able to take it that confluence of like really talented AI researchers in the ecosystem and the warfighters to then s be in service to that mission? Andy: Yeah, I I think so when when we think about like values, you know, we think about, you know, we're here to serve. And and I mean that, you know, for people that have been in the military, like that's like let's like in all of us. But there's kind of like two audiences for serving, right? Like I can serve national security and I can serve my team. And and for the people that were vets, both is kind of like so ingrained in us, it's just like obviously like what else would I do with my life? For a lot of the kind of scientists. some of them are really interested in national security, but like I've found like for the most part, like that archetype they're more interested in like I have like this this gift for science, for like for something, and I need I owe it to the world to make sure that I'm on the cutting edge of applying it in an area where it's gonna have an impact. And I feel like it's very rarely category specific, but it's much more like, is this. the cutting edge of my field. And so like it is a different profile and you have to have room for that. Like if you're ever if I was like screened, everyone has to have always wanted to do national security, like we would never hire the team that we need. But like what we can hold Andrew Couillard: Yeah. Yeah. Andy: is you have to be willing to do it and you're gonna be there for your team. Because with Andrew Couillard: Yeah. Andy: we're like we you're not gonna let your team down. And so that's on the service side. If they can meet me Andrew Couillard: Yeah. Andy: at like, I'm not gonna let my team down, national security doesn't have to be like your one thing that you ever wanted to do. Then we kind of go into like we're here to win. I mean, that's kind of actually like the high performing people all like want that anyway, and like that's kind of easy. And then the third, which does narrow down, I would say, the kind of pool of engineers and scientists that want to here is I say, like, we're a culture of and, which comes from for me it's like a rejection of this iron triangle of you can have it faster or like you know costless or like you know quality and but you can only pick two I'm just like the military doesn't yeah it's Andrew Couillard: Yeah. But there's always a trade off, yeah. Andy: like the military doesn't give a shit that you think that like you can only pick two. They want it like faster yesterday and better and like you just better figure it out. It it Andrew Couillard: Yeah. Yeah. Andy: it it's and and like that's like that's a lot of our Mentality's like, I want people, obviously there are trade-offs, and they'll be obvious when we have to make them. And everyone just pre-optimizes to make trade-offs wherever they've exhausted everything they can possibly do. Yeah, that yeah, it's Andrew Couillard: But the default is and the default's not like, I have to give a trade off. Yeah. Andy: and and as I say to the growth team, and and everything. They're like, are we doing this? I'm like, everything. and they so they Andrew Couillard: Yeah, yeah. Andy: they don't they'll I think they like that less than even the engineers do, but that that does kind and and so like if you meet those three. you can be on our team. And and I think it but but making room for people that are willing to do national security, but it's like not like the only thing they've ever wanted to do is important if you're trying to get best in class people in tech, because they're gonna want to move between different industries wherever the cutting edge thing is. And you just need to be cool with that. Andrew Couillard: Yeah, and speaking of it, I think you're probably one of the few companies, especially in the Elsa Gunda era, that actually has a mat and you even do a little bit of sparring with the team. Like there's such a unique crowd, like obviously the Gundo Bro crowd, they've got gyms, they've got bench presses and American flags. I think you take it one step further. So how have you chosen that deliberately in service that culture Andy: Yeah. Andrew Couillard: that you have that's so unique to the team? Andy: Well, I feel like the first thing is where I take all of my hobbies and I put them in the office so they never leave. So we have air conditioning which I don't have my house. We have yeah, like no one ever I I mean, at least I don't feel like I ever have to leave. so Andrew Couillard: so nobody ever has to leave, yeah. Yeah. Yeah, yeah, yeah. Andy: all of the things I love are here. I I think I think it is important to be grounded in what we're actually doing. And, so we are building technology for warriors, right? And and at the end of the day, there's a physical violence that happens, which we hope doesn't, right? Like no one wants war. I don't want that to happen. But there but but there's like a Andrew Couillard: Right. Yeah, deterrence is the ultimate goal for sure, but look at the world right now. Yeah. Andy: but there's like a there's like there's like, you know, a physical aspect at the end of the day, it is deterring through the application of violence. And I think it's just easy to be like, we do AI and we do AI and defense tech to do it in this vacuum of like that it's not Andrew Couillard: Yeah. Andy: about that. But I but it is. And this is A, an opportunity to kind of participate in, you know, things that like warriors do in a safe way, but like, you know, interpersonal violence controlled in a constructive way, to just kind of be grounded in like this is the culture that we are serving. I think that's one really important aspect of like we are serving warriors, yeah. I mean, like we're not like peer-level conflict doesn't look like an MMA, it doesn't look like a UFC fight, but there's just like the kind of values and self-discipline, and you know, doing hard things against a resisting opponent is all kind of like part that culture that think you know you and I experience. So this is an opportunity for the engineers. And Andrew Couillard: Yeah. Andy: the growth folks that haven't done that to experience because a lot of what I want is for the company to feel like the c if if you wanted to serve and you didn't have the opportunity to, that this should feel like the closest thing to doing that that you could have without actually being in the military. So I mean I think like that that's in Andrew Couillard: Yeah, and you can instill that warrior ethos. Yeah. Andy: and that and and the other piece is like, you know, we could we could have like nine million snacks and like, you know, beanbag chairs and scooters and all kinds of like weird stuff that I've seen like in other places. But the reality is They've got our employees get what would be like a thousand dollar gym membership. they've got the gym. I mean, a dude that's literally still cornering. He was like Anderson Silva and Leota Machito's like wrestling coach, who's, you know, also my coach, Andrew Couillard: Yeah. Andy: but like he's now the team's coach, who's like here four hours a day teaching kickboxing, wrestling, and self-defense. They've I mean, it's like they have this opportunity to get training that's usually only available to like Pro UFC athletes. And maybe that's not for everyone, but if you ever thought that'd be cool, that's just part of just being here. Andrew Couillard: Yeah. And I love that. And at the end of the day we try to get people and find pirates like you that challenge the the st status quo and the norms, right? And I think you are in a very unique way. and you've s you've hired a chief scientist, R L H F, quant researchers. So how does the team look like now and for the founders or aspiring founders in the audience, what is some advice you have as you've been growing and scaling this team? Andy: Yeah, I I think so the team today, the kind of the way we've organized has is if I look at so we have we call it mission operations, growth, product technology, internal operations. So internal operations is you know, kind of people people ops, like just like making sure that like yeah, yeah, it's Andrew Couillard: Yep. It's not grok bots yet. You get you still have people doing it. Andy: ex yes, maybe maybe one day soon. The mission operations so growth is what it sounds like, right? Like go win contracts. mission operations, execute contracts and get feedback about how well is the product working to influence tech and product and and provide that feedback back into how do we fix the thing we're building? What should we build next based on what people actually need? So it's the eyes and ears on the ground with the user is mission ops. Growth is going out and looking at the feedback from the users and going up to the program offices in Congress to be like, will you pay for this? and they're kind like Andrew Couillard: Right. Andy: and then you know, driving the contracts home. So when we talk about the tech side, product and tech, the easiest way is I think if you think about just one of the AI labs, you just take ChatGPT, right? ChatGPT is an app. GPT, like whatever version we're on, is the model. So tech builds the models, product builds the infrastructure, the application, and like the kind of deployment of those things into you know military production environments. and like a lot of What my role is is the kind of seam between, when I'm not fundraising, is the kind of seam between tech product and the feedback from mission operations about what should we build versus what can we build and how do we mediate that? So that's what what we look like today. the as far as like kind of like for you know aspiring finers. I mean, look, would say like hiring Andrew Couillard: Yeah. Andy: AI talent, if you're starting an AI company, I mean, like the market is insane. I mean, it's insane. Andrew Couillard: Yeah. Andy: And the reality is, especially in, national defense, there is you as a series A or B company are like not gonna be able to compete on cash comp with someone that's making what what Andrew Couillard: Yeah, how do you compete with an anthropic or an open AI or one of these leading model companies? Andy: It I mean, I'm I'll be honest, there's some in in some respects like we can't, right? And I think like the Andrew Couillard: Yeah. Andy: hard learning is like I can tell myself, you know, and all day long, but like I, you know, Andrew Couillard: Yeah, yeah. Andy: there are there are people that are making three to fifty million a year in cash. And so you you're ever's like, like Andrew Couillard: That's insane. Yeah. Andy: I'll give you equity, but then you have to look at like, okay, but like what is the equity maybe worth? And if they're making Andrew Couillard: Right. Yeah. Andy: 50 million a year in like real hard cash, it's hard to argue that their equity that you would give them is actually even equal to, you know what I said, So we've looked very heavily at like what are the best in class college labs for the types of things that we're building. How can we make relationships with professors there? You know, do some work with them, get access to, you know, their students who are interested in national defense that wanna come. And I mean, that's been the most productive pathway has been that, because then they're kind of not Andrew Couillard: Really. Has there been any academic institutions that are top of mind? Any that are really leaning into this? Andy: I I I think it it's interesting. Stanford has a great program. I mean, obviously Andrew Couillard: yeah. Andy: I don't think that's shocking to anyone. You you've got MIT, but I you know, it's interesting, like I cause I like all I stare at this problem all day long. we we joke that I'm like chief like LinkedIn recruiter officer. But the Carnegie Mellon is probably one of the densest talent pools for the type of work that we're doing out of like anywhere in the country. Andrew Couillard: Interesting. Yeah. Andy: and I think if you look at a lot of defense tech companies and you look at their early employees, especially in the autonomy space, a lot of them came out of Carnegie. Andrew Couillard: I knew that about robotics. I didn't know that about autonomy. That's fascinating. okay. Yeah. Yeah. Andy: Yep. It c and it's it's some of it's 'cause it's related. It's a it's a related enough skill set, but like the but yeah, Carnegie Mellon is is fantastic. Andrew Couillard: That's good to know. Yeah. So I we've even talked about he off record before that defense tech feels like it's in a bubble, right? You got forty nine billion dollars in funding raised last year, even Andy: Yeah. Andrew Couillard: Brian Schim from Andrew says it's in a bubble and they just are now raising it allegedly a ninety billion valuation, right? Which is now more than Northrop Grumman, which is crazy. So I mean, Andy: Yeah. Andrew Couillard: how are you thinking about government contractor revenue versus the way maybe a Sand Hill Road partner, maybe look an ARR and how are you thinking about making this company defensible in the long run given the fact that it may not be the Yeah. Andy: Yeah, look, I I absolutely agree that there's a bubble. there it is, Andrew Couillard: Yeah. Andy: it's a good problem for the country that we've that there are all of these defense tech companies. There's absolutely not room for everyone to win. in a venture scalable way, right? So there are Andrew Couillard: Totally. Totally. Andy: not enough venture scalable outcomes for the Andrew Couillard: Yeah. Andy: number of companies that are raising venture. Now, we could talk, is that a problem with venture math? that's a whole separate debate, but from the way venture looks at this and the returns that they're looking for, there's just not that many, there's not enough outcomes. And so the way I think about that is there's room for a couple more big primes, and you're either going to become one of those primes or you're going to get rolled up or you're going to stay small, having raised venture and kind of like not have a venture scalable outcome and probably be trying to get bought. I mean, I I think that that's like kind of the reality. Andrew Couillard: Yeah. Andy: and Andrew Couillard: Yeah. Andy: so when we think about that, it's like, well, we we believe that we're gonna become when people are like, Well you who do you compete with? Like, what do you build? It's actually not really about like the thing that we build. Lockheed builds Andrew Couillard: Yeah. Andy: a ton of different stuff, Andrew builds a ton of different stuff, but they're but people think of them as like, you know, the new primes or the old primes. Like we want to be a prime. And we think that our you know, right to win on that is just like we were literally like the company was founded and built by war fighters. And Andrew Couillard: Yeah. Andy: and not just Tactical warfare. people that like I'm I'm not one of these people. There are other people in the team that knows this that understand campaigning and have done it at the highest level. and and and so we're building a decision-making tool for all of that. But I think you we're starting to attack hardware on it. But really, what we think we can deliver is joint and coalition vertically integrated capabilities to the Department of War and to our allies and partners, and that we have a we believe our understanding of the requirement. Is our right to win there. And we earn the right to be a prime by providing value. The value we believe we're gonna provide is through the small domain-specific models that we're building and getting them deployed in this Alpha Omega concept. But over time, that we'll provide additional value as integrators because we understand the other things that need to come together. But for the government to trust us as an integrator, we've got to have hardware experience. We need to get alpha the edge. So we're building hardware. But the other piece of it is like you gotta understand hardware if you're gonna integrate it. And so that's I think the other piece of this. I I don't think you can just be a software company anymore and have a venture scalable outcome unless you're do unless you're selling commercially. So if you're if you're military and yeah, if you Andrew Couillard: Right. And you're talking defense, yeah. Totally agree. Yep. Andy: if you have if you have a commercial market too that's big, you can be software only. But if you're only gonna sell department, you know, let's call it national security, you've gotta do hardware. because the software this yeah. Andrew Couillard: Yeah. look at the brilliance of Andreil, right? They basically wrapped lattice in hardware and then they sold the hardware. So you can bake in software Andy: Yep. Yep. Andrew Couillard: prod profit margins into that product. So I agree. That's crazy. I I mean in in the last year, do you think the Andy: And I and I think and I think that's the that's the thing we've thought about. Andrew Couillard: yeah, do you think the government's gotten better at buying things? Like there's been so many memos that have come out. There's so much rhetoric about how we need to prioritize speed and volume and you have the highest levels of leadership that are saying it. Are you guys feeling it at the company level? Like are we as a country as a department of war buying things quicker? Andy: I I think I think that it is definitely getting better. everyone there's Andrew Couillard: Yeah. Andy: all kinds of stuff that still drives me crazy, but Andrew Couillard: Yeah. Andy: the OT pathways, like having CDO, DIU, these kind of like bodies that are trying to like help like bridge from like prototype to production. Is it perfect? Like absolutely absolutely not. Like not even close. Is it better? Yes. But if I were to say, what are this what we're trending Andrew Couillard: Yeah. But it's better. Yeah. So we're trending the right direction. Yeah. Andy: in the right direction? The the biggest gaps are are Our system can't buy joint and our system doesn't buy technology beyond tactical platforms. Like those are the real gaps. And like here's the thing. Like I mean, I always say about like campaigning. Like there is no business incentive for us to build a campaigning tool because they no one will no one buys those. So we know. I mean, when I take my like business hat off and just put like mission hat on, or like, I don't care, I'm just like here to win like like the actual. Andrew Couillard: Yeah, yeah, yeah. Andy: I don't know, life mission is we know we need this, we're gonna build it, but we've got to sell it through tactical service-based capability programs. Because the users know they want campaigning, but like the program offices don't buy that way. And so like there's just this, there's that gap. There's a certain like there's lack of joint vehicles. It's getting better. You know, CDO is an example of a joint vehicle, like kind of like, you know, Golden Dome initiative is like a joint. capability. So maybe we're we're trending in the right way. There's a lot of work to do, but I feel like overall I feel optimistic. Andrew Couillard: But if you're a king for a day and you can make one policy change, one change at the Department of War level, what would you change? Andy: Joint joint we need more joint program offices. Andrew Couillard: joint requirements, joint program offices, buy things for the joint force. that's interesting. Yeah. I wonder if they could make a c an Andy: Yeah. Yeah. I mean the the the idea, I mean Andrew Couillard: acquisitions billet that is joint and you have to take over some sort of joint program, right? Andy: that that's it, that's it's this it's one of the single biggest problems in acquisitions. It's and it's the reason Jad C2 is a disaster, is because it's purchased via services that don't all agree on the spec and the interface. And like there's just that it need like we need joint vehicles. I mean, so like look, we are on the verge of being deployed across all six services. We've had to sell into every single service, Andrew Couillard: That's incredible. Wow. Congrats. Andy: and it's the same thing. Like it would be easier to just have it. in a vehicle that all the different services could buy. Yeah, we can have a slight navy flavored workflow and Marine Corps flavored workflow and and stuff like that in there. But yeah. Andrew Couillard: Course. Yeah, there'd be tuning and training like you would for each customer, but you're right, yeah. But the enterprise you would sell into is the joint enterprise, right? Andy: Exactly. Andrew Couillard: Yeah. I think that's a unique way to put it. So Andy, you're someone that can chew glass better than most. What glass are you chewing right now? What are some of those problems that those things that you're just grinding through? Andy: A AI researcher hiring. That's a never it's a never ending. It's Andrew Couillard: yeah, it's still you're still chewing that glass. man. The war for talent's real. Yeah. Yeah. Andy: a new I we we've we've we've had some big wins on that that we'll we'll announce in the in the next couple months. But that is I I think look, I I everyone always was like hiring is like the number one like executive skill, or at least that's what like my you know mentor palantur always told me. I was like, yeah, yeah, yeah, whatever. But I mean I honestly think it is and d honestly it's kind of embarrassing Andrew Couillard: Yeah, Andy: how much I feel like I suck at it. And so I think that, you know, we have landed some near-term, let's call it like tactical recruiting wins. A lot of my, you know, anxiety and stress is how do we make that repeatable? so how do we get like good top of funnel inroads for recruiting for that pool that I like? We get a new contract, we raise another round, I need 10 more of them. It can't take a year Andrew Couillard: Yeah. Andy: for me to find them. I want to look, look myself in the mirror and say, like, we got high odds of hiring in two months. It's like a lot of my work is around doing that work now because it's a long game. and and so I think, you know, that's that's one of them. And then look, I mean, the the other piece of it, which I feel like is like a tale as old as time in the startup world, is just we went from like, you know, a team of like 20 to now a team of 75. We certainly didn't have all the internal processes and infrastructure and and we're we're in that like Andrew Couillard: Yeah. Andy: weird space where like we're still a really early stage startup. But there's enough stuff going on that we need more process. And we're like doing the yo yo game Andrew Couillard: Yeah. Yeah. Andy: of like over correcting in every single direction and like trying to deal with like ha like that and like onboarding people at a reasonable rate. And I mean we're we're struggling with it just like everyone else does. I think that's like the other thing. And you know, the third is just I feel like Andrew Couillard: What yeah. Andy: I never stop fundraising. It's like always it's like my only job. Andrew Couillard: Always hustling, always selling, right? Yeah, I love that. Yeah. Andy: I t I joined the Marine Corps to like to do to play with explosives and shoot guns and they gave me a radio and they're like sit on your hands and maybe don't even use it, but here's your toy and go in the corner and then you start a tech company, Andrew Couillard: Yeah, yeah. Andy: you'd be like, I'm gonna make like AI and then like all I do is raise money. Andrew Couillard: Raise money, hire people, chew glass. Yeah. Andy: Yeah. Andrew Couillard: Yeah, yeah. Depending on the day, depends on the order. So all right, what's some final advice you'd have for some of those? Because again, a lot of our listeners are either founders, aspiring founders, hopefully thinking of applying to Black Flag. What's some parting wisdom you have for these founders, especially those that want to be in this defense tech ecosystem? It's a bubble, it's getting crowded. What would you say? Andy: Yeah, I think hm. I think for for me a lot. I mean, like everything else, I think, you know, you start you to start a company, you're gonna have some experience that gives you the right to do that. And that could be like deep technical experience, it could be deep domain experience, it could be fundraising Andrew Couillard: Yeah. Yep. Andy: experience. And you're gonna have you're gonna have some experience that gives you that anchor to do it. and there's gonna be a lot of things you know nothing about that you're gonna have to learn really, really quickly. And I think there a lot of people want to try and figure out, okay, how do I do fundraising? How do I do this? learn it, get it all lined up and set up, and then and then go try and execute. I just feel like the reality is it's a lot more chaotic and you're gonna end up you're gonna Andrew Couillard: Yeah. Andy: you need to know what your gaps are. You need to learn as you go and just any building a product, just quickly iterate and get better. fundraising is a perfect example for me. I didn't know anything about it. And no one gave me any advice on how to do it before we started doing it. And I was terrible Andrew Couillard: Yeah. Andy: at it. And no, what I was terrible at was telling our story. horrible. And every time I would just get better and I'd mess it up again. And everyone tell what I did wrong and I get better. But I think it's just not like learning the same mistakes twice, learn it once and then fix it, but just speed of iteration to like make a mistake in the area that you don't know and learn from it. But you just kind of have to move quickly, especially because how crowded it is. Like speed is like the most important thing. I think I have a weird one for defense tech that I think is probably like somewhat controversial, but it goes back to my and and everything. I think like traditional NBA wisdom would be get really, really good at one thing, do that really, really well, and then branch out. I think if you do that in defense tech, you die. Or you're a non-venture scalable outcome that gets acquired. And if that's what you want, that's fine. But the reality, Andrew Couillard: Yeah. Andy: you have so little control. Over certain aspects of the contracting process. And it and you need to have a core set of things that you can move to different stuff. And I'm not telling you that there's a balance. You can't do everything actually do everything. But I've seen a lot of companies get really, really narrow to try and do that well. And they just can't pull the revenue in because they missed the two or three opportunities on that particular thing and they weren't prepped to go do the other stuff. And like that, Andrew Couillard: Yeah. Andy: I think that's it's like if you're only if you only have one customer. And you take the contract opportunities in a given year down to you have three or four shots on goal, you're gonna lose. and I I think that's a really Andrew Couillard: Yeah. I totally agree. Andy: hard thing about this space that I I think is probably maybe like different than the commercial space because it's just because of the size of the market or the fact that you have like Andrew Couillard: Yeah. Andy: one customer. Andrew Couillard: So true. Well Andy, thank you so much for your time. We're grateful to have pirates like you Andy: Yeah. Andrew Couillard: again, right? And I think the amount of progress you've made at Smack Technology is absolutely incredible. For those that are looking for jobs, you could always where could they go and find maybe some open roles or positions to go work for an amazing guy like you. Andy: Yeah, we I I believe we've got posted on smacktechnologies.com, but honestly, like ping me on LinkedIn. I'm Andrew Couillard: Awesome. You got it. You heard it here first. Well, Andy, thank you so much for your time. Thank you for joining Pirates Only. We'll talk soon, man. Andy: Hey, thanks for having me.