Joel Anderson is the founder and CEO of ZeroMark, a defense tech startup that used AI and robotics to build a “gun that never misses” to protect warfighters from a growing drone threat.


ZeroMark is a defense tech startup. We are backed by Andreessen Horowitz, a number of angels, and some smaller funds. Our mission is pretty simple: We produce technology that gives soldiers and frontline defenders a tactical edge or ultimate supremacy in any kind of combat or threat situation.


Today we’re focused on an interdiction platform. What I mean by that is we fuse sensors to interpret scenes and understand threats on a battlefield or in general defense situations like protecting public infrastructure. One of the big use cases for our platform is drones. The ways that drones are being used maliciously is skyrocketing and the problem is that there is no real good way of stopping them. Any time a company has come up with a method like jamming remote frequencies to the operator new autonomous drones come out and those countermeasures become ineffective. There are, of course, tried and true methods of air defense. You can absolutely detonate a drone with a missile. Like a missile used to recently shoot down Houthi drones that came at our ships in the Red Sea cost $2 million, but those drones cost them $2,000 to make. The economics are not on our side in that situation.

deep tech newsletter
A weekly dispatch featuring exclusive interviews with deep tech founders & a roundup of the most important deep tech news.


So we’re building a platform that allows you to interpret and target drones for kinetic interdiction with cheap bullets and small arms. We build what’s called a fire control system that helps you aim that device in an automated fashion. So every time you pull a trigger you hit those drones. The aiming piece is the cool part, but the real value comes from not just that but the context that we provide in addition to the interdiction platform. So when we’re looking at drones in the sky, we can do things like say this drone is not carrying payload, it’s flying around this building. It’s probably surveilling, maybe don’t intercept it. We can also coordinate data with known assets. We have what are called battlefield management systems that say this is our drone, don’t shoot it. We can also do things like say, hey, this drone is carrying payload and is in an attack pattern, but if you shoot it down that grenade payload is going to fall on an airplane or infrastructure or friendly troops. From a few hundred yards away we can show that this drone isn’t flying at you, it doesn’t seem to be a threat, but it’s actually flying at a vector that’s going to intercept with an armored convoy 10 miles away so intercept immediately and inform those people. Those are the kinds of features we put on top of auto-aiming.


There are two required pieces you have to mount onto a gun. The first looks like a little rectangle with little camera sensors on it that is about the size of 3 iPhones stacked on top of each other. On a classic M16 or AR15 type weapon platform, it’s going to be mounted on the rail where your handguard surrounds the barrel. It’s about the same size and weight as existing accessories that attach in the same spots like infrared lasers or flashlights. The second is where the computer, sensors, and motorized component that moves the point of aim of the gun exist. There are two ways to accomplish that: the most common method is a replacement stock—the part that makes contact shoulder when you’re firing the weapon. To replace one you just take your hands and pull a tab, slide the old one off, put this one on and voila: plug it in and you’re good to go. You can also have a foregrip that mounts under the handguard on the barrel where your supporting hand is usually holding the firearm that is motorized like a gimbal. Each configuration uses a different mechanical approach, but when you’re aiming at a target it is doing the hard part of leading the shot by making subtle angular adjustments so that every time you pull the trigger the shot hits the effective kill area on the drone, which is really difficult to do without any assistance.


It’s hard to compute where you need to actually aim to hit one, and that’s largely because of the way people train. We shoot at moving targets all the time on the ground, but you’re used to shooting a few degrees above ground level. When you’re shooting up into the air, it’s a very different calculation. You can’t just point the reticle of your sight on a drone. The bullet will not hit the drone because the bullet has an arc in the way that it flies, there are environmental factors, and the drone is moving as well. For example, if there’s a drone 200 yards away flying 100 miles an hour and you point your gun at it and shoot, by the time the bullet reaches that drone the it will be 72 feet past the aiming spot in ideal conditions. So you have to aim 72 feet ahead of that drone and predict the flight path in that particular scenario. Computers are very good at doing that, but people suck at it.


Yes—there’s no cloud computing or anything like that. We certainly run large models and train datasets and single shot detectors and all those fun components on larger hardware. But it’s all done on the device. It can run without any connectivity on total standalone.


We have it in the hands of end users that are in a variety of special mission units in a couple different countries. We obviously care a lot about the US DOD, but it’s a lengthy process. We have clientele within the US that we’re in evaluation stages with and we’re a little more mature in terms of international go-to-market. That’s what I can say at this point.


The reality is that there are no real alternatives to this other than spending $2M per drone that you disable. And that’s just not tenable. The standard current method is to stand and unleash a wall of indiscriminate lead. That’s exactly what the Ukrainians and Russians do every day. The problem is it’s ineffective. They never hit.


The realization that led to ZeroMark came one day when I was driving a Tesla with Full Self-Driving Beta to a shooting range. I had this epiphany that the technology at this car’s disposal could solve more than just self-driving on city streets in New Jersey. It struck me as insane that I’d never seen anything even remotely close to that even on large platforms in the military. Dismounted soldiers especially have nothing of that level of sophistication. So I had this realization and then I was immediately in a shooting range where I was reminded of the fundamentals and difficulty of marksmanship. So I put two and two together and asked if we have small camera gimbals and high compute power in small silicon packages, why is there not something that can help with the hardest part of firefighting and help take skill out of the equation?


Shooting in general is a really difficult thing to do in combat environments. Take a qualified marksman in the US Army who had to hit 36 out of 40 targets in varying ranges of positions on the range to get expert qualified. As soon as you take that same expert marksman and give them a combat simulatingscenario –say where they do 100 jumping jacks, sprint 200 yards, and then shoot, they’ll drop to 15 out of the 40 targets. Now put them in real combat when you hear bullets cracking overhead, there are detonations going on, you have like all sorts of shock and trauma, you’re tired from 72 hours of patrolling with only a handful of hours of sleep….then the hit rate is an order of magnitude worse than the simulated scenario. Try to shoot a gun accurately while you’re moving, taking in all this information, scared for your life, and your body is trembling—it’s just not a thing that people can do well, unless they are exceptionally well-trained. And very few people are. There’s a big difference between those that shoot 10,000 rounds a year and those that don’t. And, believe it or not, most military outside of special operations do not shoot more than 10,000 rounds a year. Nor do any police for that matter. So I wanted to solve that. I wanted to solve the problem of accuracy because I felt that that would give our infantry and dismounted soldiers an advantage in actual firefights and urban combat for the first time. So that’s what we set out to build.

We do have other systems that give us tactical edge and advantage in peer warfare of course. But in peer warfare in urban combat, a $250 AK47 is just as effective as an M16 or whatever other system is being deployed by the US or our allies. It’s our air support systems and other big systems that give us advantages in those situations. But left to your own peer-to-peer, it’s an even playing field.

But now drones are changing that. Today, a huge portion of injuries on the frontline in Russia are from FPV drones. It makes sense. They’re fighting in trenches where artillery doesn’t matter that much. There’s certainly no one charging up the trenches on a common basis. But drones are quite effective at trench fighting and they’re very cheap. So their use has spiraled out of control without effective countermeasures. So that’s what we set out to solve with a capability that otherwise wouldn’t exist. Our only alternative solutions are either just cost prohibitive or impractical like carrying a jammer, which is another 20 pounds and doesn’t work every time. There are towed trailer systems that can microwave hundreds of drones at the same time, but you’re never going to use that in an offensive operation because it’s limited by the trailer. But by modifying an individual weapon that every soldier will have at all times is a much more potent and broadly applicable approach.


I think that at a philosophical level, humans have a tendency to invent and create methods to have power. It’s their way of safety. The sad part of it that I don’t agree with or that I think should be consciously fought against is just the never-ending knack for people to build systems that are good at hurting other people. I think that’s terrible. I’m not building a platform that is meant for offensive lethality. Sure, there can be applications where that is the case and there are situations where violence is necessary. But that’s not the intent here. We’re trying to build a lifesaving tool. This technology gives us a chance to effectively fight autonomous machines because increasingly that will be what people engage with. Wars won’t be robots fighting each other like we see in the movies. It’s humans fighting robotic adversaries. We see that with drones already. I’m trying to build a platform and a tool that helps people effectively defend against that.

People hear “a gun that never misses” and that can sound scary. It’s scary in the context of, say, a police officer using unlawful force or a hypothetical situation I hear a lot which is ‘what happens if a mass shooter gets their hands on a system like this?’ The short answer is they won’t. This is military only. We have very strict activations for the device. It doesn’t just work indefinitely. If it was stolen from a military base, it won’t work anymore. It has to be activated.

The scariest thing is not for this technology to exist. The scary thing would not be putting it to good use to defend against lethal autonomy.


One of the most horrific short sci-fi films ever made was Slaughter Bots, which is about autonomous drones that can target and kill specific populations. It’s terrifying. We’re building a tool that can help a person defend against something like that. It’s a real threat vector and it’s very scary. But most people would rather put their heads in the sand than actually think about it.

For context, the military is only a third of the market for counter-UAV technology. There’s a ton of money that pours into counter-UAV for different purposes. Protecting stadiums is one example and golf courses, believe it or not, are big spenders too. But in any case it’s all related to public safety and defense in some way, shape, or form. While we haven’t had a big incident today, it’s inevitable. Malicious drone use will affect civilians in the future and already has in some ways. We haven’t had something like a school shooting that captures the minds of all of America, but 2,000 drones cross the border in Texas each month smuggling things like fentanyl that kills so many Americans each year. There are drones smuggling cell phones and drugs into prisons every day. These drones are popping up in so many different sectors and areas that you wouldn’t expect. People are very vulnerable to drone threats right now and they don’t realize it.


In our case we have both software and hardware, and hardware companies are hard mode for startups in general. The defense component makes it even harder. The reason that it’s harder is because it is a complicated procurement or acquisition environment. Selling is very different than it is for enterprise software or consumer products and things like that. It takes a lot of money to be able to navigate that because it’s slow. So it requires extreme capital efficiency and discipline.

There’s this term “the valley of death” in the defense startup world that touches on this. Most companies that can operate effectively will get a couple million dollars in investment or cooperative research agreements or other transactions like that. But then between that and a $100 million purchase contract for indefinite quantities there’s nothing in between. That’s where the startups die, that’s the valley of death.


Defense is coming back though and what’s changed is that every VC under the sun is getting into defense right now, even if they don’t know anything about it. I think that’s created an incredible opportunity for startups right now because of the international context. That’s interesting because generally defense is like a nationalist topic. But with NATO and the conflicts that are existing in the world today there’s a lot of opportunity to bridge the gap and change the way the DoD operates, which currently incentivizes primes and these “large exquisite systems.”

I know for a fact that if left to our own we could have a better system than any prime out there. But if we’re not successful in areas like the Eastern Front of Europe and showing that it can be effective, the DoD would never buy something like ZeroMark. They’d rather spend the $2 million on missiles or big systems or million dollar trucks to defend against drones just because they’re incentivized to. That’s just the way the sector is built in terms of purchasing officers and things like that. So the international opportunity is very real and meaningful.