Inside the Trusted Disruptor Strategy with CTO Andrew Puryear
Transcript
Maggie 00:04
In this episode of the Mission Matters podcast, we're joined by Andrew Puryear, the Chief Technology Officer, CTO of L3Harris, formed six years ago from the merger of L3 Technologies and Harris Corporation. L3Harris is the sixth-largest U.S. defense contractor. It builds a wide array of defense products, including tactical radios, night vision systems, electronic warfare systems, rocket motors, and much more. L3Harris positions itself as the defense industry's trusted disruptor, combining the reliability and mission understanding of a long-standing defense prime with the speed and innovation of a challenger. The strategy includes partnering closely with startups and emerging technology companies to accelerate capability delivery, break traditional acquisition cycles, and move new technologies into the hands of warfighters faster than the conventional defense industrial model. As part of this strategy, L3Harris established a strategic partnership with Shield Capital to gain access to disruptive innovators for technology transfer, teaming arrangements, direct investments, and partnered contracts. Since establishing this partnership, L3Harris has partnered with dozens of startups, including Shield Capital portfolio companies like SeaSats, Overwatch Imaging, and Code Metal AI. Andrew joined L3Harris three years ago and has been instrumental in driving the company's trusted disruptor strategy and expanding its partnerships with emerging technology startups. Andrew has had a long career in the national security technology space, having worked as a technical leader at the Stanford Linear Accelerator Center, Sandia National Labs, MIT Lincoln Labs, MITRE, and Ultra Electronics, as well as serving as an Engineering Duty Officer in the Naval Reserves for more than 14 years. Andrew is one of the sharpest thinkers in the defense technology ecosystem, and I have personally learned a tremendous amount from my conversations with him over the past few years about where the industry and technology are headed. In this episode, we cover everything from the future of technologies like electronic warfare, attritable systems, and generative AI in the national security space to how startups should think about partnering with defense primes and much more. Now on to the conversation. All right,
Akhil 02:46
Andrew, thanks so much for joining us today. Super excited to dig in. It's been awesome to get the chance to work with you over the past years here at Shield Capital. Let me start right now. You're the CTO managing an amazing portfolio across technologies, but you started thinking on Vassar Street doing your PhD in electrical engineering at MIT, and now here you are leading initiatives across a variety of technology domains, from the traditional communications realm to AI to the platforms and payloads and sensors that make it happen. Let me ask you, when you're not doing your leadership work and you do have a chance to get behind code or get on a chipboard, what gets you excited, and what do you still want to do technically that maybe you don't have as much time for these days?
Andrew 03:33
Yeah, that's a great question. So back to my time in the Stata Center there at MIT, I always loved that MIT was an integral part of inventing radar, right? And given those roots—those deep roots that MIT has in radar—and today's current trends in artificial intelligence and machine learning, in my off time I build models. I fine-tune models that do everything from, hey, this is how you classify RF, to generating novel waveforms, whether that’s radar waveforms or comms waveforms. It’s just super exciting to apply something that was so important historically—a decisive factor in the winning of World War II—and apply modern technology to that space.
Maggie 04:16
Andrew, are you a vibe coder at all? Are you using Cursor or anything like that?
Andrew 04:21
Yeah, and I love the trend of vibe coding because it really does help accelerate workflows. I actually tell every engineer at L3Harris that they should be looking for how to use AI-enabled tools such as Code Metal, which is a Shield Capital portfolio company. How do you use a tool like Code Metal to accelerate your workflow so that we can be more productive as engineers? What we all want to do as engineers is deliver capabilities into the hands of the warfighter faster, right? And Code Metal allows us to do that. The vibe-coding ecosystem allows engineers and non-engineers alike to start to build code. In terms of whether vibe coding is high quality or not, you see various studies that have somewhat mixed reviews. But one of the things I've noticed in my career as I monitor AI is this: every time someone says, “AI does X, but it cannot do Y”—okay, AI can beat the world chess champion today, but it'll never be able to do complex reasoning. A couple of years later, it's able to do complex reasoning, but it'll never be able to pass an eighth-grade science exam. A couple of years later, it passes an eighth-grade science exam. What I've noticed is that the time constant between “it can do this” and “it can’t do that” is shrinking over time. And so I'm very bullish on AI and its ability to transform and have a profound impact on the way that companies in the United States operate, the way that we fight wars, and the way that we live our lives.
Maggie 05:49
Andrew, one of the terms that we hear a lot, both in talking with folks in primes, in the government, and at startups, is the term “software defined warfare.” What does that term mean to you, and what role do you see for startups and primes within that future vision?
Andrew 06:08
Yeah, okay, so this is an important one. One of the United States’ enduring asymmetric advantages is that we can innovate faster than anyone else in the world. Full stop. We innovate faster, and that's incredible for our ability to maintain an asymmetric advantage. So what does innovating faster look like in the midst of the Fourth Industrial Revolution? It means deploying operational capabilities in microseconds, not months. We’ve got to go from months historically, to minutes, to microseconds. Software defined warfare, to me, is all about that: how do we deploy new capabilities into the hands of the warfighter in microseconds?
A couple of real-life examples: the kill web has to turn into code. We’ve got a sea of sensors out there looking for nefarious objects, and then a tremendous number of effectors—kinetic and non-kinetic. The decision logic to connect those together has to optimize for a tremendous number of things. If you have a large number of heterogeneous incoming threats—hypersonic, class-one UAV, cruise missile, and in cases we've seen recently, all of them simultaneously—a modern, mature kill web will sense the incoming threats and then optimize effectors for those heterogeneous forms so you maximize your probability of kill at the lowest possible cost. The cost curve is important.
To be able to do that and adapt, as soon as we deploy a kill web it will be obsolete, because the adversary will immediately start to adapt. We have to be able to do over-the-air upgrades. Radars, electronic attack modules—these sorts of things need to be updated with new techniques and new waveforms on the fly so we can act and react faster than any adversary.
I want to give one real-world example of software defined warfare. I’m going to pull from an exercise that was recently run: Talisman Sabre, a joint exercise between the US and Australia and INDOPACOM. One of the challenges they were trying to demonstrate at Talisman Sabre—and an area where we've been doing a lot of work with a product called DISCO—is this: currently, from the time an adversary deploys a new RF threat, like a radar we've never seen before or a new jammer that’s trying to jam our communications, to the time we detect that novel waveform, classify it, and then create a new electronic attack technique to counter it, takes months. That is far too slow.
With DISCO, if the adversary comes up with a new jammer, we’re able to detect, classify, and then create a new electronic attack technique or waveform on the fly, deploy it back to the edge platform in minutes, and we're targeting microseconds. Something that currently takes the US and our allies months, we’re trying to compress to minutes and ultimately microseconds in the electromagnetic domain.
Akhil 09:38
That's great, Andrew, and it’s something we did in the past during the Cold War. It was the blind man’s bluff across all domains. But to your point, it’s just getting faster, and the iterative process is only accelerating. I had two questions on what you just mentioned, which I thought were great. The first is this: you see AI and novel technologies—not even novel anymore, they’re here—in the commercial space and being applied in the perception space from a computer vision standpoint or in the multimodal space. But unlike startups, you’ve seen how this needs to get deployed at scale. What are startups missing when it comes to getting some of these software-defined applications ready for production scale with thousands of camera balls or potentially hundreds of thousands of sensors, which is the ecosystem we should have?
Andrew 10:29
Yeah, it's a great question. One of the things that we focus on a lot at L3Harris is what I'll call missionizing capabilities. So a startup will come with a capability, like a SeaSats UAV. It's a small UAV about the size of a surfboard. How do you take that from “this is a really capable platform” and turn it into something that's useful to the warfighter? You have to have resilient communications, and you have to have ISR assets, meaning intelligence, surveillance, and reconnaissance. That ties back to the EO/IR balls we talked about earlier, or SIGINT detectors and similar systems.
One of L3Harris’s sweet spots, and one of the reasons we're so excited about working with Shield Capital portfolio companies, is our ability to work with a startup to operationalize and missionize their capabilities and get them into the hands of the warfighter. That applies to hardware, such as SeaSats or APEX, which we're also working with, and to software as well. Overwatch Imaging is a good example. The DoD has very specific rules of engagement for authority to operate and related requirements. Thinking through the pipeline from having an eye-watering, world-class capability, through the ATO process, all while having the budget to do so, is critical.
Maggie 11:56
Andrew, I wanted to ask you about Ukraine in particular. What are some of the lessons we should be learning from the conflict in Ukraine, and what are some lessons we shouldn't be over-indexing on from this conflict?
Andrew 12:12
Oh, that's awesome. Do learn and don't learn. Let's do that. And I think that's important, because sometimes we overfitted from the Gulf Wars, and we should not remake those mistakes.
So what have we learned, or relearned, from the war in Ukraine and the conflicts in the Middle East? First, mass and cheap autonomy can win. For many years, the focus was on exquisite, low-quantity precision missiles and precision strike. Now we've seen first-person-view drones and Shahed-class systems that, simply because of the cost curve, present a formidable challenge. Affordable mass with autonomy is certainly important.
Electronic warfare, or EW, is decisive in modern warfare. In layman’s terms, it's the ability to jam, spoof, and survive in a contested spectrum. Systems are being jammed into the Stone Age. One way Ukraine has overcome that with FPV drones is by trailing a fiber because they cannot communicate in these contested environments. With the right communications and resilient approaches, it is possible to burn through the jamming the Russians are putting out, but at what cost? FPV drones are cheap. A high-end radio to defeat that jamming is not.
Layered, integrated missile defense works, but it must be affordable. We’ve seen this repeatedly, whether in Iran’s attacks on Israel or heterogeneous strikes involving drones and various missile types. Integrated air and missile defense absolutely works, and the United States and our allies are learning how to optimize these systems for current threats. But again, cost matters. A multimillion-dollar Standard Missile-3 or Standard Missile-6 for a Shahed drone is not sustainable. As a nation, we have to learn to mix effects in an optimal way: high-power microwave, laser, free-space optical laser, and other options combined with exquisite missile defense. The goal is to protect critical assets with the highest assurance possible but at a survivable cost.
Next, maritime drones have changed the naval calculus off the Crimea peninsula. These low-cost USVs, essentially missionized jet skis or Sea-Doos, forced Russian ships to redeploy away from Crimea and took many out of the fight. Some were destroyed, including the flagship, and others were pushed far out. At the same time, we do not want to over-learn this lesson. What works in a confined area like the Black Sea does not necessarily translate to the Pacific or Indo-Pacific theater. In some domains, maritime drones are critical. In a blue-water fight, you have a different operational picture.
Quickly, what do we not want to learn? I've heard rumblings that tanks and airpower are obsolete. I do not believe that. You still have to be able to deliver mass and lead on target. Tanks are not dead, but they will die without electronic warfare cover, deception, and combined arms integrated into a sensor-saturated fight. If you deploy a tank by itself, it will die. If you provide EW and the right ecosystem around it, it's still part of modern land warfare.
We also should not conclude that exquisite systems are unnecessary. We have learned that over-reliance on exquisite systems is brittle against cheap mass and EW, but that does not mean abandoning exquisite platforms. We need a balanced portfolio of high-end exquisite weapons and lower-end, less capable but highly cost-effective weapons.
The key to all of this, and something we've talked about already, is intelligence and C2. Those are what allow you to combine effects in a way that keeps things safe at the lowest possible cost.
Akhil 17:01
You talk a lot about some great lessons there, both the ones that we should emphasize and the ones we should not. There’s a lot of learning lessons, and then there’s acting upon those lessons and actually being able to implement them at scale. To you, what are some really good success stories that you hold in high regard from past history where either the US — or maybe it wasn’t the US — was able to quickly internalize a lesson and then field what was needed from a technology standpoint, integrated with what the users and operators needed?
Andrew 17:31
Yeah, it’s a good question. And, you know, we are on a modern podcast here, but I want to go back. The United States has been doing this for a long time. Back in the early 2000s, the United States was fighting wars in the Middle East, and one of the biggest issues was IEDs — improvised explosive devices — which were having an outsized toll on American and Allied soldiers and sailors. I know the US Army had a huge effort to go out there and understand: what is this threat doing, and how is it having such an impact on our soldiers in the field?
What they did was deploy the MRAP, the Mine Resistant Ambush Protected system. It’s basically like a boat, so when one of these IEDs blows up, it vectors the force of the explosion off the side of the hull. Super interesting how they did it, and it was probably one of the fastest acquisitions of a major platform the US has done in decades. The point is that there are now all these new RCOs — rapid capability offices — and DIU doing excellent work, but the United States has been doing this for a long time. When sailors or soldiers’ lives are in harm’s way, we absolutely figure out how to get it done.
I do see OTAs, Other Transactional Authorities, as a great way to move fast. While the authority to use OTAs has been codified in law for a while, the department is expanding their use specifically to buy commercial platforms. This is one of the places where L3Harris really likes to partner. The Defense Department says it wants to buy UAVs, commercial UAVs, commercial drones, these commercial things — but there still has to be someone who really knows how to take that commercial thing and adapt it. In fact, Joby is a great example. They make these commercial flying cars.
Joby is a super innovative company. They build commercial flying cars, and when we talked to Joby initially, we said, “Hey, this is an awesome platform that can bring a lot of capability to the warfighter. We would love to work with you, Joby, to figure out how we put resilient comms in here so that as soon as the adversary turns on the jammers, we’re still able to establish command and control. How do we put ISR assets on there — intelligence, surveillance, and reconnaissance? How do we put things like EW payloads — electronic warfare — on there?” That’s where a partnership between companies like Joby, or more broadly other commercial capabilities, can work with L3Harris through an OTA to get really eye-watering capabilities into the hands of the warfighter more quickly.
Akhil 20:27
Thanks, Andrew. I will say, on the MRAP note and referencing Secretary Gates’ book, it did sort of take the Secretary to emphasize and prioritize initiatives like that. Hopefully things have changed. I’d be curious about your perspective: have you seen structurally, sitting where you are now, that the demand signal — not just from the operator, because you’re staying in touch with what is happening there, seeing how your technology is being used and where the limits are — and then the demand signal from the voices that matter, has it improved and evolved in the way that you would like in the last 20 years?
Andrew 21:06
Yeah, absolutely. It is interesting. At the end of the Cold War, the United States and our allies had a peace dividend, and we built an acquisition system over the years that was really focused on delivering a peacetime military. What’s happened is the increasingly assertive China and a major land war in Europe that, quite frankly, a few years ago was unthinkable. I think it has forced us all to take a step back and look at whether the acquisition system, as it’s structured today, is what we need — not just because the threat is changing, but because the technology is changing so incredibly quickly.
People ask whether this is the fourth industrial revolution or just another technology wave. Why is it actually an industrial revolution? To me, it’s the pace of change that drives why it is having such a profound impact on the United States and our allies.
Maggie 22:01
Andrew, speaking of some of these acquisition trends, one of the big buzzwords we hear with respect to acquisitions and the DoD budget is Golden Dome. I know that L3Harris is intimately working in the sensor space and the space domain. I would love to get your take on what exactly Golden Dome is, what it means in the first place, and where you are seeing opportunities for startups developing new technologies.
Andrew 22:35
Yeah, absolutely. Golden Dome is super exciting. As a technologist, it is super exciting, and I think it is also super important from a national homeland defense perspective. The architecture for Golden Dome is still taking shape, or if it has taken shape, I haven't seen it yet, but we broadly know what it looks like and what functions it will have to perform regardless of the final architecture. A key part of this is that the earlier you can detect a threat— and not just one threat but many—the better. This is going to drive space sensing and custody. Space sensing and custody will proliferate across LEO and MEO with persistent tracking on orbit, fusion, and crosslinks to do all that.
I mentioned on-orbit fusion because I think it's really important. Historically, if you didn’t have to operate on timelines that included hypersonics and react at machine speed, you could have an on-orbit sensor that detects a missile launching somewhere, pipe that information down to a ground station, process it, and then decide what to do. Honestly, that is just too slow. There will need to be the ability to deploy things like automatic target recognition to space so you can close the kill chain more quickly. Crosslinks are really important because, preferably, we will have multiple assets tracking these missile launches, and you can improve your probability of detection and correct classification by fusing all this data together. We will see things like optical crosslinks— which are incredibly high data rate—continue to proliferate so that we can do this on-orbit fusion, classification, and detection.
We talked earlier about battle management command and control, maybe in the counter-UAS scenario, but the same thing applies to Golden Dome. You have to have a high-assurance data fabric. The foundation of that will likely be things like optical crosslinks as the physical layer. Once you have that data fabric, you must be able to do AI-assisted track correlation and discrimination. These are easy problems when there is one object in the air, but imagine thousands: some hypersonic, some decoys, some UAVs with warheads. That is why you need AI-assisted tracking and correlation to discriminate between threats, decoys, and civilian air traffic.
Once you have all that and you are connecting the data and fusing it and identifying the tracks you need to prioritize, you must be able to pair those with shooters at a continental scale. This is not just a theater anymore; we are protecting the entire continental United States, Alaska, Hawaii, and hopefully all of North America. Being able to do that sensor-shooter pairing is a non-trivial problem even for AI, especially in an environment with incredibly high state space and sometimes unreliable information. Track quality can vary, there are decoys, and quite honestly, it's just the fog of war.
So you have to be able to sense and track and then do the sensor-shooter pairing. Once you've paired it to a shooter, we cannot afford to have hundreds of millions of dollars' worth of shooters, where each interceptor is super expensive. We have to have an affordable effector stack. That includes hit-to-kill interceptors— we will still have to hit a bullet with a bullet— but we also need more cost-effective layers. We will need high-power microwave and lasers, particularly for counter-UAS and counter-cruise-missile saturation, so you can use lower-cost effectors and interceptors for volume and reserve the high-end interceptors for exquisite threats.
Underlying all of this is resilient communications and PNT—positioning, navigation, and timing— all incredibly important. I cannot overstate the importance of resilient comms. Back to your earlier question about what we learned and what we should not overfit from past conflicts: one thing we overfit from the Gulf War era was the assumption that we would always have electromagnetic superiority, which means uncontested communications. That will not be the case, and we have seen this in Ukraine. We are going to have to have resilient communications, which means spread spectrum, frequency hopping, and incredibly high-end systems to maintain connectivity.
The last thing I want to mention— because I’m an engineer— is digital engineering and test. This gets to our ability to iterate more quickly. We need national-scale digital twins so we can do everything from testing to, perhaps most interestingly, training sensor-shooter pairing algorithms via reinforcement learning. You only get one chance to do this in real life, so to train these algorithms, you need to run them millions or trillions of times in simulation.
You also asked about how startups can plug in. There are tremendous opportunities. At L3Harris, we are looking for startups to plug in everywhere we are plugged into Golden Dome. Data and AI are obvious areas. AI starts and ends with data and access to data, and to the extent that startups can build capabilities for sensor fusion and discrimination, these can be microservices within a broader architecture, especially when we can provide real datasets. The same applies to training non-kinetic effectors like high-power microwave systems. There are a couple of startups actively working in high-power microwave. And, gosh, I don’t want to sound like Austin Powers— or Mr. Evil— talking about “lasers on sharks,” but high-power laser systems will be increasingly important. In the startup world, that might mean companies developing high-power laser capabilities, but there is an entire system required to actually use them. For example, beam directors: if you have a high-power laser, you then need a mirror with a steerable beam director to aim it at a target.
Digital test ranges and manufacturing technology are also important. There is opportunity across the board.
Akhil 30:00
Andrew, so many awesome things right there—I don’t even know where to start. Let me dive into a couple of them. Let me start with SCA. You mentioned the comm-resilient piece. To you, that golden dome is not new in concept, right? We’ve been talking about it; I think The Economist ran a piece comparing Star Wars to Golden Dome. What has fundamentally changed that might make Golden Dome different today? And is the long pole in the tent, from your standpoint, the technology, the resilient comms, or something else?
Andrew 30:31
No. So I think we have all the technology to build Golden Dome today. That is true. We have everything from solid rocket motors that perhaps could survive in space—material advancements—to the sensors we need to detect these things. Really, it’s a monumental engineering effort to create a missile defense system that covers a continent. To me, it’s probably the ninth wonder of the world. The technology is there; it’s a question of how we integrate it and whether we have the national will. It’s like the space race. It took the will of a president to get to the moon. I think similarly here, it will take the will of a president to drive this forward.
Akhil 31:23
Thanks, Andrew. And on that note, you talk about digital twins. We think of this beautiful nexus between the physical and digital world. Can you unpack a little bit what you mean by digital twin? In the news you hear about the B-21 Raider being designed in the digital world. At the same time, a lot of us who have served or worked in the automotive space know that at some point you need to get the thing out into the world because you’re not going to be able to model everything. How are you all thinking about that?
Andrew 31:52
Yeah, a couple of things. To the extent that we can use digital twins to move faster, so that you can take your lumps in ones and zeros rather than in hardware and a full redesign, that’s where we focus a lot. At L3 Harris we are laser-focused on moving faster. The Navy and other services are saying the fight is tonight, and that’s how we use digital twins.
I’ll say the most exciting place for me in terms of digital twins and simulation is how we can use them for training and fine-tuning AI models. An example is in the RF space. Cognitive EW is increasingly important. Any waveform you deploy—radar or comms—the adversary will figure out how to jam it, and they’re deploying cognitive EW so they can adapt incredibly quickly. We as a nation have to be very good at generating new waveforms, new EA and EW techniques, and so on.
Back to digital twins: as a nation, we probably need to set up an electronic warfare simulation environment. The goal is to create AI agents for electromagnetic superiority. If you have a model you want to fine-tune for this complex space of electromagnetic battle management—where you have to communicate while an adversary jams you and both sides are adapting—you need an EM playground. But EM propagation is so complex that you can never train a model at full fidelity. Even with future computing, you will never do that. So what we’ll need is a digital twin with varying resolution.
You start with incredibly coarse resolution in the simulation environment and use that to get a gross training of your model—you run it a trillion times. Then you turn up the resolution by a factor of ten and fine-tune. Turn it up again and fine-tune. You take this model that started coarse and refine it. Remember how AlphaGo Zero was trained? The first generation of chess-playing AI relied on data scientists hand-coding techniques. Then someone realized it’s more interesting to have one AI agent play another without even telling them the rules. Very quickly, by running trillions of games, it reached superhuman capability. Now my iPhone can probably beat the best chess players because of reinforcement learning in a simulated environment.
Doing the same thing in the RF space will be increasingly important as we move into cognitive EW, cognitive comms, and cognitive warfare.
Akhil 34:57
Thanks, Andrew. Real quick, for those who aren’t as steeped in EW, what is cognitive EW?
Andrew 35:02
Yeah, so cognitive EW can mean a number of things. Very roughly, a waveform is the way that I communicate in the RF spectrum, the way that I communicate with you. FM radio is an example of a particular waveform. AM radio is another. 5G is another; it's actually a suite of waveforms. Cognitive waveforms can mean a number of different things, and you can think of them in levels. A relatively basic cognitive waveform would be a cognitive frequency hopper. Let’s imagine I’m communicating with you on a certain frequency—one megahertz, for instance. It’s not a good frequency, but let’s imagine it. If someone starts to jam one megahertz, I’m going to jump to two megahertz. If someone starts to jam that, I jump back to one. If they're still there, I’m going to jump to three. So the most basic cognitive EW involves clever ways to do frequency hopping and clever ways to do your modulation. I might switch from AM to FM, for instance. That would be your lowest-level cognitive waveform.
Some of the higher-end stuff—let’s call them generative waveforms—allow you to create the waveform on the fly, just like generative AI creates a poem on the fly that is optimized for whatever you want to optimize for. In this space, you can say, “I want to optimize for throughput given this specific jamming environment,” or, “I want to mimic 5G or Wi-Fi and hide in the noise.” Because generative waveforms leverage the most advanced AI, you're able to change the objective function and do essentially limitless things with it. I’ve seen tests where these cognitive waveforms are able to outperform classical waveforms in very harsh jamming environments.
The reason they aren’t deployed to the edge right now is that they are incredibly compute-intensive. It takes quite a bit of compute to do this. But as compute cost comes down, I expect to see these deployed more and more—probably first through SOCOM and other early adopters, and then more broadly across the U.S. as cognitive waveforms and cognitive EW mature.
Maggie 37:30
Andrew, I want to shift gears a little bit and specifically talk about how you view startup–defense prime collaboration. I know L3Harris has really been a leader in this, and it has become a big part of your strategy as the trusted disruptor in the defense space. So maybe the first question I’ll ask is: where do you see opportunities for startups to really excel, versus what kinds of technology areas do you think defense primes will continue to remain dominant in?
Andrew 38:06
Yeah. I’ll say that defense primes, and where I put all of my cycles as the CTO of L3Harris, are focused on operationalizing AI. There’s some really interesting space here where the U.S. technology ecosystem, including venture-backed companies, can work with the larger primes. Why is this a good marriage? L3Harris is really good at having exquisite domain expertise. We spend a lot of time in customer spaces understanding their hardest challenges. We tend to have access to a lot of data—we build EO/IR balls, we build radars, we build radios. All this data can be used to take a model or a capability developed by a startup or partner and fine-tune it for a mission application.
The final thing we have as defense primes is real estate. I helped stand up and was part of one of the premier divisions in the National Lab system that does AI—the AI and Data Analytics Division at Pacific Northwest National Labs. They’re great at AI for national security and high-consequence environments. They have some of the world’s best data scientists and data engineers. Their challenge is access to real estate and access to operational systems. If you’ve got models, whether they’re coming out of the national labs or the venture tech ecosystem, you have to have places to actually deploy those so they can start to have an impact on the warfighter.
Missionizing AI requires us to be really good at selecting the right model. Not every use case is an AI/ML use case. So we need to be good at determining which startup model, open-source model, or partner model we want to adopt, how we fine-tune that model with real operational data, and then how we get it through the ATO process so that we can actually deploy it to the edge. Deployment to the edge is sometimes non-trivial. This is where Code Metal is helping us. It’s pretty easy to deploy AI to GPUs. It’s less easy if you’re in a SWaP-C—size, weight, power, and cost—constrained environment with only access to FPGAs. These systems need to be very power conscious because you’re on the edge, on a radio a soldier has to carry for 10 hours—or really 10 days. It has to operate for those 10 days. Imagine if your iPhone lasted for 10 days. I’d be much happier.
Code Metal helps us translate code from Python or MATLAB into embedded code on the edge—FPGAs and similar hardware. That’s where primes like L3Harris and others are focusing: operationalizing AI. Startups, on the other hand, bring novel AI/ML, edge autonomy stacks, and synthetic data generation. These are spaces where there have been tremendous commercial investments in Silicon Valley and the U.S. technology ecosystem. If you’ve got novel models, particularly ones that could work in SWaP-C environments, we’d love to talk and figure out how we can start the process of missionizing and operationalizing them.
Maggie 42:00
Andrew, how do you all determine when you're going to partner with another company or startup versus when you're going to build some of that technology in house? I know L3Harris has what, 10,000 engineers or something employed?
Andrew 42:15
Twenty thousand engineers. Yeah. And so it's a good question. All right, so we would always rather accelerate, accelerate, accelerate. And the way that I look at partnerships with startups is that we have technology roadmaps, and our technology roadmaps actually don't start with technology. If anybody starts talking to me about technology for a technology roadmap, they're wrong. You start with a mission. These are the hard problems that we at L3Harris want to be able to contribute to. So that's one end of the bound. The other end is what technology we have today, meaning L3Harris capabilities that we have right now. From that, I've got two endpoints. I've got the mission, the customer problem we're trying to solve, and the technology we have today. That's the gap. If there is a company that can accelerate the closure of that gap, then I absolutely want to talk to them today.
Maggie 43:06
What do you see startups most misunderstand about working with a large defense prime like L3Harris?
Andrew 43:14
Yeah, I think there's sometimes a misperception that large defense primes don't want to go fast. And to be honest, a company as large as L3Harris is not a monolith. There are parts that will still be slow, but by and large, we care about exactly the same things as everyone else in this ecosystem. We want to go fast and hard and solve hard problems. From the flip side, if I take that question from the other angle, I get a lot of misperceptions from my engineers about startups. Some of the misperceptions they have come from engineers who, particularly if they've never worked in a startup, see a startup almost as a supplier. The startup brings a widget, and the engineer says, this is five degrees off from the widget that I needed, sorry, don't want to talk to you. What they don't realize is that if they just engaged a little bit with the startup, the startup would be more than happy, on an incredibly short timeframe, to shift that five degrees and build something that is absolutely applicable to the problem the engineer is trying to solve, that gap they're trying to close. A lot of my internal work at L3Harris is helping engineers understand that you've got to invest your time in startups so they understand the problem you're really trying to solve. You also have to be conscious of their time horizon. If you've got a pursuit where it's going to be a problem in three years and you'll finally get money for it in eight years, that is not an appropriate place to engage a startup. So we have to be aware of what problems are amenable to startups in terms of what a startup's traditional run rate is.
Maggie 44:58
What advice do you have for startups to work with a company like L3Harris? What can a startup do well, or what have you seen startups do well? Sort of the flip question to what I asked before, which is what do they not do well?
Andrew 45:10
Agreed. So startups that do well, and this is sort of the flip-side answer, are adaptable and willing to fit into, and I'm not going to say the L3Harris need, but the customer and mission need. That adaptability is incredibly important. I think it's actually inherent in most startups’ DNA. They don't run agile. If you talk to a founder about agile, they’ll hit you. It is pivot, pivot, pivot until you find product-market fit. I would advise that startups not confuse a great demo with a deployable capability. You have to understand your concept of operations but also your concept of employment. Give us a call and we will help you understand the concept of employment. A lot of startups don't have access to SCIFs and high-side information, so we can help shape your product in appropriate ways so that it is actually solving the problem the warfighter has and addressing the threat they're facing.
Akhil 46:20
Thanks, Andrew. Maybe to flip it again, there’s a perception in certain spaces that the primes are like the Evil Eye of Sauron. They've existed for a while, they're there. But I think there's this interest from startups in some of the points you're bringing. There's also real value in finding the right partnerships at the right time and place to achieve effects ultimately for the mission. My question is, if you're a startup, how should you be evaluating and what should you be expecting out of a partnership with someone that's a lot bigger than you, who may have a direct relationship with the customer, when even the startup wants that direct customer feedback and engagement? How do you navigate the limited time a young startup has in finding which partnerships make sense and which ones are actually going to drive value across both organizations, and ultimately for those who need it most from a mission standpoint?
Andrew 47:17
Yeah, absolutely. I think you're looking for partners who are interested in a true strategic partnership as opposed to a transactional relationship. That is really important. When you talk to various primes, make sure they're willing to invest their time in you. I'll give you an example. At L3Harris, we have internal money that we use to run things we call validation projects. If you come in, Akhil, and you've got a novel widget that you've invented or a software app, we invest our own money at L3Harris to figure out how that app can be used to fill one of those gaps I talked about to solve a hard customer need. That's an example of a strategic relationship where we're investing in the relationship, you're investing in the relationship, and together we're able to move faster and solve harder problems than we'd be able to do individually. I'm really looking for that. A lot of times when primes engage with startups—and I haven't seen this at L3Harris, but I have seen it elsewhere—they want to be a bully around IP rights, like “We want to own IP rights.” What they don't realize is that absolutely kills the startup. It makes them unappealing to investors. So watch your IP rights. Make sure the prime you're working with is looking to generate mutual value for the prime, for the startup, and for the customer, and not just for themselves.
Maggie 48:48
If you had all the CEOs and CTOs, the senior leadership of other large defense government contractors, in a room, what advice would you give them?
Andrew 48:58
The first thing is definitely focus on open systems. The reason this is important is that the United States’ asymmetric advantage is our ability to innovate faster, and open systems—while they take a bit more time up front—really open up their architecture on the back end. If some startup comes up with a new capability, because it's an open architecture they'll be able to plug it in quickly, get it operationalized, and make a real difference. Whereas if we have a closed ecosystem where everybody owns their own IP and data rights and nothing plays well with each other, we're never going to get to that future where we're able to rapidly deploy capabilities and iterate more quickly than adversaries.
Akhil 49:47
Andrew, maybe one level deeper on open architecture: is the government providing the right incentives for that ecosystem? And what would be the one policy change you would make? When you think about modular open architecture, people are trying to build a business that's defensible and not dependent on year-to-year uncertainty about whether their system will survive. There’s a healthy tension between what a company needs to defend its position and what’s required to provide value and create the right incentive structure. How do you see that?
Andrew 50:18
I would say I'm never worried about defending anything. If you have to defend based on a closed architecture, then you're not moving fast enough. Our strategy is to continue to move faster than anybody can adopt, and we will continue to do that. You also need to really understand where you make your money. You make your money off your implementation of the open architecture. Waveforms are a good example—we’ve talked a lot about communications waveforms. Many of them are open; they are government-owned protocols. That doesn't mean L3Harris doesn't make money off them. We still take that waveform, translate it into software, implement it on hardware, and sell an operational capability. Open architecture doesn't mean you can't make money.
What we need more of from our customers is consistency. Probably 90% of customers say “open architecture, open architecture, open architecture.” There's a vocal 10% who say, “Well, with closed architectures we can go faster,” because they see the upfront initial investment—maybe you're 10% slower—as the impediment. But let's look at a concrete example. We talked about tanks earlier. You could create a tank that is a completely closed system. You don’t allow anyone to plug in a new EO/IR ball, nobody can plug in a new communications module. You can do that faster; you can deploy a closed ecosystem right now.
But let's say you spent a little more time and built this on an open architecture, which means you can switch out modules. As a new radio comes online, one of these cognitive radios or cognitive EW systems, you just take out the old one and put in the new one. That could come from the startup ecosystem or from a traditional defense contractor. A new radio comes online, a new EO/IR ball comes online, and it allows a much faster upgrade cycle. In this tank example, you can see how the proponents of closed systems are right that you can get a fixed capability to the field more quickly, but the long-term ability to iterate and upgrade is far better with an open architecture.
Maggie 52:31
What’s a technology that you are bearish on in the short term but bullish on in the long term?
Andrew 52:37
All right, I did mention high-energy lasers and HPM earlier. There is still a way to go before those are real operational capabilities. I think everyone’s seen a couple of interesting prototypes and examples where people have shown, hey, we're able to kill a UAV or do some other interesting things. Today, weather, power, and beam control are the limiting factors. In the future, these will be indispensable for a layered, cost-effective defense. The other is quantum. People tend to think about quantum in three buckets: quantum sensing, quantum communications and networking, and quantum computing. At some point, quantum systems such as quantum sensing with Rydberg atoms are going to far outperform RF antennas and RF capabilities for sensing, comms, and SIGINT. In the near term, though, I don't see an impact in the next five or six years. So again, bearish in the near term, but long term these things are going to be differentiators and game changers.
Maggie 53:41
And I know you've been in the national security technology space for decades now. What's the technology that you thought would be further developed than it actually is today, back in your days at MIT or one of the labs or elsewhere?
Andrew 53:56
The technology that I thought would be further along is true CJADC2. For those of you in this field for a while, it was net-centric warfare before CJADC2, and before that it was something different. We still fight in data silos with bespoke messages more than we should. Really being able to bring that “any sensor and every shooter, every sensor and any shooter” capability to the battlefield—I thought we'd be further along. It's been a huge focus for the DoD for at least two decades, and I hope we continue to make progress around true CJADC2. Less developed but pleasantly surprising is autonomy at the edge. Small teams are absolutely deploying incredible capabilities. Everything from—gosh—we talked about Overwatch AI to Shield AI (not to be confused with Shield Capital) to Overland AI. These are just incredible advances in true autonomy at the edge. What are some tech—
Maggie 55:02
—areas where our adversaries are ahead of us in developing and deploying, and what will it take for the US to catch up?
Andrew 55:13
One place I'll point out is hypersonics and long-range strike. So PRC and Russia—munitions like the DF-17. I will say that the United States, and we here at L3Harris Aerojet Rocketdyne, have eye-watering, exquisite hypersonic capabilities, with the ability to do computational fluid dynamics and advanced materials. The United States does have some incredible hypersonic capability. That said, in deployed, operational hypersonics, we are behind. We are absolutely behind. That’s the space where we want to keep up. Shipbuilding and the industrial scale and industrial might that China is able to bring to the table—our fleet growth is currently being outpaced by large numbers. Tactical electronic warfare—actually, I’ll lump these next two together: tactical electronic warfare and low-cost drone ecosystems. Currently, Russia is beating us there, and it's largely because they're actively engaged in a war in Ukraine. What this does for them is tighten their learning loop. They deploy an EW technique or system, and if it doesn't work they are forced to continue to update and cycle it. Our EW capabilities aren't nearly on the war footing that Russia’s are.
Akhil 56:36
Thanks, Andrew. I want to end where we started, with you being a technologist. If you had one extra hour in your busy day—you’ve got a ton of things, the leadership aspects of running the organization, the connecting-the-dockets piece, and then somewhere in there you get a chance to vibe code—what would you spend that extra hour doing, and how would you break that apart?
Andrew 57:00
You know, I think as technologists it's easy to become enamored with the technology, and I spend an incredible amount of time, like I said, even developing my own models. If I could find an extra hour, I would spend that time with the operators, with the people who are actually using the technology. At the end of the day, as a CTO and technologist at L3Harris, I believe every engineer at L3Harris is here to solve hard customer problems. The better understanding I can have—and that every engineer can have—of what those problems are and how we might be able to solve them more creatively, the better.
I was recently talking with a customer, and they said, “Hey, Andrew, I just need more bandwidth. I need more throughput on these radios.” We talked for a little bit, and it turns out that’s not actually what the customer needed. They needed the right information at the right place at the right time. That drives you to think: yes, I could give you a radio with a bigger power amplifier and make this poor soldier carry around a big dish antenna. Yes, that could be a solution to that problem. But because we sat down and talked and understood the problem and the threat they were facing, it turns out the right answer is deploying intelligent agents to the radios to prioritize the data so they’re getting the right information at the right time to provide decision and information advantage. More time with the customer, more time understanding what they’re facing and what their threats look like so we can create better solutions—that’s where I’d spend my time.
Akhil 58:25
Awesome. Andrew, thanks so much for joining us.