Shield AI’s Ryan Tseng on Building an Autonomous Future for the DOD

Summary

The AI Policy Podcast: Shield AI’s Ryan Tseng on Building an Autonomous Future for the DoD

Host: Gregory C. Allen (CSIS)
Guest: Ryan Tseng, Co-Founder & President of Shield AI
Date: June 10, 2025

---

Brief Overview

In this episode, Gregory C. Allen sits down with Ryan Tseng, co-founder and president of Shield AI, to discuss the evolution of defense technology startups, the company’s origin story, and their work in building AI-powered autonomous systems for the battlefield. The conversation dives deeply into Shield AI’s operational support in Ukraine, the unique challenges of defense entrepreneurship, and where military AI and autonomy are headed—particularly in GPS- and communications-denied environments.

---

Key Discussion Points and Insights

1. Origins and Motivation Behind Shield AI

• Ryan Tseng’s background as a serial entrepreneur, having sold his first company (focused on wireless charging) to Qualcomm, and his search for a “noble mission,” “extraordinary people,” and the opportunity “to define the possible.”

  • “If I could find the intersection of three things, I’d have that fire for the rest of my life... A noble mission. ...Extraordinary people. ...A chance to define the possible.”
    (Ryan Tseng, 04:29)

• The company’s inception during an era when defense tech was considered “idiotic” by Silicon Valley—before the sector’s recent popularity.

  • “I think it was categorically considered idiotic to start a technology company focused on defense.”
    (Ryan Tseng, 06:24)

• Early struggles with venture capital, including a notable refusal to take funding that required abandoning the national security mission.

  • “I’d rather go out of business trying to make mission impact, than take this money.”
    (Andrew Reiter, as relayed by Ryan Tseng, 11:47)

• Shield AI’s original focus: providing autonomous quadcopters for special forces—solving GPS-denied navigation, autonomy in loss of comms, and other technical hurdles.

---

2. The Climb: Scaling from Start-up to Major Defense Player

• The “climbing the aviation food chain” strategy:

  • Start with quadcopters → Progress to more consequential platforms (e.g., VTOL, jet aircraft).

• Key milestones:

  • 2016: First meaningful funding ($800,000) and contract with the Defense Innovation Unit (DIU).

  • 2017: Andreessen Horowitz leads Series A, signaling a shift in VC perception of defense tech.

    • “Sometimes I just look for what seem like the dumbest possible investments because it’s either crazy or it’s brilliant.”
      (Peter Levine of Andreessen, via Tseng, 14:02)

  • 2020: Acquisitions of Martin UAV (VTOL aircraft: VBAT) and Heron Systems (reinforcement-learning for jet aircraft; winner of DARPA’s AlphaDogfight competition).

    • “The AI system not just winning, but like smoking them by the end... it was a wake-up call to the fighter pilot community.”
      (Greg Allen, 15:13)

  • 2022+: First live F-16 dogfight with AI pilot; human-vs-AI live dogfights; Shield AI named Collier Trophy finalist.

---

3. War in Ukraine: A Test Lab for Military Autonomy

• Ukraine as a “wakeup call” for military innovation:

  • Initial failure: Early Shield AI aircraft deployments in Ukraine were unsuccessful; “not ready for the EW [electronic warfare] environment.”
  • Key challenge: GPS and comms denial are omnipresent; U.S. defense systems built for GPS-permissive environments failed to perform.
    • “People will take microwave ovens, turn them inside out, blast them up in the sky, and GPS is no longer functioning over very large geographic regions.”
      (Ryan Tseng, 19:14)

• Pivot and Breakthrough:

  • Redeploying VBATs with alternative navigation (AI-powered).

  • First successful mission: Provided GPS-denied targeting of a Russian SA-11 ($100 million asset).

  • The learning curve: Software malfunction on first live attempt; fixed and redeployed in just 24 hours—a rapid innovation cycle unheard of in traditional defense acquisition.

    • “It was like a scene straight out of the movie Interstellar…eventually they found it orbiting 60km away and they were able to bring it down and land in a sunflower field, actually not a cornfield.”
      (Ryan Tseng, 23:18)
    • “Changing the software on an aircraft in 24 hours…traditionally, anything that's flight-critical software…once it works, for the love of God, never touch it again.”
      (Greg Allen, 24:05)

  • Ultimately: Over 170 sorties, all performed without GPS at any point, vindicating an iterative, mission-driven culture.

• Cultural lesson: Mission stubbornness and the necessity of agile, rapid software updates in combat environments.

  • “The only way to be relevant, the only way to survive is to embrace rapid iteration, rapid turnaround times and rapid change.”
    (Greg Allen, 25:43)

---

4. Policy, Process, and Lessons for the DoD

• Need for process and regulatory reform:

  • Traditional U.S. certification process is too slow for dynamic battlefields.

  • Ryan’s “three pillars” to win in autonomy: performance, assurance, and speed—ideally, all achieved within a 24-hour development cycle.

    • “I have sort of three pillars…performance…assurance…and speed. You have to be able to deliver one and two in a time frame of relevance. …if you can get through that cycle in 24 hours.”
      (Ryan Tseng, 28:17)

  • Current state: Even in the DoD, changing flight-critical software can take a year due to risk aversion and process complexity.

• The U.S. is learning—cautiously:

  • EW environments and affordable/dense air defense systems (even from non-peer adversaries like the Houthis) are changing fundamental assumptions from earlier wars.

  • The need for “mass” (large numbers of affordable drones) is critical—Ukraine targets over 1.5 million drones/year.

    • “How can you send a million and a half drones downrange per year and not get effect?...there is a big… targeting challenge…It’s been hard to get eyes and ears deep across the forward line of troops.”
      (Ryan Tseng, 33:10)

  • Takeaway: Intelligence, coordination, and effective targeting (with drones/AI) are as essential as sheer numbers.

---

5. AI, Autonomy, and Future of Military Operations

• The real state of battlefield autonomy:

  • Much AI & swarming showcased globally is GPS-dependent and non-robust for EW conditions.
  • Shield AI’s work: Building robust alternatives (e.g., vision-based navigation, inertial systems), but there are still “eight big conditions” to solve for.
  • The field lacks the industrialized tooling and infrastructure that mature software sectors enjoy—the process is not yet scalable or democratized.
    • “There are only so many of those people in the world…how do you…make it accessible to the defense industrial base?”
      (Ryan Tseng, 41:33)

• Shield AI’s vision: Mass deployment of “hyper-intelligent AI on every vehicle or weapon in the arsenal.”

  • First step: Get reliable, scalable “stand-ins for GPS.”
  • Second step: Enable battlefield-relevant swarming and teamwork without comms or GPS.

---

6. Challenges of Defense Entrepreneurship & Acquisition Process

• The daunting complexity of U.S. government acquisition systems—too many stakeholders, “so many people…on the chessboard who can say no.”

  • “It’s shocking… the number of stakeholders that have a vote.”
    (Ryan Tseng, 44:05)

• Contrast with Ukraine: Lean, evidence-based, and rapid procurement cycles—test on the EW range, prove efficacy, deploy if it works.

  • “Everybody has given us PowerPoints, those just don’t fly here. …They took us out to their EW range… slammed VBAT for a couple days…and then we started to get momentum.”
    (Ryan Tseng, 48:46)

• Potential U.S. solutions: Promoting problems-based, rather than requirements-based, acquisition to foster innovation and accelerate the deployment of needed solutions.

  • “I think that there’s a huge time savings and unlock there… usually much quicker to identify you have a problem than specify the solution to that problem.”
    (Ryan Tseng, 47:02)

---

7. Looking Ahead: 2030 and Beyond

• Shield AI’s aspiration: Mission impact at scale—deepening values of honor, service, and excellence.
  • “I hope we’re making mission impact at tremendous scale, still grounded in the values of honor, service, and excellence.”
    (Ryan Tseng, 50:01)

---

Notable Quotes & Moments (with Timestamps)

• On finding purpose:

  • “If I could find the intersection of three things, I’d have that fire for the rest of my life... A noble mission. ...Extraordinary people. ...A chance to define the possible.” (Ryan Tseng, 04:29)

• On early investment struggles:

  • “I’d rather go out of business trying to make mission impact, then take this money.” (Andrew Reiter, 11:47, recounted by Tseng)

• On battlefield software innovation:

  • “Changing the software on an aircraft in 24 hours... the traditional ethos... is once it works, for the love of God, never touch it again.” (Greg Allen, 24:05)

• On success in Ukraine:

  • “Since then, I think we've done over 170 sorties with the aircraft in the conflict. Zero of those sorties have had GPS at any point in the mission.” (Ryan Tseng, 23:53)

• On DoD’s need for reform:

  • “Future battlefields…will be defined by the agility of the software that's behind our forces, that enables them to operate effectively as conditions are fast-evolving.” (Ryan Tseng, 25:55)

• On scaling AI autonomy:

  • “There are only so many of those people in the world. And so how do you take something…that's required very high-end expertise and make it accessible to the defense industrial base so that…we can deploy it not once, but across the number of vehicles and weapons that the mission demands?” (Ryan Tseng, 41:33)

• On “problems-based” acquisition:

  • “I think that there's a huge time savings and unlock there…usually much quicker to identify you have a problem than specify the solution to that problem.” (Ryan Tseng, 47:02)

---

Important Timestamps for Segments

| Segment | Timestamp Start | |------------------------------------------------------|---------------------| | Introduction & company origins | 00:00 | | Early days, venture funding, founding moments | 03:17 | | Strategy: quadcopters to jet aircraft | 09:12 | | Key milestones: DIU, Andreessen Horowitz, Heron | 12:30 | | DARPA AlphaDogfight & implications | 15:07 | | Ukraine deployment (failures, lessons, breakthroughs)| 18:31 | | Software change breakthroughs; rapid iteration | 23:12 | | DoD/US process versus Ukraine speed/approach | 25:43 | | Mass in drone warfare, targeting challenges | 32:49 | | Cultural and process barriers in U.S. acquisition | 43:27 | | Need for regulatory reform and problem-based buying | 46:38 | | Shield AI’s vision for 2030 | 50:01 |

---

Summary

This episode is a candid exploration of the intersection between cutting-edge AI, startup hustle, and the realities of military conflict and bureaucracy. Through Ryan Tseng’s journey, listeners gain insight into the technical, cultural, and organizational shifts propelling military autonomy forward—and the hurdles that still stand in the way. Ukraine’s war serves as both crucible and warning, highlighting the urgency of rapid innovation, flexible procurement, and software/game-changing agility.

The future of defense tech, as Shield AI envisions, is one where ultra-adaptable, intelligent autonomous systems are as crucial as the traditional platforms they supplement or replace—and where mission-driven stubbornness can literally tip the scales on the battlefield.

Transcript

[A]

Foreign I'm Gregory Allen and welcome back to the AI Policy Podcast. Today you're going to hear a conversation I had on June 9 with Ryan Tseng, the co founder and president of Shield AI which is one of the most successful and important startups focused on AI power technology for the military. We're going to talk about the his experience as a founder and creating this company, his experience supporting military operations in Ukraine and what the future holds for military AI and autonomy, and a world in which drones are an ever more important part of combat. You can listen to the conversation here on the podcast or you can watch the full video event on our website csis.org I hope you enjoy it. Thanks. Good afternoon, I'm Gregory Allen here at CSIS where we've got an incredible event, a discussion with Ryan Singh, the co founder and president of SHIELD AI. We're going to get into a lot today, including Ryan's background, the evolution of SHIELD AI as a company and the current state of of AI and autonomy in warfare and its adoption not just in ongoing conflicts like the war in Ukraine, but also by the United States Department of Defense. Ryan is a serial entrepreneur, having previously founded the company Y Power, which was acquired by Qualcomm and now leads SHIELD AI from its inception almost a decade ago to now being one of the largest defense technology startups in the world and at the forefront of so many of the trends of AI and autonomy. Ryan, thank you so much for coming to csis.

[B]

Thanks for having me, Greg.

[A]

So you and I actually have kind of an interesting overlapping life story. You obviously vastly more successful than me in this overlapping life story. But your brother, Brandon Singh, also a co founder of SHIELD AI, was a Navy SEAL and then went to Harvard Business School, where I also went to graduate school. And on the campus at the business school they have this thing called the Ilab where they try and incubate exciting new startups. And 10 years ago, one of those exciting startups was Shield AI a company thinking about the future of defense technology. And I have to say, at a time when defense technology was not a very attractive word to be uttering in Silicon Valley venture capital firms. So Brandon, as I said, a former Navy SEAL and you, a entrepreneur at the time with a different company, decided to join fellow forces and create this new company. And at the time I was like, well, there's a lot of startups in the Ilab. Some of them will be interesting, some of them won't be interesting. But wow, SHOF AI actually had a pretty good run over the past 10 years, the company has raised over a billion dollars of venture capital investment, was recently valued at being worth more than $5 billion, enjoys contracts with the Department of Defense, and has been a real supporter of the war in Ukraine, not just in words, but also in deeds, providing technologies used operationally by Ukrainian armed forces. I want to get into the meat of so much of that story, but I want to start with your story. So how did you come to be in the technology field? How did you come to be in the defense technology field?

[B]

Yeah. Well, just on the Ilab. I remember visiting it once, and it stood out to me because SHIELD AI was the only company. It had a desk in the Ilab. And I asked Brandon how it happened. They said they wouldn't give me a desk. So I just put my name tag on it. And he claimed a desk for his. For his own in the Ilab. And that was the beginning of. Of SHIELD AI but my story. Look, I started a company right after undergrad, focused on doing wireless charging. I sold that company to. To Qualcomm, and I ran the wireless charging group there for about four years. And in my last year at Qualcomm, for the first time in my life, I didn't feel like I had the fire in the belly that I had had throughout my career. So growing up, my parents had a couple of messages for me that became deeply ingrained in who I won, who I was, and, frankly, am today. One was work hard. Two was work hard. And three was work hard. And so in my last year at Qualcomm, not feeling that fire for the first time in my career was this sort of confusing moment. And I decided to take time off and reflect on what it was that motivated me that would give me that fire, not for the next year or five years, but for the rest of my life. And after some searching, I basically decided that if I could find the intersection of three things, I'd have that fire for the rest of my life. And one was the opportunity to contribute to a noble mission. Two was the opportunity to work with extraordinary people. And three was a chance to define the possible. And so I set about looking for companies or opportunities that sat at the intersection of those things. And it was my brother getting ready to prepare for his interviews at Harvard that led me to SHIELD AI. He asked if I would help him prepare for his business school interview. We're very competitive. We're brothers. And so this was a very unusual request, but, of course, I agreed. And it started with a simple question from me to him, which is what's the most challenging circumstance you've ever faced, and what did you do about it? Having been to business school myself and, you know, knowing what I said and a lot of people said, I was ready for a lame answer. What he shared with me was a story that brought me to tears. It was a story of service and sacrifice on behalf of teammates and people that couldn't stand up for themselves. And I remember thinking to myself, what an incredible mission. What an incredible person. And after that, several weeks later, when I encouraged him to think about what he wanted to do in life, he came back and said, I want to bring the best of what's going on in the autonomous driving sector to the mission of protecting service members and civilians. I thought, amazing mission. I also thought it was a dumb business. So I encouraged him to have a great time.

[A]

So I think people forget today because you've got exciting companies like shield AI, like Anduril, like Palantir, that have made defense technology a pretty hot sector. But in 2015, it was not a hot sector. This was not a fashionable idea for a startup. So what was going through your head when you thought this was not a good idea?

[B]

I think it was categorically considered idiotic to start a technology company focused on defense. And Certainly Palantir and SpaceX existed at that time, but the refrain from the venture community was, well, those are started by billionaires, so they don't count. But yeah, it was just. And I think at the time there was the maven thing going on at Google with Silicon Valley, and I think there was this letter, and people were just sort of openly against supporting the national security mission. So it was a pretty unusual step. And basically having started and sold a company to Qualcomm and knowing the sensibilities of that community, I just thought it was gonna be very hard to get the attention of investors.

[A]

And so what changed your mind?

[B]

My brother's a persistent person, and it seals if you haven't met them or persistent people. And Brandon, stubborn in a good is on brand. And what I told him is, look, go talk to 100 people that you don't know and get a sense of how real the problem is. And he said, I'll do it, but I want you to come with me. And I wasn't actually doing anything at the time. I was in between, looking for something to do. So I went with him. And the thing that struck me was the scope of the problem and how little was being done about it. I was shocked to learn how my brother and his teammates would clear buildings of threats. So this not having eyes and ears in the places where you need them was costing the lives of service members.

[A]

You're talking about the clear hold and build, you know, mission set, where you're going into a building, Somebody has to be the first human through that door. And you don't know if there's a machine, like, 10 machine guns pointed at that door or if there's a bunch of friendly people who mean no harm on the other side of that door.

[B]

And my brother gave me an opportunity to, you know, put on kit and, you know, experience it with him. And I was just blown away that there are people that go and execute this mission and frankly, humbled to live in a place where so many of those people sign up for that mission. But the problem not having eyes and ears in the right places can be just around the threshold of a doorway or it can be at a theater scale. And so it was a huge problem, and one that I felt like the technologies in the autonomy sector could truly contribute too.

[A]

So you mentioned your brother Brandon was sort of thinking about autonomous driving as his mental model. Your company's initial work stream was focused around aerial drones. Can you talk a little bit about that evolution?

[B]

Yeah. So it was a mission that we were familiar with. And then it also turned out that nobody would give us the keys to a jet, you know. Yeah. And probably a good decision. You know, when we had three people in a garage in 2015, who knows what would have happened to that jet? So we had a strategy that we called climbing the aviation food chain, where we would build an AI backbone that we would apply to quadcopters and allow those to be the first thing through a threshold, which requires solving a lot of hard problems. GPS denied navigation, autonomous decision making when you lose communications, because in many ways, the inside of a building, like a concrete structure or a cave, is a small version of an EW battlefield where GPs and communications are anything but assured. And so we started on the path to bring AI to clearing buildings with autonomous quadcopters.

[A]

This is for primarily a special forces customer community, initially for the special forces.

[B]

Community, with the intent of being able to use that AI backbone as we climbed the aviation food chain, getting to progressively more consequential aviation platforms.

[A]

That's amazing. So, you know, you talked a little bit about where you were in those early days. Now today, you know, you're a really exciting startup with a ton of growth and a ton of capability that you're already bringing. Can you sort of just walk us through the journey of who was Shield AI? You've talked a little bit about this in 2015, but who were you in 2017? Who were you in 2020? Who are you now in 2025? Talk us through the story of climbing. Climbing the ladder as you.

[B]

Yeah, I'd say the one common through that entire journey has been commitment to mission and values. I think we've been very clear eyed about who we've been from the very beginning. Our mission is to protect service members and civilians with intelligent Systems. And in 2015, that made our journey very difficult. Turns out by the time Brandon converted me to believing in the mission and the ability for the company to contribute to the mission, still nobody else in the world had been converted. And I remember just going to investor after investor and being told, no, this is never going to hunt. And when we finally got an offer for a term sheet for $5 million, it came with the criteria that we change our mission.

[A]

That's.

[B]

Whoa.

[A]

This is a pretty remarkable thing. So you've gone to venture capital company after venture capital company. Everybody's slamming the door in your face, and the first people you meet who give you, yes. It's like, yes, but you have to give up on national security. Is that right? That's incredible. I mean, that speaks to the time.

[B]

Yeah. I remember we had convinced our other co founder, Andrew Reiter, to quit his job. He was a rising star in AI and autonomy. And I assured him that, hey, we're going to raise the money, it's going to be fine. And so I remember coming back with, hey, I've gone through my entire Rolodex. The only option we have is this one. And it was a short conversation, but I remember Andrew saying, I'd rather go out of business trying to make mission impact, then take this money. And so we passed on that term sheet. It would be another year before we got a term sheet. Not for $5 million, but this time for $800,000 in 2016 that let us build the company that we wanted to build.

[A]

And that $800,000 letting you pursue the mission of national security. Yes.

[B]

Yes.

[A]

Great. And so who was your first customer?

[B]

First customer was Defense Innovation Unit, actually.

[A]

Okay. So this had this organization that DoD set up to actually help startups.

[B]

It's funny, it was sort of a joint coming of age of DIU because.

[A]

They were pretty young too.

[B]

They were. They didn't even have a location on their website. We just saw it get announced around the same time that we were starting the company. And we remember seeing a picture that, you know, somehow, through a reflection, my brother saw that it was at Moffett Field. And so he took an Uber there when we were in San Francisco and like, ran. Tracked down this building and tried to walk in.

[A]

Well, you're like, ask the guard at the gate. Yeah, where's the fifth animation?

[B]

So we got sent home and said, you don't just come in. Anyways, it was so anyways, that was the first program. But you asked about the evolution and coming back to that 2017, when we met, a big milestone, I think was in Andreessen Horowitz deciding to invest and lead our Series A round.

[A]

And which was a kind of a big moment for the entire sector. Right. It was like the. The big venture capital companies finally saying, it is. It is okay, you know, it is okay to invest in national security. And that was. You were part of that first wave, being taken seriously.

[B]

Yeah, yeah. And I think even for Andreessen, it was a bit of an unusual move. The partner who led it, Peter Levine, made a comment along the lines of, sometimes I just look for what seemed like the dumbest possible investments because it's either crazy or it's brilliant. And at the time, he just didn't know. But he was willing to make a bet on it, and national security was something that he was passionate about and willing to take a flyer on it. Just very quickly fast forwarding through. In 2020, we took our first step up the aviation food chain with the acquisition of two companies, one called Martin U that makes a VTOL tail sitting aircraft called the V Bat, can fly for about 12 hours. And we can talk about that later. It's like 12ft tall, 12 foot wingspan. And then we acquired another company called Heron Systems that had been doing some pretty remarkable work with reinforcement learning applied to jet aircraft. So all the way at the top of the aviation food chain and had famously one, I think the Alpha dog fight trial.

[A]

Yes. So this was certainly. I remember because I was in the DoD at this time, and the Heron system winning the DARPA competition where they had an AI pilot facing off in simulation, but facing off against real fighter pilots in dogfighting simulations. And the AI system not just winning, but like smoking them by the end of the story was kind of a wake up call to a lot of generals. It was a wake up call to the fighter pilot community that like, look, this may not be ready right this second, but we have to invest in this technology because the potential here is really exciting. And so. So that company then becomes a part of SHIELD AI and so where does that take you?

[B]

Yeah. And by the way, on that point, I think that one of the things that both DARPA and the team at Heron did so well, there was many pockets of people were starting to develop that really believe that AI and autonomy were going to be a significant part of the future of defense and national security. But the question always was like, how do you make that point? And I think that DARPA and Heron did such a good job of picking an example that really made the point in a powerful way that, hey, this technology is coming and when it matures, it's going to be a force to be reckoned with. So where it took us from there was we set the objective to fly an F16 with an AI pilot in real life by 2022. We met that objective. We set the objective to do a human versus AI dogfight and live flight in 2023. And we hit that objective. And then in 2024, that program, along with where the program, including SHIELD AI was one of four finalists for the Collier Trophy, which is for the highest achievement in aerospace and astronautics in the preceding years. The Wright brothers won an Chuck Yeager won it, the moon landing won it. So it was pretty amazing to be one of four finalists for that award on the back of the nine year journey to that point. Along this same pathway, we continue to expand the application of AI and we could talk about it later, but brought our technology to, brought our AI technology to vbat achieved some pretty remarkable outcomes in Ukraine, some of which were written about in August last year by the Wall Journal.

[A]

So I, I definitely want to go there because you've talked about, you know, the evolution of the company, climbing the autonomy ladder, climbing the aviation food chain. But at the same time, you know, you're evolving as this, as this company. The world is changing in a big way. Waking up to the reality that we're in a new era for national security. You know, Alpha dogfight, a very crystallized point there. Russia's full scale invasion, Ukraine, a very crystallized point there. And so you have been really out front amongst defense technology executives talking about the importance of supporting Ukraine and putting your money where your mouth is, putting your kit where your mouth is. So walk us through the story of how SHIELD AI got involved in Ukraine, how it went initially, how it's going now.

[B]

So SHIELD AI got involved in Ukraine, I think, alongside many companies right at the outset of the conflict. And we deployed some aircraft pretty early on. We suspected that they wouldn't have some of the right capabilities, but we weren't sure. And those aircraft were unsuccessful for the same reason that I think basically most systems today continue to be ineffective. They weren't ready for the EW environment.

[A]

What do you mean by that? For folks who are not familiar with electron warfare and what it really means in today's conflict?

[B]

Yeah, There are two very fundamental assumptions that a lot of weapon systems make. One is that you have gps, right? And gps, because it's coming from a satellite far in space that has to cover a large area, it's just a very weak signal by the time it gets to Earth. So basically, people will take microwave ovens, turn them inside out, blast them up in the sky, and GPS is no longer functioning over very large geographic regions. And then two, your communication channels, same story, those end up being degraded. And so if you had a remotely piloted aircraft, and, you know, one of the foundational assumptions for anything to be successful is, where am I in the world? It no longer knows.

[A]

Right. So, like, if you think about the, the drones, the, the really charismatic drone aircraft of the Iraq and Afghanistan conflicts, like a Predator or a Reaper, they spent most of their operational life having perfect comms, perfect gps, and they were designed, you know, to live in. We go to Ukraine, facing a different kind of adversary, those assumptions are no longer able to be taken for granted at all. And everybody, all the US Companies who are used to always having the best stuff, you know, come up against this, you know, very painful realization that what they have is not fit for purpose for. For this conflict. So it's as you said, you know, you had to realize whatever you're going to make, it has to work in a world with no GPS and no comms.

[B]

Yes. And so, you know, we were faced with a decision, focus somewhere else or solve the problem. And we just felt strongly we could solve the problem because that was fundamental to our autonomous quadcopters that had to work inside of buildings. And so we set about bringing that to the vbat. And so August 2024, it was disclosed, but shortly before that, we made the decision to deploy new VBATs that would work without GPS and with heavily contested communications. And the culmination of it was the Ukrainians using it to find a Russian. I think it was SA11, like $100 million air defense asset, finding it without GPS targeting it and resulting in a successful strike and destruction of that asset.

[A]

And to that point, the VBAT in this case is the spotter.

[B]

The VBAT is the spotter, and it's generating targetable coordinates without gps, which basically blew the minds of Everybody up and down the Ukrainian side because, number one of how deep in the territory we were. And the fact that we were able to do that without GPS was just this insanely compelling outcome just on the path to do it. Interesting story and I think important for the future of autonomy in any military system. When we re attacked in 2024, it initially actually wasn't successful on our first attempt. So we decided to solve the problem. We got all of our notes and intelligence from the Ukrainians about when we would have gps. And so we use that to inform the design of the system. Basically, at 200ft altitude, we'd start getting hit by Ukrainian jammers, which of course are active to protect from Russian attack. And after that point we would have no GPS. So we designed the vbat to start with GPS on the ground, do its thing, and at 200ft, transition to an alternative navigation solution. When we went to go fly, GPS was lost at about 2ft, a bit earlier than we expected. And so the aircraft takes off and it just goes the wrong direction. And it flew for about 60km just in the wrong direction. It was like a scene straight out of the the movie Interstellar with the Ukrainians and some of our team hanging.

[A]

Out the side when Matthew McConaughey is chasing the drone in the cornfield. This is like you. You're Matthew McConaughey?

[B]

Yeah, not me, but some people on our team and they're on a laptop trying to figure out where this airplane is. And eventually they found it orbiting 60km away and they were able to bring it down and land in a sunflower field, actually not a cornfield. Oh, wow. But the team was able to change the software in the span of 24 hours, replan the operation 24 hours later, and then execute the mission that became very successful. And since then, I think we've done over 170 sorties with the aircraft in the conflict. Zero of those sorties have had GPS at any point in the mission.

[A]

I mean, I think this is the story, I think is a lovely crystallization of what I'm beginning to understand as like one of the defining cultural traits of Shield AI, which is this just stubbornness towards the mission. Because, you know, your first time getting involved In Ukraine in 2022, at the outset of the full scale invasion, you really have a really tough experience, you know, meeting the reality of that conflict. You, you re engage now with the souped up, up vbat in August 2024, once again, you know, have a pretty fall in your face moment, and you get right back up and then you finally, you know, reach this culmination of 170 sorties and that really exquisite example of blowing up. Well, helping blow up, I should say $100 million worth of Russian air defenses. I mean that's, that's a pretty amazing vindication of being stubborn for both you and your brother and the entire SHIELD AI team. I want to ask though this part of that story that I just hope everyone in the audience understands how astonishing what you said is changing the software on an aircraft in 24 hours. Because the traditional ethos of the Department of Defense for anything that's flight critical software is once it works, for the love of God, never touch it again. Because getting through flight certification is this incredibly painful ordeal. And so once you have something that's past that flight certification, you never want to touch it again. But what you're encountering in Ukraine is an entirely different approach to software in a national security context, which is the only way to be relevant, the only way to survive is to embrace rapid iteration, rapid turnaround times and rapid change. What was that like?

[B]

Well, one, I just want to double tap the, the difference between going home and never having any impact at all and doing now 170 sorties, many of which that have culminated in successful strikes on strategic assets, was the ability to, within a short time frame, turn around the software to make it mission relevant and to change fundamental assumptions about it and push it forward. I believe that future battlefields in large part will be defined by the agility of the software that's behind our forces, that enables them to operate effectively as conditions are fast evolving. And yes, we have programs where to change software can take the better part of a year. Like that's not an exaggeration. If you wanted to change flight software, it can take a year.

[A]

And that's a mostly a regulatory and process story. Right.

[B]

And I think it's, it's just where the risk rheostat is set.

[A]

Yeah, right.

[B]

Crashing a plane is of course a very big deal. When it's vbat is, you know, like a million dollar airplane by. So just on, just for context of the audience where you set the riskier stat. But there's the practice like you play practice like you fight. I think that there's a real question that, look, in the Ukraine they're in duress. Their risk reestat is set differently. Accepting software Updates on a 24 hour basis makes a lot of sense given that situation. The United States in some programs, and you know, I can't speak for all programs but has a very conservative risk reestent. So then there's a question of, well, is it the best thing to sit at this very conservative dial when we're not rehearsing what the future could look like? Right. Because I don't know anybody in industry that's truly practicing 24 hours or 48 hours from software update to combat deployment.

[A]

Right.

[B]

And if you've never done it, is it reasonable to expect that if we find ourselves in extremis, we will be highly effective at suddenly pivoting to a very agile motion of development test deployment?

[A]

Right. And it's not that you're abandoning testing before you deploy, it's not that you're giving up on the canonical things of aircraft development and responsible development of aircraft. It's just that you're saying this has to be done in a relevant time frame.

[B]

Yeah. It has to be done at the speed of relevance. And basically I have sort of three pillars that I think are essential for winning an autonomy. Basically, the industry government team has to find a way to sit at the intersection of performance. Right. It has to do the mission and has to do it well assurance. Right. It has to be airworthy, it has to have the right cyber hardening and speed. You have to be able to deliver one and two in a time frame of relevance. What's that time frame? I would argue what we should be aiming for is you can get through that cycle in 24 hours. And I think there's a lot of work to go to get it to a 24 hour cycle. But I think if you have, and I think this is where just autonomy in general needs a lot of work. It's not a well industrialized process compared to many other industries. A lot of autonomy programs, they're brilliant PhDs and people like, it's a pretty young industry and you've got these collections of super smart people to sort of bootstrap every new project from the grounds up. But you don't have this foundational development infrastructure that you have, say for mobile app development. Right. Where if I want to develop a mobile app for Android or iOS, they've put guardrails on the process to, to enable me or anyone to go 100 times faster and be a hundred times more effective than was possible 15 years ago.

[A]

Right.

[B]

And so this happens in industries through time. It has not yet happened in the autonomy space where the infrastructure has been put in place that allows developers to sit at the intersection of performance assurance and speed.

[A]

Yeah. And you know, we talked about Diu and you partnering with Diu in the early stages of the company. Are there any parts of the DoD that stand out to you as maybe not at the, the speed of Ukraine, but as effectively drawing lessons learned from Ukraine and incorporating them into a changed operational approach?

[B]

I'd say that some of the, the, the, the, the big takeaways have been, I think, positive, which is if it doesn't work in Ukraine, it's probably not going to work, you know, anywhere else. Just on the count of if you can't operate without GPS and if you can't operate with severely degraded communications, I think a lot of people are starting to see that, I think correctly, they do not pass GO capability.

[A]

And just, just to your point, right, I mean, I think there's obviously the, the hypothetical of what if we're in a fight with China over the Taiwan Strait? I would expect China is going to have some formidable electronic warfare capabilities. But even when you're not fighting against a near peer adversary or a peer adversary, in the recent conflict against the Houthis, we've had to re examine fundamental assumptions about the nature of American air dominance as we experienced it in Iraq and Afghanistan. Right. We lost reapers at a significantly higher than typical attrition rate just in this most recent conflict against the Houthis because suddenly ground based air defenses are within the reach of an actor like the Houthis, which don't have the world's biggest defense budget or anything like that. And so the urgency of learning these lessons from Ukraine, I hope is being very deeply understood by the leadership in the dod. And so sorry, I've interrupted you. Please keep going.

[B]

Well, I think that we talked about ew, I think that there's another pillar there which has integrated air defense systems are more capable and less expensive or let's just call it air defense systems because some of them might not even be that well integrated in the case of the Houthis have become much more effective. And so these are very low cost Iranian surface to air missiles that are shooting down very expensive aircraft. And certainly you see some of that same capability in Ukraine as well. So that's been another, you know, I think wake up call in addition to the EW environment. And the third one that I'll call out is, is, is a lot is written about mass. So a lot of people, you know, and, and I think it's another great takeaway. Like clearly the world is going to involve or conflicts in the future will involve much more mass than has been historically the case, much more use of drones. I don't know what the latest numbers are. Do you know what the latest numbers are for Ukrainian production?

[A]

Well over a million and a half is what they're targeting for annual production.

[B]

Well over a million and a half per year. So just st numbers of drones yet.

[A]

I mean, you think about Predators or Reapers, we're talking. I don't even know if they made it to more than a thousand. I mean, it was probably hundreds of aircraft.

[B]

Yeah.

[A]

Not hundreds of thousands.

[B]

And yet despite that production, these lines don't really move.

[A]

Right.

[B]

And there's no, you know, victory for either side. And so I think an important question is how can you send a million and a half drones downrange per year and not get effect? Right. And so I think another reminder, because I think certainly a lot of people know this, but I think they are having some effect.

[A]

Right. There's over a million Russian dead or wounded in this conflict, and Ukraine has not fallen. That's an effect.

[B]

Yeah, I think that's absolutely fair. But I think that there is a big, I think, targeting challenge in the world. I mean, the world is big. You can't just fire a million and a half things downrange and, and hope for effect. They have to be well placed. And a gap that we have seen has just been the ability to get these drones to have the effects that is intended for, for basically information gaps in terms of, you know, where to send them and what effect did they have because it's been so hard to get eyes and ears deep across the, the forward line of troops.

[A]

And so as we connecting it back to the, to the dod, Right. You talked about the big lessons learned about what, what is the exchange ratio of how much does our drone aircraft cost, how much does their surface to air missile cost. You talked about the importance of mass and how we're like multiple orders of magnitude away in production from where we probably need to be. As a Department of Defense, are you, are you seeing signs of movement in the DoD that are encouraging? I think about, for example, the collaborative combat aircraft program, which is putting in flight autonomy in a formal program of record with a multi billion dollar budget. I think about the Biden administration's Replicator initiative, which was looking specifically at that affordable mass and especially equipped with autonomy type capabilities. How have these moves, in your estimation, affected the DoD's position and what else needs to be done?

[B]

Look, I think there's been substantial influence and some of the initiatives that you called out, I think, are some things that were already happening. Maybe some of them were adjusted and informed by what's happening. In Ukraine. But certainly I think that there's a lot of positive momentum around updated technologies, platforms and programs informed by what's happening in Ukraine. I also think there have been some call outs to make sure that we don't learn the wrong lessons from Ukraine. And I think that's absolutely a great call out that some people are making because you can't just wholesale take what's happening here and expect it to translate to other theaters.

[A]

Yeah. I mean, a ground war is different than a naval war to a first approximation, if we're thinking about implications for China. So. So it's important that we calibrate as we draw lessons learned. So as you're thinking about SHIELD AI and where you are in this moment, you have the vbat, which is this exciting $1 million platform that can be airborne, I think you said, for 12 hours, which is for a million dollars. That's a nice capability.

[B]

Right.

[A]

I think a Reaper is somewhere in the 24 to 36 hour range and it's like $25 million. So it's a really interesting sort of return on investment kind of capability that you bring. There's also this autonomy part of the story. And I'm curious, there's a lot of discussion out there and has been for well over a decade. Right. Think about when Google originally pulled out from Project Maven about bringing AI into a war fighting context and whether or not that's a desirable thing that America should be engaged in. Where are we now in terms of the state of AI and autonomous capabilities and where should we be trying to go?

[B]

I'll start with the end state, which I think is just mass deployment of hyper intelligent AI on every vehicle or weapon in the arsenal. In terms of where we are today, I think frankly there's just a lot of work to be done to claw back to, let's just call it table stakes functionality. Right. GPS existed for a long time. It's this magical, brilliant sensor that tells you where you are in the world and tells you what time it is. Precisely. No questions asked for a long time. And so much was built on that infrastructure. Today, that infrastructure cannot be trusted in large swaths of the world. And so a lot of the energy for AI and autonomy needs to be applied to building back to the base case of knowing where you are in the world. And so sometimes people wonder if you can do all this great selfless swarming or pick your capability. Why don't we see that right now? And it's just a lot of effort has to go back to creating a Stand in for gps. And AI is a great foundational technology to, you know, figure out where you are.

[A]

Yeah. And this is something that I've talked about with your colleagues, which I think is just a helpful illustration. You know, talk to me about the difference between swarming in like the Chinese Lunar New Year demonstration, which people maybe see a YouTube video of and see like, oh my gosh, China already has swarming technology. Talk to me about the difference between that kind of swarming and battlefield relevant swarming capabilities.

[B]

Yeah. So it's just that you have, you either have GPS or you don't. And if you have GPS, putting up 10,000 drones to do these pre programmed displays becomes relatively well, becomes dramatically more accessible. And so as, and it's all technologically possible, we're doing it, but in terms of what's actually deployed, there's very little that has weaned itself fully off of.

[A]

Yeah, I think just to crystallize it. Right. There's a difference between have GPS tell you where you are, then fly up 10ft. Right. 10ft back three feet. You know, that pre programmed sequence of instructions that when you look at it make this beautiful display that appears highly, highly coordinated. There's a difference between that and we're on a battlefield, there's 100 drones. Literally none of them can effectively talk to each other. And yet they all have to work like a team by watching each other, by watching the environment and making intelligent decisions. And that's kind of the gap that you're trying to cross.

[B]

Yes. And I will say that, you know, the state of technology is such that pretty quickly we will have effective stand ins for gps. Of course, like SHIELD AI is running a lot of sorties. Those technologies are becoming more accessible, easier to deploy. And once you have that, can you.

[A]

Talk a little bit about like what, what is effective in a GPS and denied environment? You know, whether that use as AI or otherwise.

[B]

Yeah. So the, the, the name in the game is first you need observability. And then you know, you, you need to combine that information to create some estimate of where you are. So people will use cameras, they'll use radio frequency, they'll use thermal cameras, STAR cameras, they'll use inertial measurement units.

[A]

When you have a camera, you're talking about like taking a picture of the ground. Yep. And being like comparing it to some map database you have. Correct. And saying like, like if I see the Washington Monument, then I must be near Washington D.C. or whatever the Ukrainian battlefield equivalent of that is. Correct. And that's, that's Leveraging AI computer vision.

[B]

It is. And so that's one event, and then it, and then it turns out that on cloudy days or at night or over the ocean, that won't work. And there's sort of like eight big conditions. And a camera looking to the ground works in one of the eight conditions.

[A]

Got it.

[B]

So. So where I think things will fast go and are fast going is building from that foundation to enable the deployment of very large teams of machines, whether vehicles or weapons, to produce coordinated effects and outcomes. And really the only thing I think that has been slowing down the deployment of that has been again, getting back to basics with the where am I in the world? Question. But I think we're well on our way there and then it's just a matter of, of taking that next step. But I do think that one of the major gaps is a lack of industrialization of the overall process. So even if it's technologically within reach, is it actually scalable if it requires a bunch of PhDs in AI and robotics to field the capability, because there are only so many of those people in the world. And so how do you take something that for the longest time has required very high end expertise and make it accessible to the defense industrial base so that, you know, we can deploy it not once, but across the number of vehicles and weapons that the mission demands?

[A]

Yeah. So we've talked a lot about how the DoD needs to change in terms of, you know, what types of technologies it needs to prioritize. You talk about mas, you talked about operating GPS denied environments, talked about prioritizing AI and autonomy. I want to talk about, you know, programmatically and regulatorily, you know, what is the difference between the DoD many times saying, you know, a software update takes a year to get it to the, through the regulatory approval process, and in Ukraine it might take a day or a week. And I kind of want to frame this just in a. As a startup founder who bet on doing work for national security well before it was fashionable, well before it was cool, well before it was considered anything remotely approaching a safe investment of time or money, what has made your life hard, right, as a defense entrepreneur trying to work with the dod. And I mean that both in a, you know, getting on contract. And I also mean that in terms of, like, what makes it hard to build the right thing? You know, what are the pressures that you face? And where have you seen the DoD improve and where could they still improve?

[B]

Yeah, so number one thing that usually makes an entrepreneur's life hard is their own dumb Decisions. So let's just, so, like, we'll just ignore that one and, you know, put it in a box. But that's, that's number one on the list.

[A]

From the outside, right. It looks like you're batting a thousand. You might not necessarily be batting a thousand.

[B]

But, but, but that, that aside, I, I think that the, it's the complexity of the government acquisition system. It's just a huge machine with so many stakeholders. This morning we were mapping out some program capture activities, and I was looking at the different work streams. And even today, 10 years later, I'm just thinking to myself, it's shocking how many people. Right. Are on the chessboard who can say no. Yeah. And so getting alignment across so many people, and I think it's just very hard for a startup company to comprehend or properly prepare yourself for the number of stakeholders that have a vote.

[A]

Right. And I think to compare that with Ukraine, the way they approach acquisition of commercial drones as they have a special pot of money. And when you're buying drones with that pot of money, you face so many fewer regulatory challenges and so many fewer process steps to the point where even individual battlefield commanders can be engaged in the procurement of actual weapons systems. In the US System, you have what's called different colors of money. There is procurement money, there is research money, there is operations and maintenance money. And different types of organizations are only allowed to have certain types of money. So a combatant command, the people who are actually engaged in the fighting or engaged in the deterrence activity only have operations and maintenance type money. So they're not allowed to go say, I want to go buy this stuff. If they want more weapon systems, they have to ask another DoD organization, you know, please generate a requirement to tell somebody else to buy this for me. And by the time that game of telephone snail mail completes, it's nowhere near the speed of operational relevance. And for an organization such as yourself, you've had to master this skill set of identifying the million stakeholders, whether it's the operational users, the requirement setters, or the acquisition community. You got to be engaged simultaneously with all of those people. And as a, a startup, that is a really formidable skill set to ask, you know, an entrepreneur in your case. Right. The team started with three people.

[B]

Yeah.

[A]

To, to try and do that work. And you, you got an assist from diu. But why are we asking, you know, why are we making it so hard when we say so? Like, five different presidential administrations in a row have said they want to make it easier to do business with the DoD, but at least to me it still seems like it's a formidable challenge.

[B]

Yeah. I think that any entrepreneur that's thinking about getting into this space and I encourage it because I think the mission is worth it. I wake up every day very proud to have an opportunity to contribute, but you need to be prepared for some complexity. One of the things that I'm excited about, my brother testified, gave congressional testimony on this I think last year. But as doing a more problems based acquisition as opposed to requirements based acquisition, I think that's a great if, if the department can find a way to lean into that, I think there's a huge time savings and unlock there. Right. I might have a problem and I don't know what the answer is, but it's usually much quicker to identify you have a problem than specify the solution to that problem. And that's where I think, where there's an amazing opportunity to leverage industry and tech companies and just say, I've got this problem and I can acquire against this problem. How would you solve it? Right. And I think that could be something that cuts considerable amounts of time out of the overall acquisition system.

[A]

Yeah. I mean there's such a difference between what you just described and the typical requirements setting process for the DoD, which is we will buy a system with all of these parameters and it must, you might even get to the level of it and it will use this engine and it will use, use this radar. Right. Like that's, that's sometimes how detailed these requirement specifications can go. And that's so different from the way still and, and that's so different from what Ukraine, you know, often has to do, which is be like we're suffering this, you know, very vicious problem and we urgently need a solution. Tell me if you've got one and if it's great, I'll buy it. And that problems based mindset, if you can have it, is, is really powerful.

[B]

Yeah. I think in, in, in and luck think I, I, I'm not an expert on Ukrainian acquisition by, by any stretch. I understand it's got all sorts of its own problems, but in terms of our own experience and, and the way that, you know, it happened for us, it, it's just very evidence based.

[A]

Yeah.

[B]

They said, you know, come out and everybody has given us PowerPoints, so those just don't fly here. No pun intended. Yeah. So they took us out to their EW range where they just slam vbat for a couple days and they're like, okay, great. And like, but that's not enough. Right. We need to see it overfly, you know, the rushing territory, whatever. And so we, you know, trained their operators and they overflew the, the flot. They did the thing that was written about in the Wall Street Journal. And it was like at that point that we started to get momentum. And even after that. And he said, We've done 170 sorties there or maybe more now. Now it's just, it's a very evidence based acquisition system, which I think works very well up until a point. I think that there are some. There are certainly programs that would not be the right. It's pretty, pretty hard to show the evidence for.

[A]

Yeah. Okay, we're coming to the end of this. I'm really grateful for the time you've taken. I want to ask one final question. Is we've talked about, you know, who Shield AI was in 2015, who they were in 2017, who they were in 2020, who they are now in 2025. Where's Shield AI going to be in 2030?

[B]

Look, I hope we're making mission impact at tremendous scale, still grounded in the values of honor, service and excellence. I think that autonomy is foundational to the future of America and we're just humbled to have the opportunity to contribute to our it.

[A]

Ryan sang. Thank you so much for coming to csis.

[B]

Awesome. Thank you so much.

[A]

Thanks for listening to this week's episode of the AI Policy podcast. If you like what you heard, there's an easy way for you to help us. Please give us a five star review on your favorite podcast platform and subscribe and tell your friends. It really helps spread the word. This podcast was produced by Sarah Baker, Isaac Goldston and Sadie McCullough. See you next time.