Summary

Cheap drones have overturned the economics of force: a system costing a few thousand dollars can take out one worth several million. The host argues that the United States should consider deploying “100 times more systems,” shifting some procurement from legacy exquisite platforms toward tens of thousands—or millions—of distributed drones.
America cannot quickly match China’s “fearsome” industrial capacity, but the speakers see a path through intelligent mass. Ryan Tseng distinguishes mass from effect: hardware matters only if software, AI, and autonomy can “find, fix, and finish targets at scale,” making every minute of flight and ounce of charge count.
Current U.S. military programs are roughly three orders of magnitude below demonstrated battlefield demand. Adam Bry says Ukraine uses small quadcopters at “millions per year,” while comparable U.S. programs operate at single-digit thousands; redirecting military purchasing toward this category could equip troops and rebuild a dual-use industrial base spanning defense, public safety, and infrastructure.
Chinese drone light shows do not prove battlefield swarm superiority because they depend completely on GPS and continuous communications. Those assumptions are vulnerable to jamming; the strategically relevant race is deploying highly autonomous, AI-driven systems that sense and navigate in contested environments—a domain Bry says “is still up for grabs.”
Electronic warfare turns drone competition into a continuous software-deployment race. Tseng recounts that Shield AI rewrote its V-BAT navigation stack in 24 hours after an aircraft lost GPS three feet above the ground; against software-defined electronic warfare, government update cycles of up to a year—and sometimes longer—are unacceptable.
Human control is not a binary choice, and autonomy can sometimes reduce destruction rather than expand it. Bry says human-machine teams are currently more effective than machine-only teams, while Tseng relays a senior military leader’s view that defensive forces may increasingly “flip the switch and go full auto” for reaction speed. Simple assurances that humans will always remain in the loop are therefore inadequate.
The decisive strategic call is that American AI leadership cannot be optional. Bry treats autonomous weapons as morally grave, while both speakers frame the game theory starkly: the only outcome worse than both sides possessing a powerful technology is “one where only your adversary does.” Bry’s founder prescription follows: the mission is foundational, the work is brutally hard, and builders must “embrace the struggle.”

Deep dive

1. Two unlikely paths converged on autonomous defense

Ryan Tseng had started and sold a company to Qualcomm but lost his “fire in the belly.” He wanted his next 50 years to combine “a noble mission,” extraordinary colleagues, and “a chance to define the possible”; his Navy SEAL brother supplied the mission by arguing that autonomous systems could protect service members and underpin future American military dominance.
Tseng initially called the business “stupid” and advised his brother to find a better commercial route to mission impact. The SEAL’s persistence changed his mind: as Tseng investigated, he was shocked by both “the scope of the problem” and how little anyone was doing about it.
Adam Bry arrived through technology rather than mission, having flown radio-controlled aircraft since childhood and studied autonomy at MIT. Skydio’s founding bet was that AI embedded in a small, light quadcopter would remove the expert-pilot bottleneck; a consumer launch created the integrated platform later adapted for enterprise and government, including the Army’s Short-Range Reconnaissance program, which began around 2018–2019 and was officially announced in 2021.

2. Drones replaced concentrated force with distributed asymmetry

Matt Cronin framed the battlefield shift as a move from concentrated, exquisite assets toward distributed, decentralized systems. He described drones launched from trucks potentially reaching 1,000 nautical miles and urged the United States to contemplate “100 times more systems” to make forces more lethal while helping service members return safely.
Bry puts the economic asymmetry plainly: “A system that costs a few thousand can take out a system that costs a few million dollars.” Ukraine, driven by necessity and “incredibly impressive ingenuity,” is rapidly iterating aerial, ground, and maritime drones; he does not think the U.S. military has fully absorbed what that implies for its traditional preference for larger systems.
The host’s sharp example is a commercial quadcopter attacking a tank, with two or three munition-laden drones potentially destroying a large armored vehicle. Bry’s warning is not that America shares Ukraine’s constraints, but that adversaries will exploit the same low-cost surveillance and strike possibilities whether U.S. procurement adapts or not.
Tseng’s qualification is that quantity alone does not create victory. World War I, World War II, Vietnam, and now Ukraine show that mass can still produce an attritional stalemate because “the world is big and targets are relatively small”; the operational divide is therefore between mass and effect.

3. China has manufacturing scale, while America’s edge is intelligence

Bry traces China’s advantage to a seemingly benign supply chain: radio-controlled aircraft were already made there in the 1990s and 2000s, and modern drones combine that hardware with phone-like consumer electronics. Offshoring accumulated not merely lower costs but an ecosystem of machining, molding, circuit-board fabrication, component placement, and skilled production.
His rule of thumb is memorable: “Wherever they’re building iPhones, they’re going to have a really good ecosystem for building drones.” Skydio has manufactured domestically since 2016, initially feeling like it was “swimming upstream into a fast-flowing river”; Bry sees policy tailwinds now but admits restoring that manufacturing depth may not be attainable.
Tseng calls China’s industrial capacity “fearsome” and doubts the gap can be closed in one year—or even a decade. His “Max Q moment” is the next decade, when U.S. force transformation, AI advances, China’s wealth, and its military investment all collide; reindustrialization is necessary, but software and autonomy are the plausible competitive multiplier.
Tseng expects a second wave beyond Ukraine’s largely one-operator-to-one-drone model. Advanced onboard autonomy and machine-to-machine communication may prove even more consequential than merely removing the pilot, so “winning on the AI front” matters more even though hardware scale cannot be ignored. He argues that software updates could close the OODA loop and make every system more effective.

4. Procurement scale is itself an industrial policy lever

Tseng’s policy recommendation starts with continued investment in the service members who use these systems and with mobilizing the bureaucracy to operationalize autonomy technologies already available. He says Ukrainians use soldier-carried quadcopters at “millions per year,” while Bry says comparable U.S. programs operate at single-digit thousands—a gap of three orders of magnitude. Bry also points to Israel’s rapid adoption of similar systems.
Civilian and consumer quadcopters constitute a single-digit-billions market, roughly comparable, in Bry’s view, to what the military should spend on the category. Defense demand at that level could sustain production serving public safety and critical-infrastructure inspection because the underlying technologies overlap heavily.
Cronin suggested adding a zero to procurement quantities. Bry responded that some legacy spending may need to be deleted—sometimes by moving the decimal point. The real trade-off is one or two exquisite systems costing hundreds of millions or billions against 10,000, 100,000, or one million AI-driven drones—not because drones replace everything, but because they are more powerful in many future situations.

5. Battlefield swarms are an autonomy problem, not a light show

Matt Cronin tests the China thesis with light shows over Shenzhen and other cities, including displays involving hundreds of thousands of drones that broke Guinness World Records. Bry’s pushback is that “swarm” hides the task: light shows are 100% dependent on GPS and continuous communications, both vulnerable to disruption and therefore “basically irrelevant on the modern battlefield.”
The displays demonstrate the ability to build a large number of drones and get them into the air, which is one part of the equation, but not contested autonomy. Bry argues Skydio’s advantage over DJI in open civilian markets is its ability to sense surroundings, react in real time, and automate complex missions—evidence, in his telling, that highly autonomous AI drones remain “a domain that we can win in.”
Shield AI’s first decade pursued “the world’s best AI pilot,” eventually reaching work on F-16s. Tseng says its next decade is about contributing that pilot to the broader industrial base so other American manufacturers can deploy sophisticated vehicles at scale; intelligence is what turns mass from an attritional pile of hardware into operational effect.

6. Ukraine exposed both product failure and procurement failure

Bry’s candid admission: Skydio’s first systems in Ukraine “basically failed.” They reflected U.S. Army requirements in which electronic warfare was absent; repeated trips and painful field lessons then drove years of radio and vision-based navigation development more directly than formal military requirements did.
Skydio had bet on computer vision in 2014, giving its drones onboard cameras for localization and environmental understanding without native GPS dependence. Bry says the resulting systems now perform in harsh, communications-contested and GPS-denied environments. Tseng adds that electronic-warfare testing is becoming part of military purchasing evaluations.
Tseng’s V-BAT example makes adaptability tangible. The 12-foot-tall, 12-foot-wingspan aircraft, capable of roughly 12 hours’ flight, was expected to lose GPS at 200 feet; instead, Ukrainian and Russian jamming removed it at three feet, sending the aircraft the wrong way until a truck team chased it for two hours and found it orbiting a cornfield roughly 80 kilometers away.
Shield AI rearchitected the navigation stack within 24 hours so the aircraft never needed GPS, validated it in Texas, then returned to demonstrate a mission whose outcome was a Russian SA-11 being “found, fixed, and finished” by HIMARS. Tseng’s lesson: “It will be the pace of deployment that is the make or break,” yet government software changes can take up to a year, and he says his company has sometimes taken two years to push new software into deployed systems.

7. Autonomy makes war a software fight—and ethics a moving boundary

Tseng’s synthesis is that autonomous conflict becomes “basically like a software-writing fight.” The behavior and capability of autonomous systems, along with the behavior of electronic-warfare systems, are software-defined, so the side able to write and deploy updates in a day gains an advantage unavailable to an institution that requires years.
Bry first insists that “this is terrible stuff”: weapons kill people and create immense suffering, so deterrence, precision, and minimizing harm remain the goal. Human judgment in the use of force is important, but the status quo may mean dropping a 500- or 2,000-pound bomb; intense autonomy that precisely identifies a target might cause far less collateral damage. He also says the military has robust controls and people thinking deeply about how authority is delegated.
Bry says human-machine teams are currently far more effective than machine-only teams and may remain so for the next several years, perhaps longer. Tseng relays a senior military leader’s framing that defensive pressure encourages machine control for reaction speed; the Phalanx ship-defense system already kills anything entering its weapon engagement zone once activated.
Bry’s historical analogy is a World War II bomb: once released, it was “autonomous,” governed by gravity, wind, and physics, with no further human correction. AI does not invent relinquished control; it potentially makes that irreversible act more precise.

8. The strategic answer is leadership, and the founder test is endurance

Bry says the game theory is stark: autonomous weapons may be terrible, as nuclear weapons are, but “the only world worse” than mutual possession is one where only the adversary has them. He acknowledges uncertainty about how machine-only operations will develop and offers a “strong conviction that America needs to lead.”
Bry calls AI “the most important technology really in the history of humanity” and treats adversary dominance as unacceptable. Tseng warns that blanket limits on parameters or compute would be “music to your ears” in Beijing or Moscow; he says safety deserves thought, but strategic competition cannot be excluded from regulatory reasoning.
Bry’s message to founders is that security and stability underpin everything else worth caring about, making the mission worth the effort. He adds the operating reality: critical hardware is hard in different ways across days, months, and years, and anyone seeking genuine impact must “embrace the struggle.”

SF Compute: Commoditizing Compute to solve the GPU Bubble forever Big Tech's Tariff Chaos + A.I. 2027 + Llama Drama