Watch any counter-drone demonstration and it ends the same way. A drone drops out of the sky, the system that killed it takes the credit, and the footage makes the problem look solved. Microwaves frying swarms, lasers burning wings off, interceptors running a target down, chain guns walking rounds onto a moving quad, a high speed drone interceptor chasing down a fixed wing flying overhead. The kill is the easy thing to film and the easy thing to sell, and it is most of what the counter-drone conversation is about right now.
What many of these demos fail to show is the part that actually decides the fight. Someone already knew the drone was coming - they knew the bearing, likely the altitude, the target count, and roughly when to expect it to cross their path. The sensor was cued, the operator was looking the right way, and the effector was pointed in the right direction before the first target lifted off. If we strip those conditions away there is a strong likelihood that the same system that looked unbeatable on the range never gets a shot, because it never sees the target in time. You can't kill what you can't see, and seeing is the part nobody is filming.
The Half Nobody Films
Counter-UAS is two sides of a coin. Sensors find, identify, and track the threat, whereas effectors bring it down. The effector side is definitely sexier, and tends to get the money, the headlines, and the demo videos. And honestly, it deserves that attention, because the c-UAS ecosystem requires that effector, whether electronic warfare, microwave, kinetic with or without a warhead, a bullet, a laser, or a net, to be capable of taking down a drone. The FY26 NDAA budget reflects that, with roughly $3.19 billion in c-UAS funding and the Air Force alone quintupling its Air Base Air Defense line to $836 million. But nearly all effectors do nothing until a sensor hands off a track good enough to provide targeting data. The sensing half is equally hard, less photogenic, works behind the scenes, and decides whether the effectors ever fire.
In practical terms, there is not a single sensor that covers the full range of threats in the c-UAS environment. A THAAD radar built to catch ballistic missiles was never built to find a Group 1 quadcopter, let alone target Group 3 Shahed-136 one-way attack drones hitting in waves. Multi-mission radars like those behind Israel's Iron Dome track multiple simultaneous targets and cue interceptors, yet low and slow Group 1-2 drones have repeatedly slipped past them. And even when the track holds, the math is backwards: interceptors costing tens of thousands to millions of dollars against quads costing three thousand, fired from a finite magazine that runs dry long before the attacker's production line does. Radars tuned for small, slow targets are fighting through birds, cars, and ground clutter to hold tracks, and are often limited to shorter ranges against smaller targets. RADA's RPS-42, one of the most widely fielded counter-UAS radars and the sensor inside the Marine Corps' LMADIS, detects Group 1 and 2 drones out to roughly 10 km and out to 30 km for conventional aircraft. Ten km sounds generous until you run the clock with an attack quad closing at 20-30 m/s that can cross that distance in a matter of minutes. Real-world detection against small airframes often happens well inside the advertised sensor maximum, limiting c-UAS reaction time. The radars built for smaller drone threats are not the same radars built for Group 3 fixed-wing, which are not the radars built for ballistic threats, so true coverage means stacking different systems for different slices of the same sky.
Radio frequency detection fills part of the sensor gap, and in Ukraine it carries much of the load. In many online videos from both the Ukrainian and Russian sides, soldiers use handheld scanners that read an FPV's video downlink, letting them watch live and figure out what the drone is hunting (often they are the target themselves, since these scanners are typically short range). However, those RF scanners only work against a drone that transmits in the RF spectrum. Fiber-optic drones fly on a spool of glass and have no RF emissions, so RF detection never sees them and jamming has nothing to grab. The fiber-optics threat design erases an entire sensor channel and an entire effector class in the same stroke, and those drones now make up a meaningful and growing share of strikes on the Ukrainian front. AI-enabled and autonomous drones that fly to the target on their own optics produce the same challenge to c-UAS teams. Breaking the communication link between vehicle and operator has traditionally been a primary means of taking down these threats, but the rapid evolution of AI is bringing online entirely new classes of drone control. These drones are not unstoppable, just changing the game a bit. Radar, optical, and acoustic still find them, and a bullet or a microwave pulse can still kill them. The takeaway is simple: a drone built without an RF link takes RF detection out of the fight entirely, and a large share of the fielded counter-drone base was built in an era when every target emitted.
Now Add Thirty More Targets
Everything I am talking about above is the problem with just a handful of drones, but that is not the threat we will be facing in the near future. A sensing picture that struggles to hold clean tracks on a handful of contacts is going to face thirty or forty, arriving together, some emitting and some dark, some real and some decoys, appearing on different sensors at different ranges with different confidence levels. Correlating that into one picture, sorting threat from bird from decoy, and assigning effectors before the clock runs out is not the single-target problem run thirty times. We are facing an entirely different problem here, and it is the one the demos aren't targeting.
Epirus' recent and much hyped microwave demonstration that downed a 49-drone swarm with one pulse is the cleanest example. Yes, the capability is real and valuable against non-EM hardened drones. However, given my own experience in testing equipment like this, I would be willing to bet a Starbucks latte that the Epirus crew knew the swarm was coming, knew where it would be, and had the system oriented before hitting send on the microwave pulse. The effector in this case was never the hard part. Knowing the swarm is coming, where it is coming from, knowing hostile intent, with time to get effectors into action is the entire c-UAS fight - in these demonstrations we are giving the defenders a cheat sheet in advance.
Take Away the Script
Traditional test runs are on a known threat axis. The drones come from over there, the sensors face that way, and the engagement plays out along a line everyone agreed to beforehand, target drones are slowed down to ensure the sensor can get a lock on them, etc. No real adversary drone gives us that courtesy in the actual fight. A forward operating base, a port, a supply node, a stadium can be hit through a 360-degree arc, low and fast on one bearing while something slower comes in high on another. A sensor sees a cone. Covering the whole sky means stitching panels and systems together with no seam an attacker can slip through, and the attacker's counter costs nothing: come from where the good sensor is not pointed.
This is not a thought experiment. On March 1, six soldiers were killed and dozens more wounded when an Iranian one-way attack drone hit a U.S. operations center at Port Shuaiba, Kuwait, flying in low and slow enough that air defense never engaged and no siren sounded. Their site had T-walls, adequate protection carried forward from two decades of GWOT-era ground fights; nothing overhead, no counter-rocket-artillery-mortar coverage, and by internal accounts almost no drone defeat capability despite requests. Personnel had sheltered in a bunker during missile intercepts that morning, got the all-clear, and went back to work. The drone arrived after the picture said clear. In the week before the strike, soldiers had spotted quadcopter reconnaissance over the site, and an Army memo later concluded Iranian intelligence had tracked the unit's move there. The adversary surveyed the coverage, found a seam, and flew one drone through it.
The official line called that penetration an outlier. The Army's own response suggests the threat is anything but: within five days of the war starting, it surged 10,000 Merops interceptor drones to the theater. Surging effector assets into theater unfortunately does not close the gap that got U.S. soldiers killed, because we aren't launching interceptors at targets nobody saw. The failure was upstream of every effector in the inventory.
From an investment perspective, our focus cannot remain dedicated to single domain c-UAS, as the threat is evolving too rapidly to look only in one direction. Russia is launching fiber-optic FPV drones from inside the hull of an USV, in serial production by its maker's account. Ground robots carry munitions on both sides of the Ukraine front. All of the parts and pieces enabling an autonomous strike against an over-the-horizon target, hitting simultaneously by air, surface, and ground, exist and are in production today. Our radar watching the sky for hypersonic targets is not looking for the USV entering the harbor, and the camera on the perimeter looking for intruders probably won't catch the quad coming over the back wall. Each domain's sensors were bought separately by a combatant commander, under different contracts and programs, different colors of money, and each dedicated to watching their own slice. Port Shuaiba was one drone, one axis, one domain, against a fixed site with exploitable gaps and seams. The version of this story where an adversary places forty drones from multiple domains across a full arc is not a hypothetical war-game situation. It is now an assembly challenge and some time with an AI coding agent - the only open question is the date it strikes.
The Numbers Come From the Easy Version
Having dedicated considerable time to UxS Test and Evaluation, this is what keeps me up at night. Nearly every detection range, time to detect, and intercept rate, every number a program office action officer is reading off a slide deck, was generated under known-axis, known-count, single-domain, cued-sensor conditions. Test ranges produce those conditions primarily for good reasons like safety and airspace control, but the result is a performance record that only describes how the c-UAS system under test was able to handle what is likely the easiest version of a threat. Port Shuaiba is what the gap between tested performance and combat performance looks like, and it is no longer a rounding error in the T&E world.
Commercial vendors are not to blame for this artificial construct, as many have nowhere to test honestly even if they want to. Government ranges are scheduled months out and scripted for safety, and the regulatory picture in the United States pushes serious unmanned testing somewhere else entirely. Commercial companies are flying their systems in Jordan, Poland, Germany, and Ukraine because that is where they can get realistic conditions, real threat replication, and in Ukraine's case a live enemy grading the homework. Government test programs are expanding multi-domain testing into the Middle East and Africa, taking advantage of more relaxed airspace and environmental standards. Every one of those flights is American test data generated on foreign soil, at shipping and logistics cost the program absorbs, on timelines set by foreign approvals, building test infrastructure and expertise in someone else's country. A manufacturer that wants to run its sensor against thirty mixed targets from every bearing, with no script and a red team hunting for the seam, is significantly challenged to legally conduct that test on a commercial range inside the U.S. today. Nine Mile Training Center is probably the closest a commercial company can get to that test range, but they are paying heavily for access. Until that environment exists at commercial speed, and until program offices start demanding results from unscripted conditions instead of accepting the slide, the demo numbers will keep standing in for the real ones.
Where the Fight Is Decided
Realistically, additional effectors will not fix this, and neither will additional radars. What I see fixing this is the layer that fuses many sensors across many domains into one picture that holds at scale, from every bearing, with discrimination good enough to tell a threat from a bird and a decoy from a warhead, fast enough to act on. Israel runs arguably the most mature version of it, a layered air defense where radars, EO/IR, and command systems feed a single engagement picture under live fire, and serious fusion work is underway on both sides of the Atlantic. What does not exist is the American version built for the small-UAS fight, at the scale and tempo the threat now demands. This layer is not producing a satisfying video of drones being decimated from the sky. This is also the layer where the future fight will be decided.
Venture capital has been ahead of the Department on most of this fight, funding interceptors, microwave systems, and the autonomy stack years before any program of record caught up. Effectors are now the crowded trade. Sensing, fusion, discrimination, and track management at scale sit in the middle of the kill chain where nothing demos well, so the money has mostly passed them over. Same goes for the test infrastructure needed to prove any of it honestly. Inside the Department, sensors get bought piecemeal, by threat class and by domain, on separate programs and schedules. The integrated picture tying air, surface, and ground threats into a single decision belongs to everyone's interface requirements and no single program. That seam decides whether the defense holds.
The next counter-drone video that crosses your feed will end with a drone falling out of the sky, and while it will be real, it will not tell you what you need to know. Ask what the sensor saw and how far out. Ask how many simultaneous tracks the picture held and for how long, from how many bearings, across how many domains. Ask where the test was run, and what the crew knew before the first drone launched. Six soldiers at Port Shuaiba had T-walls, a siren system that tested fine the week before, and an all-clear. The kill is the easy ten percent. Everything before the kill is the part we are underfunding and testing under conditions the enemy will never grant, and the next attack is not an if.