No Hands on the Stick

At an Army aviation panel in October 2025, Brig. Gen. Travis McIntosh described what modern frontline drone combat actually demands: right now, it takes four soldiers to launch a drone ambush. One flies the aircraft, one pulls security, one hauls the gear, and one sets up antennas. Then McIntosh, deputy commanding general of the 101st Airborne Division, reportedly offered a clear dividing line: "Let me give you a threshold that's easy to understand: when we can fly drones by command, not by pilot. When your drones can understand commander's intent—that, ladies and gentlemen, is the threshold for AI autonomy to help us."

That operational burden brings us to the question behind this episode: as militaries build autonomous battle networks, will onboard artificial intelligence transform reconnaissance drones into autonomous digital commanders directing fire across the battlefield, or do they remain tactical routing aids while humans supervise who holds the trigger? Understanding that shift matters, because the systems being designed today will dictate how nations wage high-speed combat and whether human judgment can survive machine tempo.

We will explore how unpiloted aircraft navigate blinded skies, why military doctrine rejects software commanders, and where the human supervisor fits when combat unfolds in milliseconds. The origin of this technological scramble lies in the sudden collapse of older battlefield assumptions. For two decades, military operations relied on uninhibited satellite relays and continuous live video. But against peer adversaries equipped with modern electronic warfare, that playbook fell apart.

From late 2022 onward, Russian electronic-warfare units pushed GPS noise levels across Ukrainian sectors high enough to knock out satellite locks. Ground jamming and signal spoofing broke drone links and blinded artillery spotters. A May 2023 report by the Royal United Services Institute estimated that Ukraine was losing roughly 10,000 drones a month to electronic warfare, averaging about 300 platforms lost every single day.

Edge artificial intelligence reportedly addresses this vulnerability by relocating computation from distant ground stations directly onto companion computers aboard the aircraft. Instead of acting as a passive camera beaming raw footage over fragile airwaves, the drone executes its own perception, inference, and control loops locally in ten to one hundred milliseconds, turning the aircraft into an automated target-cueing node. To see why that local processing is mandatory, consider the communications bottleneck that erupts when dozens of drones attempt to broadcast video all at once.

The traditional way to run a reconnaissance drone is to pipe continuous high-definition video straight back to an operations center. But streaming live video demands between 6 and 10 megabits per second. In a 2018 study, computer science researchers calculated that just fifty drones broadcasting high-definition feeds at once could completely saturate a 500-megabit wireless uplink. In contested airspace, trying to pump that much raw footage across contested radio waves turns the network into an immediate choke point.

Edge processing bypasses that wall by analyzing the imagery directly aboard the aircraft. Author Christopher Jenks has argued that raw video is rarely needed offboard. According to a defense engineering analysis, onboard machine-learning models can classify objects, score their importance, and discard empty sky and terrain before anything transmits. Instead of streaming gigabytes of visual noise, the aircraft sends lightweight metadata packets and event thumbnails. In an industry scenario analysis, transmitting actionable triage data instead of full video yields an estimated 20-to-1 data reduction, allowing squads to share actionable tracks over tiny fractions of bandwidth.

That filtering fundamentally reshapes tactical tempo. When data is computed at the collection point, sensor-to-shooter timelines shrink from tens of minutes down to mere seconds, as Major Nickolas Lupo noted in an Army doctrinal analysis. Ground units receive actionable cues almost instantly, feeding target coordinates directly into automated battle management software like GIS Arta to coordinate distributed artillery strikes before an adversary can relocate or electronic warfare severs connectivity.

The scale of that speed advantage surfaced during the Air Force's September 2025 DASH 2 experiment. Evaluating human-machine teaming across twenty tactical problems, participating software microservices generated potential courses of action in under ten seconds. According to Air Force public affairs reports, the automated tools produced thirty times more options than human-only teams, delivering thousands of proposed solutions in a single hour.

Compressing sensor feeds down to instant metadata solves the network bottleneck in theory. Yet finding and striking a target still depends on knowing precisely where that target sits in physical space—a calculation that falls apart the moment enemy jamming blinds a drone's satellite navigation entirely.

Stripping down the data stream only works if an aircraft actually knows where it is. By spring 2024, GPS denial was observed across flight corridors over the Baltic Sea, the Eastern Mediterranean, and the Persian Gulf. In Ukraine, Russian electronic warfare units pushed noise levels high enough to drown out satellite navigation, causing frontline drone operators to lose satellite locks routinely. For an autonomous scout, relying on external satellite signals had turned into a fatal vulnerability, echoing a documented 2011 incident where a U.S. RQ-170 drone was reportedly captured over Iran using GPS spoofing.

To keep flying through electronic interference, autonomous systems reportedly fuse their internal movement sensors with downward-looking cameras. Visual-inertial odometry tracks frame-by-frame motion, using real-time camera views to correct the short-horizon drift of onboard gyroscopes and accelerometers. For long-range positioning, terrain-relative navigation compares live landscape imagery directly against preloaded digital elevation maps or satellite databases, letting onboard models calculate coordinates without touching a satellite constellation.

According to industry analysis, dead reckoning alone accumulates errors over long flights, but continuous terrain matching can bound that drift. At Jammertest 2025, engineers evaluated that exact layered approach under live electronic warfare. When counterfeit signals were directed at a 10 kg electric vertical takeoff drone, its onboard system detected the spoofing warning, purged satellite inputs entirely, and completed its planned reconnaissance route by relying on optical template matching.

Once navigation is solved, platforms still have to share their targets across contested skies. Rather than beaming coordinates up to satellites, technical reports describe drones forming decentralized tactical mesh networks where each aircraft acts as a relay. According to industry technical analysis, passing data peer-to-peer using directional millimeter-wave radios operating from 24 to 100-plus gigahertz allows swarms to pass targeting data peer-to-peer across the area of operations without relying on satellite relays. Yet this rising technical autonomy forces the central question: does operating without human pilots turn these machines into digital commanders?

That operational independence might suggest an aircraft navigating and coordinating on its own is becoming an autonomous commander in the sky. Yet military doctrine and defense analyses directly refute that idea. Emerging software architectures function strictly as tactical edge nodes and decision aids, designed to improve data flow rather than hold command authority. Because military frameworks view sensitive operational choices as an exclusively human responsibility, command authority and the legal power to authorize lethal force remain firmly reserved for human officers.

What changes instead is the nature of human control. As Army Major General Clair Gill reportedly observed at an association meeting, the era of a drone operator acting as a hands-on pilot glued to flight sticks is fading. The traditional human-in-the-loop setup—where operators manually fly the aircraft, steer the sensor gimbal, and formulate every tactical move—is giving way to a supervisory human-on-the-loop posture. Proponents of this supervisory model explain that onboard computers handle the initial burden of triaging, prioritizing, and sequencing targeting recommendations, leaving the human supervisor to make the final operational calls.

Major armed services began actively institutionalizing that transition in autumn 2026. The Navy demonstrated its Collaborative Autonomy Mission Planning tool at the Point Mugu Sea Range to translate human battle plans into machine instructions, followed by the reported establishment of the Robotics and Autonomous Systems Warfighting Development Center under Commander Melvin Smith to shape unmanned doctrine. Around the same time, military doctrine writing in Line of Departure outlined an operational shift away from assigning four soldiers to launch a drone ambush. Instead, new tactical roles like the 15X operator are envisioned to oversee and orchestrate squads of eight to ten autonomous vehicles under a single supervisor.

Across these evolving frameworks, doctrine insists the human soldier must remain the ethical anchor whenever autonomous capabilities are deployed. Proponents point to tangible safeguards built into recent live testing: when DARPA announced its evaluation of an autonomous AI flight agent in a modified fighter jet in July 2026, a pilot remained seated in the cockpit, retaining the ability to override the software and retake full manual control with the flip of a switch. Yet that mechanical safeguard prompts a harder question: can human supervision remain truly meaningful when targeting algorithms begin processing battlefield data at machine speed?

Military commanders insist that this supervisory boundary will hold. In trials testing human-machine teaming, Air Force Colonel John Ohlund explained that while software triages, prioritizes, and sequences recommendations, the human on the loop still makes the final decisions. Lieutenant Colonel Shawn Finney affirmed that doctrine will always keep a human as the final decision-maker. Operational analysts emphasize that the character of war remains a clash of wills demanding ethical oversight, where soldiers verify positive target identification before any strike. Technical analyses point out that algorithmic guardrails, such as control barrier functions, can mathematically restrict software actions to verified safe regions.

Yet critics warn that when machine speed collides with human cognition, the question of who holds the trigger becomes dangerously blurred. In the Air Force's DASH 2 experiment, software generated more than 6,000 course-of-action solutions for about 20 tactical problems in just one hour, forcing decision tempo faster than human capacity. Researchers warn that false alarms cause severe cognitive overload, while machine-learning perception can be fooled by minor alterations that no human would miss, like stickers confusing computer vision on road signs. When operators receive a constant barrage of filtered alerts, supervisory review risks collapsing into automatic rubber-stamping.

That supervision faces an even harder physical barrier under dense electronic warfare. An operator's ability to veto a machine's recommendation depends entirely on continuous low-latency communications. When adversary jamming completely severs tactical datalinks, remote intervention becomes physically impossible. In a cockpit, safety can rely on the flip of a switch, but across degraded electromagnetic airspace, an isolated swarm cut off from its human supervisor exposes a deep uncertainty that doctrine has yet to resolve.

The vision of future military reconnaissance drones is not one of autonomous digital commanders deciding who lives and dies. Instead, edge-AI turns these aircraft into intelligent, resilient routing nodes—systems that can navigate GPS-denied skies, triage targets locally, and pass target markers directly to ground troops, artillery batteries, and space satellites. The transformation achieves what Brigadier General Travis McIntosh reportedly called the real threshold for autonomy: flying drones by commander's intent rather than by hand, shifting soldiers from piloting single airframes to directing autonomous swarms across the network.