The Joint Force Is Buying Autonomy Faster Than It Can Measure Whether It Can Fight Without It
Every automated system the joint force fields is a system an adversary will try to switch off. The neglected question is not whether anyone can still fight when it goes dark, but whether the force knows, unit by unit and service by service, who can.
Earlier this year, in an expansive training area, soldiers from one brigade loaded an autonomous ground vehicle and watched it move supplies forward without a driver. It worked.
Across the joint force, the same story is repeating with different hardware: autonomous resupply, AI-assisted targeting, machine-speed sensor fusion, uncrewed systems threaded through nearly every warfighting function. The Air Force is developing collaborative combat aircraft to fly alongside crewed fighters, and the Navy is teaching uncrewed surface vessels to cross oceans. The American military is fielding autonomy at a pace that would have seemed fantastical a decade ago, and much of it works as advertised.
That success frames a harder question. Every one of those systems depends on a link, a signal, a position fix, or a software service that an adversary has every incentive to deny. The Army has invested enormous energy in acquiring autonomy. It has invested less, and far less consistently, in a harder question: when the automation is jammed, spoofed, or simply fails, who on the crew still knows how to accomplish the mission by hand, and does the Army have any reliable way to know the answer across a force of uneven readiness?
The link is the target
The war in Ukraine has delivered one lesson with brutal consistency: the connective tissue of modern systems is the first thing to come under attack, and it comes under attack constantly. Russia entered 2022 with more tactical electronic warfare capability than any military had deployed in decades, and it has used it relentlessly. Loss figures for Ukrainian drones circulating in front-line reporting vary widely and resist independent confirmation, and they deserve to be treated with care; what the reporting consistently agrees on is the mechanism, which matters more than any total: jamming, not kinetic defense, is the dominant killer of small uncrewed systems. The physics favor the jammer: a transmitter close to a receiver will overpower a friendly signal originating kilometers away.
The adaptations that followed are instructive because they are not about building better radios. Both sides turned to fiber-optic control that trails a hair-thin glass filament and emits nothing to jam, and to onboard machine vision that lets a drone complete its run after the link is cut. In other words, the winning move under electronic attack was to reduce dependence on the vulnerable link and push capability, and judgment, closer to the edge. When a Starlink outage in July 2025 disrupted Ukrainian military communications, the fragility of centralized connectivity became a headline rather than a hypothetical.
None of this is really a drone story but a dependency story, and it generalizes to any force that has wired its operations to signals an enemy can contest. The United States military is more networked, more automated, and more dependent on space-based position, navigation, and timing than any force on earth. That is a strength when the spectrum is permissive, and an exposed flank when it is not.
Ask what happens when the signal goes away, and the follow-up questions arrive quickly: what are the alternatives, how functional are they under real interference, and what do they cost? For satellite navigation the answers exist as a portfolio rather than a replacement. The military’s hardened M-code GPS signal raises the jamming bar without eliminating it. Beneath it sits a layered set of complements: inertial navigation that drifts without updates, terrain- and magnetic-field matching that works where maps are good, celestial fixes, timing holdover from precise clocks, and signals of opportunity. Each layer works somewhere and none works everywhere, which is exactly why proficiency in choosing and blending them is a trainable skill rather than a procurement line item. The Government Accountability Office reached the institutional version of the same conclusion: the Department is developing alternative navigation systems without measuring its progress toward actually fielding them. Alternatives that exist as programs, like manual skills that exist as doctrine, are not yet capabilities anyone can count on. The measurement gap runs from the rifle squad to the enterprise.
The connective ambition of the joint force is Joint All-Domain Command and Control, the concept that any sensor should be able to feed any shooter across services at machine speed. The concept is sound and the dependency it creates is enormous, so its institutional condition deserves precise statement. As the Government Accountability Office reported in April 2025, each military department is building its own contribution, the Air Force’s Advanced Battle Management System, the Navy’s Project Overmatch, and the Army’s Project Convergence, several of them begun before the joint strategy existed; the Department still lacks a comprehensive framework for linking the hundreds of systems involved, and funding flows through separate service initiatives rather than a single program. None of this is a scandal; it is how large institutions build. But the implication for the present argument is direct. A network that has not yet achieved interoperability on a good day has certainly not rehearsed its degraded modes on a bad one, and no service’s fallback procedures are currently measured against the possibility that the link they assume from another service is the one that fails. The question a joint commander needs answered is the same one a platoon sergeant needs answered, scaled up: when the network degrades, who still knows how to fight, and how would we know?
A doctrine that names the problem, and a force that rehearses it unevenly
To its credit, the Army has named this condition, and in places it trains to it seriously. Its doctrine speaks of the denied, degraded, and disrupted operational environment, and a related body of work addresses the denied, degraded, intermittent, and limited conditions that units should expect. Combat training center rotations increasingly impose degraded voice, digital, GPS, and satellite communications, and some units come out of them genuinely hardened. The 2024 Army Space Training Strategy, endorsed by the commanders of Space and Missile Defense Command, Training and Doctrine Command, and Forces Command, explicitly aims to train the force to fight in a denied space environment. The intent is real, and in the best units the training behind it is real too.
The problem lies elsewhere: how much a given soldier actually experiences fighting blind varies enormously, from unit to unit and even from one training-center rotation to the next, and the Army has no consistent way to measure that variation or act on it. A brigade with a demanding commander and a hard rotation may be genuinely proficient in degraded operations; the unit next door, or the same unit two years and one leadership turnover later, may barely have touched it. Readiness that depends this heavily on local initiative is not readiness the institution can count on, because it cannot see it clearly enough to know where the gaps are.
Part of the unevenness is practical. Realistic electromagnetic denial is genuinely hard to stage on American soil. As commanders at the Army’s special warfare center recently told Defense One, GPS jamming can be exercised regularly at only a small number of ranges, principally White Sands Missile Range and the Nevada Test and Training Range, because civil authorities tightly restrict jamming that would bleed into commercial aviation and telecommunications. One officer noted that the service is preparing formal requests to open more space for this kind of training. When realistic denial is available in only a few places, whether a unit gets meaningful exposure to it becomes partly a function of where it is stationed and how its rotations fall, which is exactly the kind of variation a readiness system should be surfacing rather than leaving to chance.
Where the threat outruns the training, a familiar pattern fills the gap: crews rehearse with the network up, the feed live, and the position fix good, because that is what the home-station range most easily allows. The degraded case gets briefed more than it gets lived. The Army senses this, which is why organizations like Futures Command run dedicated denied-environment experiments. But an experiment run by a specialized organization tells you what is possible, not what the average rifle company has repeated often enough to rely on.
The skill that atrophies when you stop using it
There is a well-documented human dimension underneath this, and it is not unique to the military. Across aviation, driving, and even surgery, researchers have found that reliance on automation erodes the manual proficiency it replaces, and that expert performance can degrade measurably once the assistance is removed and not deliberately maintained. The pattern is consistent enough to state directly: a skill you delegate to a machine and never practice is a skill you are slowly losing, whether anyone notices on a good day.
Consider what that means for ground combat. A generation of soldiers is learning to navigate by the moving map, call for fires through a digital system, and trust a sensor picture assembled by software. These capabilities are enormous advantages, but the corollary is that manual alternatives, terrain association with map and compass, degraded-mode gunnery, voice procedures under jamming, analog backup for digital fires, get exercised only as much as a given unit chooses to exercise them. In some formations that is often; in others, rarely. The manual layer does not disappear on a schedule, and it does not thin at the same rate everywhere. Where it thins, it does so invisibly, and it presents its bill on the worst possible day, in the first hours of a fight against an enemy who has done its homework on your dependencies.
Underneath all of this sits a force-design question that deserves honest framing: what is the right mix of crewed and uncrewed systems, and which combinations actually work best under fire? Ukraine’s answer has changed every few months, which suggests the honest position is that nobody’s answer is final. That is an argument for humility in acquisition and for measurement in training, because the only reliable way to learn the right mix is to test candidate mixes under realistic denial and count what completes the mission, not what survives the briefing.
Treat manual reversion as a measured standard, not an assumption
The answer is not to slow the adoption of autonomy. The advantages are real, and an adversary uninterested in fair fights will exploit any hesitation. The answer is to close the variance, to make the ability to operate without the automation an explicit, measured, and rehearsed standard rather than something that happens to be strong in some units and absent in others. Three commitments would move the force in that direction, and none requires a new program of record.
The first is to build the denied environment into ordinary training rather than reserving it for specialized experiments or the luck of a hard rotation. If national ranges are the only places full jamming is lawful, then the more common conditions, GPS-denied land navigation, comms-out mission command, sensor-degraded targeting, can still be imposed by exercise design at home station: turn the systems off, or have observer-controllers rule them dead, and require the unit to finish anyway. The point is not fidelity to a specific jammer but making the experience of losing the crutch and continuing the mission a routine event rather than a rare one.
The second, and the one that most directly answers the measurement gap, is to make degraded-mode competence something every service assesses and reports the way it assesses gunnery and physical fitness. The problem today is not only that some units under-train it. The institution cannot readily tell which units those are. A crew that can only perform with its automation up is not fully ready, and readiness reporting should be able to say so. What gets measured gets trained, and what gets measured also gets seen: a degraded-mode standard would both pull manual proficiency out of the category of nice-to-have and give commanders a way to know where, across an uneven force, the real gaps lie.
The third is to treat the manual layer, the people and the procedures that can carry a mission when the network cannot, as a capability to be resourced and protected rather than a legacy skill left to atrophy at whatever rate each unit allows. That means retaining analog backups where they matter, documenting the procedures before the people who remember them retire, and rehearsing the handoff from automated to manual under stress rather than discovering it during one.
A final consideration belongs to the adversary, who studies this force more attentively than any friendly audit. Every automated system fielded teaches an opponent what to counter, and every downed system on a battlefield teaches its recoverers something about how the sender fights. Adaptation now runs in both directions at a wartime tempo. The side that knows its own degraded-mode floor, precisely, unit by unit, learns faster from those exchanges, because it knows which losses change the plan and which do not. Autonomy will keep earning its place across the joint force, and it should. The mark of a serious military is that it also knows, and can prove to itself, exactly how well it fights on the day the autonomy cannot.
The quiet edge
There is a version of the next war in which the side that automated the most wins decisively in the opening hours. There is another version, closer to what Ukraine’s experience suggests, in which both sides rapidly deny each other the very links their automation depends on, and the advantage flows to whoever adapted fastest to fighting without them. The United States should intend to win the second war, not merely the first, because the second is the one a capable adversary will try to force.
The Army is right to buy autonomy, and right to buy it quickly. But a force is defined not by what its machines can do when everything works, but on the day that matters, by what its people can still do when the machines go dark, and by whether the institution knew in advance which of its units could do it. That capability is perishable, it is unevenly held today, it is cheaper to preserve than to rebuild under fire, and the time to make it a measured standard is now, while the stakes are still counted in training rotations rather than in casualties.
Burak Oktenli holds an MBA and a Master of Professional Studies in Applied Intelligence from Georgetown University, where his work focuses on the governance of autonomous decision-making systems. His writing on autonomous systems and defense has appeared in the Modern War Institute, RUSI, and RealClearDefense.
Source link
