The Six-Month Reversal on Autonomous Warfare

In February, Donald Trump was asked about Ukrainian drone expertise.

We have the best drones in the world, actually. — Donald Trump

We don't need their help in drone defense. — Donald Trump

By July 22, the United States had signed a pact to import roughly 100 drones each from six Ukrainian manufacturers and establish joint production on American soil, including MAGURA maritime drones in Oregon ^[1]. Between those two positions lay five months in which Ukraine demonstrated, under fire, that autonomous systems could solve two problems American doctrine had no answer for. The first was electronic warfare. In July, Russian security services foiled an attack on the Rostov region: 13 drones equipped with autonomous targeting, designed to operate without a radio link to their operator and therefore immune to jamming ^[2]. Russia's counter was not electronic. It was a human informant who compromised the operation before launch. The drones' terminal guidance — the final sprint to the target — ran on onboard AI that needed no signal to jam. Electronic warfare, the centerpiece of Russian air defense, could not stop a drone that required no radio link. The second was manpower. By February, Ukraine faced a 300,000-personnel shortfall ^[3]. By April, the NC-13 robot combat unit of the 3rd Assault Brigade had conducted over 100 systematic ground drone attacks, combining aerial and ground unmanned systems with infantry into a single integrated strike package ^[4]. The Ministry of Defence described it plainly.

A new model of warfare is being introduced, drone-assault units that combine aerial and ground unmanned systems with infantry into a single integrated system. — Ministry of Defence of Ukraine

These were not reconnaissance tools. They were infantry substitutes, fielded because the soldiers were not there. The proof traveled through three channels. NATO sent study teams to examine Ukrainian production and procurement cycles, with Deputy SACEUR Sir John Stringer concluding that the West was now in a race ^[5].

The West is now in a race. — Sir John Stringer

Co-production deals followed: the State Department drafted a drone manufacturing agreement with Ukraine in May, reversing Trump's February dismissal ^[6], and the July pact made it binding ^[1]. In June, the 101st Airborne Division conducted a fully robotic trench-line breach on American soil — 500-plus drones and Hunter WOLF unmanned ground vehicles clearing landmines and destroying bunkers before a single rifleman entered, directly replicating the Ukrainian robot-first assault model ^[7]. Colonel Ryan Oliva, who ran the exercise, was explicit about the logic.

We need drones at scale. We need to treat them like ammunition. — Ryan Oliva

That same month, Trump signed a National Security Presidential Memorandum directing the military to accelerate AI adoption for warfighting and ordering Secretary Hegseth to update autonomous weapons policies within 90 days ^[8]. The doctrinal shift was formal. And it was already operational: Project Maven, the Pentagon's AI targeting system, was being used as a primary targeting tool in strikes on Iran, compressing the kill chain from hours to seconds ^[9]. The Pentagon had begun replacing Anthropic's Claude model specifically because the company refused to allow its AI for fully automated strikes.

A system that compresses a kill chain from hours to seconds makes an adversary obsolete. — Alex Karpovsky

The US Air Force was building toward its own autonomous fleet in parallel: the Collaborative Combat Aircraft program planned at least 1,000 drone wingmen for F-35A and F-47 fighters, with autonomy software decoupled from the airframe so algorithms could be updated independently of the hardware ^[10]. The speed of the pivot suggests something about its cause. Ukraine's turn to machine-centric warfare was not a strategic preference; it was a response to quantitative exhaustion — 300,000 soldiers short, artillery stocks depleted, drones accounting for over 80% of destroyed enemy targets by the end of 2025 partly because shells were not available ^[3]. The US faced its own version of the same problem: Patriot interceptors and precision munitions stretched across two simultaneous wars. Autonomous systems were the only asset class that could scale fast enough to fill both gaps. No US official has stated this explicitly as the driver, but the timing and the direction of investment point in the same direction. The shift has a confirmed ceiling. Colonel Oliva, after the 101st's robotic breach, noted where AI still fails.

Large language models don't really understand three-dimensional space, so they're not good for developing courses of action. — Ryan Oliva

Targeting and terminal guidance — the final seconds of flight, the identification of a bunker, the path through a minefield — are now machine tasks. But developing a course of action, the three-dimensional planning that turns targets into a fight, remains human. The division of labor is clear and it is consequential: officers remain in command, but they are no longer at the point of contact where attrition concentrates. The machines go first, and the humans follow after the breach is cleared.