Watch as a team of drones destroys their target in a demonstration of Fury. That is, the Fury Autonomous Vehicle Orchestrator by Scout AI that uses proprietary AI software to carry out autonomous combat based on simple verbal orders from a human.
It’s often difficult to see trends in a particular field, and the history of technology is strewn with bad guesses that have fallen by the wayside. In 1876, telephones were seen as a fad. In the 1950s, nuclear-powered cars were thought to be just around the corner, along with jet packs, flying cars, and Moon colonies. In 1996, the internet was dismissed as a flash in the pan. In 1981, Bill Gates allegedly declared that no one needed more than 640K of computer memory. Of course, he also thought Zune and Clippy were good ideas.
The military sphere has often fared no better. France believed its Maginot Line of fortresses was impregnable – until the Germans simply went around it. The US Navy scoffed at the idea that airplanes could threaten battleships. During the First World War, neither Axis nor Allied generals fully appreciated the devastating effects of machine guns and barbed wire. Then there was the post-Second World War belief in the “Push Button War,” supposedly fought antiseptically with missiles, which left many a squaddie shaking his head while hauling a 3-in mortar across a muddy field by hand.
Fury
On the other hand, some trends are as plain as a pikestaff. Over the past decade, whether on land, sea, or air, drones have become increasingly common and effective weapons of war, with no clear end in sight as autonomous vehicles grow more diverse and sophisticated.
One notable aspect of the latest generation of drones is their move beyond simple remote control and automated piloting toward full-fledged autonomy, and the ability to act as a team with humans and other drones to achieve shared objectives. In other words, a shift from automation to autonomy.
Until now, autonomous systems have largely relied on pre-scripted mission plans based on rigid, hand-engineered code and conditional logic – essentially elaborate if-then-else flowcharts offering predefined alternative courses of action. They also required electromechanical or digital interfaces to receive instructions.
Silicon Valley startup Scout AI seeks to change this with its Fury software, which replaces traditional coding with a proprietary “Vision-Language-Action” (VLA) foundation model family.
In practical terms, Fury acts as what the company calls an “agentic interoperability layer,” bridging the gap between robotic hardware and high-level human intent. Rather than simply translating orders into actions, it functions as a large language model (LLM) or Vision-Language-Action (VLA) model that can perceive its environment, reason toward a goal, and take independent actions to achieve it based on its training.
Scout AI
In other words, it can self-correct if it encounters obstacles not explicitly covered in its code and devise workarounds. It can also operate using natural language, allowing a commander to issue orders much as they would to a human soldier (minus the swearing), rather than inputting precise machine instructions.
For example, instead of typing “move forward 10 meters,” a commander could say, “find the red vehicle,” and Fury would determine how to accomplish the objective. It could even dispatch other drones as scouts if it cannot immediately see the vehicle and adjust its plans accordingly.
According to the company, Fury operates at a high level, enabling it to control drones from different manufacturers using their existing flight controls and drive systems, without replacing the low-level code that keeps each platform stable. Because Fury uses an LLM-based architecture, it can read the technical documentation of a new robotic platform and bridge the gap between high-level goals and that platform’s capabilities.
At the same time, it translates human language into machine-specific code in real time. The result is the ability to make disparate robotic systems work together and provide commanders with a universal command interface for controlling a heterogeneous fleet of drones. It can also operate using off-the-shelf components and is designed to rely primarily on simple cameras to passively gather data, rather than radar or lidar.
“AI agents are becoming mainstream in the digital world, we’re bringing that same agentic intelligence into the physical world for the US warfighter,” said Colby Adcock, CEO and Co-Founder of Scout AI. “Fury Orchestrator acts as the AI interoperability layer between C2 systems and heterogeneous unmanned assets across domains. It keeps the human at the center, and enables coordinated mass and force projection against our adversaries.”
This all sounds promising, but if Fury can deliver as advertised, it could change the battlefield as profoundly as the rifle or the telegraph. Drones that roll on wheels, fly on rotors, streak through the skies on jets, sail the seas with diesels, or lurk beneath the waves with electric motors are already demonstrating their advantages.
Not least among these is their relative cheapness compared to manned systems, combined with their expendability. Now software like Fury promises a double advantage: turning robotic combatants into adaptive problem-solvers that can be briefed like a squad in a ready room and enabling them to coordinate with one another as seamlessly as computers on the internet.
Scout AI
It’s a future that’s hard to accurately assess. Tomorrow’s battlefield may prove as different from today’s as those of the Second World War were from the age of cavalry. Of course, countermeasures or deterrent effects could render the technology far less decisive, as has happened with many strategic systems over the past 80 years that never saw meaningful combat.
Then there are the unforeseen consequences. In my experience, the future turns out to be not be so much frightening as aggravating. So many promised breakthroughs simply fail to perform as advertised.
During the Second World War, the Soviets trained dogs carrying explosives to run under German tanks in suicide attacks. Unfortunately, the dogs were trained using Soviet tanks, and their first combat deployment ended disastrously. At the same time, the British developed the Sticky Bomb, a grenade coated in powerful adhesive so it could stick to armored vehicles. Unfortunately, it also stuck to everything else – including the soldier wielding it – making it distinctly unpopular.
Perhaps something similar will happen with drones designed to operate like the AI systems we use on our computers. They may prove tremendously useful. Or they could wander off into fantasy land, fabricating plans detached from reality and insisting on executing them while their human commander bellows for them to stop.
Perhaps the ultimate weapon will be the first with built-in idiocy.
We can only wait and see.
Source: Scout AI
