What Makes a Swarm a Swarm?

The term “swarm” is used too loosely. GAO defines drone swarm technologies as systems in which multiple drones coordinate using algorithms, local sensors and communications with reduced need for human attention.

The important word is coordinate.

A traditional remotely piloted system can receive a detailed command from a human operator. A collaborative system may instead receive a mission objective.

The machines then determine how to distribute parts of the task. At the simplest level, that can mean synchronized movement.

At a more advanced level, it can mean distributed sensing, dynamic role assignment, route adjustment, local collision avoidance and recovery after loss of a member. That creates several different architectures:

The Operator-to-Machine Ratio

Drone warfare changes dramatically when autonomy reduces the attention required for each platform. Imagine one operator controlling one aircraft.

To deploy twenty aircraft, the organization may need roughly twenty active control relationships plus supervisors, analysts and support personnel. Now imagine one mission commander supervising a coordinated team.

The human specifies:

the objective;

the operating boundaries;

the priorities;

and the actions that require explicit approval. The machines handle lower-level coordination.

DARPA’s completed CODE program explored exactly this model: groups of unmanned aircraft operating collaboratively under one person’s supervisory control, adapting to changes while presenting recommendations to a mission supervisor. The program is complete, but its architecture has become increasingly relevant.

The scaling bottleneck is no longer only aircraft production. It is cognitive bandwidth.

The Collaborative Autonomy Stack

A swarm is not primarily a collection of airframes. It is a software and communications architecture embodied in multiple machines. The critical layers include sensing, navigation, networking, shared state, task allocation, autonomy policy, human command and safety boundaries.

Replicator and Affordable Mass

The Pentagon’s Replicator initiative made “small, smart, cheap and many” part of the public defense vocabulary. Replicator 1 was designed to accelerate fielding of multiple thousands of all-domain attritable autonomous systems.

The initiative included aerial and maritime systems, small unmanned aircraft and software enablers intended to coordinate large numbers of assets. The most important feature of Replicator was not any single drone.

It was the acquisition model.

Defense officials described Replicator as a mechanism for compressing the path from commercial technology to fielded capability. That is significant because autonomous systems evolve more like software than traditional aircraft.

A five-year procurement cycle can deliver a platform whose autonomy stack is already obsolete. Replicator attempted to shorten that cycle.

The initiative also demonstrated the strategic logic of attrition. Traditional high-end military systems are designed around preserving scarce, expensive platforms.

Attritable systems assume that some machines may be lost. That changes cost, design, maintenance and mission planning.

But it does not make losses free. Every autonomous platform still consumes: electronics; sensors; batteries or fuel; motors; communications hardware; software; manufacturing capacity; and trained support personnel.

Collaborative Combat Aircraft

The Air Force CCA program is one of the strongest current examples of collaborative autonomy becoming a real acquisition architecture. In June 2026, the Air Force awarded contracts covering both Increment 1 air vehicles and mission-autonomy software.

That separation matters.

The Air Force is treating the aircraft and the autonomy as distinct competitive layers. Its YFQ-42A and YFQ-44A prototypes are now in flight testing.

In July, the Experimental Operations Unit operated both aircraft at Creech Air Force Base during a realistic exercise environment. The Air Force emphasized servicing, refueling, relaunch, distributed operations, integration with crewed aircraft and operator-driven experimentation.

This is a maturity transition.

The aircraft are no longer merely aerodynamic prototypes. The service is learning what it takes to operate them.

But the correct label remains:

experimental / developmental semi-autonomous combat aircraft moving toward fielding. They are not yet a mature fleet of fully autonomous combat aircraft.

When Autonomy Becomes Software

The Air Force’s most important 2026 CCA decision may be architectural rather than aerodynamic. Its Autonomy Government Reference Architecture is designed to support platform-agnostic autonomy.

In June, the service described mission autonomy as effectively “software sold separately.” That means the government can potentially compete airframes and autonomy providers independently. A software provider does not necessarily have to own the aircraft.

An aircraft manufacturer does not necessarily control the mission-autonomy layer. This reduces vendor lock.

It also creates the possibility of faster autonomy updates. But modularity creates new assurance problems.

If autonomy software can migrate between platforms, certification becomes more complicated. A policy that behaves safely on one aircraft may behave differently when:

sensors change;

performance envelopes change;

communications differ;

or payloads differ. Software portability therefore does not eliminate platform-specific testing.

The Maritime Autonomy Experiment

The Navy is approaching autonomy from a different physical problem. Ships and maritime robots operate for long periods.

They contend with corrosion, weather, waves, navigation rules and enormous geographic areas. In 2026, the Navy expanded real-world robotic and autonomous system exercises in the Arctic, Baltic, Caribbean and off the U.S. coast.

Long-dwell unmanned surface vessels have already been used for maritime surveillance missions in the Caribbean. The Navy says one autonomous vessel detected and tracked a target of interest in October 2025 and helped cue a later interdiction.

That is operational evidence of autonomous sensing and persistence rather than a swarm strike. In May 2026, the Navy also selected seven companies for at-sea Medium Unmanned Surface Vessel marketplace demonstrations, with successful systems eligible for follow-on production.

The broader direction is clear:

the fleet is trying to turn unmanned systems from special experiments into ordinary force structure. That requires more than autonomy algorithms. It requires: ports; maintenance; communications; launch and recovery; fuel; cybersecurity; rules for navigation; and operators who know how to command mixed human-machine formations.

Communications in a Denied Environment

The simplest multi-drone systems depend heavily on continuous communication. That is a weakness.

An adversary can attempt to jam links. Terrain can block radios.

Satellite connections can disappear. Network capacity can collapse.

A resilient machine team therefore needs enough local autonomy to continue functioning safely when communications degrade. This does not mean machines should continue every mission without contact.

It means the architecture must define what remains permissible when command connectivity is intermittent. DARPA’s earlier CODE work explicitly emphasized operation in denied environments.

Replicator software enablers were likewise described as supporting coordination among hundreds or thousands of unmanned assets while remaining resilient against jamming and other countermeasures. The key distinction is:

communications loss should trigger a defined degraded mode, not undefined behavior. That connects directly to Articles 015 and 016.

Autonomous systems depend on resilient communications and trustworthy PNT. Swarming does not eliminate those dependencies.

It makes them more consequential.

Heterogeneous Teams

The popular image of a swarm is a cloud of identical small drones. The more consequential future may be mixed teams.

One platform may have better sensors. Another may have longer endurance.

Another may provide communications relay. Another may operate on the surface.

Another may operate underwater.

Another may remain crewed.

The system becomes less like a flock of identical birds and more like an orchestra. Collaborative autonomy coordinates specialized machines around one mission.

DARPA’s CODE program explicitly explored mixing systems with different capabilities. The Air Force is building platform-agnostic autonomy for CCA.

The Navy is experimenting with air, surface and subsurface systems across common command architectures. This is why “swarm” may ultimately become too narrow a term.

The deeper concept is:

distributed machine teams.

Human Judgment and the Use of Force

Autonomy creates a question that does not arise in the same way with a warehouse robot: Who decides to use lethal force? Current U.S. Department of Defense policy requires autonomous and semi-autonomous weapon systems to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.

DoD Directive 3000.09 also requires appropriate testing, reliability, safety and compliance with the law of war and applicable rules of engagement. The Air Force has been unusually explicit about CCA.

After its July 2026 live-fire test, the service stated that CCA will not autonomously employ weapons and that weapon-release decisions remain with a human operator. That is an important distinction between:

autonomous navigation;

autonomous formation;

autonomous sensing;

autonomous task allocation;

and autonomous lethal decision.

These are not the same capability. Public discussion often collapses them into one phrase: “autonomous weapon.” Technically and legally, the boundaries matter.

The Counter-Swarm Problem

Every autonomous-system revolution creates a defensive industry. NATO’s current exercises show how quickly counter-drone defense is becoming a systems problem.

At Technical Interoperability Exercise 2026, roughly 300 participants tested more than 60 counter-UAS systems and 40 command-and-control applications. Allied Command Transformation’s Layered Counter-UAS Initiative is explicitly designed around integrating:

detection;

tracking;

command and control;

and multiple defensive mechanisms.

The strategic problem is saturation. A defensive system designed to track and defeat one object may be overwhelmed by many simultaneous objects.

That is why NATO’s current capability requirements emphasize: simultaneous tracking; multiple axes; interoperability; depth of defensive magazines; and cost per engagement. The economics are uncomfortable.

If an inexpensive drone forces a defender to expend a far more expensive defensive resource, the attacker can impose costs even when the drone fails. But cheap defense is not automatically easy.

Defenders still have to distinguish: drone from bird; hostile from friendly; decoy from real threat; and one track from dozens. The counter-swarm problem is therefore as much about information processing as interception.

Attrition and Cost Exchange

Traditional airpower has moved toward extraordinarily capable—and extraordinarily expensive—platforms. That creates strategic scarcity.

Autonomous systems introduce another approach: distribute capability across more machines. If some are lost, the system can continue.

This is one reason the Pentagon uses the word attritable. But the term should not be mistaken for disposable.

A system still has to be valuable enough to justify production. It needs enough capability to matter.

It must be manufacturable at sufficient volume. It must use components that are actually available.

And the software stack must be supportable. This creates a different equation:

The Logistics Underneath Autonomy

Autonomous warfare is often described as if software removes people from the system. It changes where people are needed.

Machines still require:

manufacturing;

transport;

charging;

fuel;

batteries;

spare parts;

motors;

sensors;

software updates;

cybersecurity;

launch crews;

recovery crews;

and maintenance.

A swarm of one hundred machines can reduce the number of pilots while increasing the number of batteries, network nodes and maintenance events. The Air Force’s July Creech exercise is important partly because it tested servicing and relaunch in a more realistic operating environment.

The Navy’s autonomous systems efforts likewise devote substantial attention to deployment, sustainment and integration. The military system that wins will not necessarily have the most intelligent prototype. It may have the autonomy architecture that can be repaired, replenished and updated fastest.

What Is Actually Mature?

The phrase “autonomous swarm” compresses very different maturity levels. This point matters because the technology should be evaluated as part of the surrounding system rather than as an isolated claim or capability.

This maturity discipline is essential. The technology is advancing rapidly.

But the most dramatic conception—hundreds of machines independently making lethal decisions as a self-directed collective—should not be treated as synonymous with current U.S. programs.

The SURVXCOM Autonomous Swarm Test

A real collaborative autonomous system should be judged on more than vehicle count. This point matters because the technology should be evaluated as part of the surrounding system rather than as an isolated claim or capability.

What to Watch Next

Warfare Without One Pilot Per Machine

The deepest change in autonomous warfare is organizational. Militaries have always sought mass.

More soldiers.

More ships.

More aircraft.

But mass traditionally required more people. Autonomy breaks that relationship.

At least partially.

One person can potentially supervise many machines. One crewed aircraft can coordinate with several uncrewed aircraft.

One maritime command center can oversee persistent unmanned vessels across enormous areas. One mission-autonomy layer can eventually coordinate specialized platforms from multiple manufacturers.

That creates military scale without requiring identical growth in cockpit crews. But autonomy does not remove humans.

It moves them upward in the decision hierarchy. The human becomes less responsible for continuous stick-and-rudder control and more responsible for:

mission intent;

boundaries;

exceptions;

authorization;

judgment;

and accountability.

That may ultimately be the defining feature of collaborative warfare. Not machines replacing commanders.

Machines absorbing lower-level coordination so commanders can direct systems at a higher level. The risk is obvious.

As machine speed increases, humans can become ceremonial supervisors rather than meaningful decision-makers if interfaces, doctrine and timing are poorly designed. Article 009 called this problem decision sovereignty.

Article 019 extends it:

the more machines one person supervises, the more important it becomes to define which decisions must remain genuinely human. The technology race is therefore not merely about autonomous capability.

It is about scalable command.

The future swarm is not a cloud of drones. It is a new relationship between human intent and machine coordination.