Who Commands the Drone Wingman?

1 What Was Actually Demonstrated

The Navy announced on September 22 that it had successfully tested its Collaborative Autonomy Mission Planning system, with the demonstration conducted during the Gray Flag 2026 event at the Point Mugu Sea Range in California using a single General Atomics MQ-20 Avenger as a surrogate aircraft. Other accounts place the flight on September 10. The stated achievement is that CAMP let mission planners work through the Collaborative Mission Planning Continuum to direct an autonomous platform without manually translating the operational plan into platform-specific code, a step that normally consumes time and specialized engineering knowledge.

The mechanism described by GA-ASI and the Navy office involved is a common language: objectives, priorities, limitations and threat context are entered in the planning system in a form the autonomy software can interpret. The architecture named for this role is the Autonomy Government Reference Architecture. The point of the test was not the airframe.

2 The Eliminated Step Is The Real Deliverable

Manual translation of intent into code is where autonomy programs historically lose tempo. It requires a human to restate a commander's intent in software terms, which introduces delay, interpretation error, and a narrow bottleneck of people who can do the translation. If CAMP removes that step, the value is measured in hours between a decision and an aircraft executing it, not in aircraft performance.

That is also why the demonstration should be read as a mission-planning result rather than an air combat result. The Avenger was a carrier for a software interface, and the claim the Navy is making is about the interface.

3 The Surrogate Problem

An MQ-20 Avenger is not a collaborative combat aircraft. It is a large, mature unmanned aircraft standing in for one, which is standard practice for early software work and also a limit on what the result means. A surrogate can validate that an interface accepts a plan and that an aircraft flies it. It cannot validate the operational concept, because the operational CCA will have different performance, different sensors, different signatures, and - most importantly - a different relationship to a crewed aircraft it is expected to accompany.

4 Authority And Workload In The Command Chain

The questions a demonstration like this leaves open are the ones that determine whether collaborative autonomy is usable at scale. How much deviation from the delivered plan may the aircraft take on its own authority? Does a detected pop-up threat permit a course change, a sensor cue, or a weapon release without a human in the loop, and are those three decisions treated identically? If a crewed aircraft is supervising several uncrewed ones, how does the pilot's workload scale with the number of wingmen, and at what count does the human become the constraint rather than the commander?

Workload is the underrated variable. A pilot who must approve every meaningful action of three wingmen has not been given three aircraft, they have been given three additional task streams. The value of autonomy is that it reduces required approvals per unit of mission, so the metric that matters is decisions-per-pilot-hour, and no public reporting yet puts a number on it.

5 What A Test Can And Cannot Establish

A range event can establish integration, interface behavior, and that the software did what the designers intended under scripted conditions. It cannot establish performance in degraded communications, behavior under adversary electronic attack, or the reliability of a plan built from incomplete intelligence. Those are the conditions collaborative autonomy will actually face, and they are adversarial rather than nominal.

6 The Questions Before Routine Operations

Four answers would need to exist before a CCA operates routinely alongside crewed aircraft. The rule set for autonomous action, written so that operators can predict it. A workload model that survives contact with multi-vehicle supervision. Verification evidence from contested, communications-denied conditions. And a clear statement of liability and authority when an autonomous aircraft takes an action nobody specifically ordered.

7 The Bottom Line

The Navy has shown that a human mission plan can reach an autonomous aircraft without an engineer in between, which is a genuine and necessary advance - it shortens the path from intent to action. It has not shown who commands the aircraft once the plan meets the unexpected. That is a policy and doctrine question as much as a software one, and it is the question that will set the pace of fielding.