US Air Force swaps AI midflight on YFQ-44A

The United States Air Force just treated combat drone autonomy like an app you can close and reopen, except this app flies at jet speeds.

In a recent test, Anduril Industries demonstrated that its YFQ-44A Collaborative Combat Aircraft can switch between two completely different mission autonomy systems while airborne, no landing, no wrench turning, no drama, as Interesting Engineering reports.

The aircraft first flew using Shield AI’s Hivemind software. Midflight, it pivoted to Anduril’s own Lattice for Mission Autonomy and ran the same test points again before returning safely.

If this sounds subtle, it is not. This is architecture strategy in motion.

Software as a weapon system

According to the company, the jet took off and autonomously navigated to a designated point where Hivemind was activated to execute a structured series of test cards. After completing those evaluations, the aircraft switched to Lattice and repeated the same mission set.

No hardware swap. No flight system redesign. Just a different autonomy brain layered on top.

That separation is intentional, and a few years ago it may have looked like magic to some of us.

Mission autonomy handles the tactical thinking once a human gives high level direction. It is the decision maker in the cockpit that does not exist. Targeting logic. Maneuver selection. Task execution.

Flight autonomy is different. It keeps the aircraft stable, safe, and airworthy. The Air Force walls these layers apart through its Autonomy Government Reference Architecture, designed so companies can plug in different mission stacks without jeopardizing flight certification.

Think of it as a smartphone where the operating system keeps the device stable, while apps can be updated, deleted, or replaced without bricking the hardware.

In combat aviation, that modularity is oxygen.

CCA and the 2026 decision

The software swap happened under the Air Force’s Collaborative Combat Aircraft program, known as CCA. The service wants autonomous wingmen that can fly alongside crewed fighters, extending range, adding sensors, and absorbing risk.

Earlier this month, General Atomics announced that its YFQ-42A flew with Collins Aerospace’s Sidekick autonomy software.

Now Anduril has raised the bar by proving not just that autonomy works, but that it can be swapped midair.

The Air Force plans to make a production decision on both aircraft and mission autonomy software for Increment 1 in 2026. That choice will determine which platforms and which digital brains become the first operational autonomous teammates in U.S. fighter formations.

If the architecture holds, future upgrades could look less like decade long aircraft programs and more like version updates.

DroneXL’s Take

This test is less about which AI wins and more about who controls the interface.

By separating mission autonomy from flight systems, the Air Force is trying to avoid vendor lock in at Mach speed. The real power sits in the architecture.

If it works as advertised, the service can compete software providers against each other continuously, forcing rapid iteration without grounding fleets.

That is a profound shift.

Air dominance used to hinge on airframes and engines. Increasingly, it may hinge on code that can be replaced between waypoints.

And if autonomy becomes modular, the battlefield becomes a software arena where updates travel faster than jets.