But it gives access to thousands of hours of training data for that particular aircraft across a range of different situations, and it allows you to test that ML model in real situations.
Moving to something lighter or more aerodynamic becomes a lot easier after that. If you look at the Boeing Airpower/Loyal Wingman UAV, it looks like its based on this concept quite closely.
That doesn't seem to be the route the XQ-58 is going, though. [0] It's an attritable stealthy missile/bomb truck - and at a projected unit cost of $2-3 million, they'd be cheaper than some US air-to-air missiles.
Moving to something lighter or more aerodynamic becomes a lot easier after that. If you look at the Boeing Airpower/Loyal Wingman UAV, it looks like its based on this concept quite closely.