fiIf the Kratos Valkyrie drone were production ready, I think we could close the skies over Ukraine without direct NATO involvement. One of my projects at Hopkins was to use gameplay from Digital Combat Simulator to create AI that could control a fighter jet: You basically break the task into three parts, plane type identification, general course of action/tactic selection, and then tactic implementation. The reason identifying the sort of plane you are up against is so important is that things like relative turning rates and thrust to weight ratios matter: If you are in an F16 and fighting an F15 (not that this would happen in the real world as both are US planes) you generally would go for a two-circle fight (one where the fighters turn in opposite directions) because the F16 turns faster than the F15. On the other hand, the F15 would want to do something that takes advantage of its superior thrust to weight ratio and greater service ceiling.
Anyway, I built something over a few months that could destroy all but the very best DCS pilots---and that was one person working alone, with other schoolwork to do and a full time job.
And drones have one key advantage that regular fighters do not: There is no human pilot with a limit on the number of g's he can handle. If you built your drone to specifically handle high g maneuvers, it should be able to beat anything else in the sky even if it were somewhat underpowered in thrust to weight terms.
This is one of the reasons I think forward swept wing designs are a natural fit for the drone space; they were of limited use for human beings because humans were not physiologically capable of taking advantage of the extreme maneuverability they offered. A computer, however, is not limited in this way. They also reduce stall speeds, meaning they could linger over the battlefield for longer, ideal for reducing fuel consumption and providing close air support. Lastly, they can take off and land on smaller airstrips, also useful if your drones have to operate from irregular locales on improvised airstrips. Of course, the great counter argument against forward swept wings is that thrust vectoring has made them unnecessary, but thrust vectoring is expensive, and drones are supposed to be cheap (or, at least, some of them are). [For the record, there is Chinese Analysis of the Valkyrie’s characteristics that suggest it may not be esp. maneuverable. It appears that it will rely on stealth to save it from enemy fire. That said, I believe its BVR abilities could still play a role in closing the sky over Ukraine.]
Similarly, building AI to operate US SAM systems that require substantial training should also be possible---though I understand why we are sending Ukraine Soviet SAM systems that are easier to use and implement. One friend of mine, who eventually became a full time programmer but started out his career as an electrical engineer and who worked on projects like the Camanche helicopter, thought it profoundly unwise to field an AI control system developed over a month: He is right, it risks dangerous friendly fire incidents, but it is doable from a purely engineering perspective. And there is a simple solution, Ukraine can intentionally avoid the corridors controlled by these systems, while operating in those that are controlled by the older, Soviet systems.
AI offers a path to closing the sky over an embattled nation, like Ukraine, while circumventing the political entanglements that come with direct human involvement. In this new Cold War, AI offers us a path by which we can conduct proxy wars while still avoiding escalation.