I do believe that autonomously drones will just make kill chain faster, but human intervention will be necessary for ages yet, and cannot imagine how an visual algorithm could differ between a Russian or an Ukranian soldier, for instance.
https://en.wikipedia.org/wiki/CBU-97_Sensor_Fuzed_Weapon
contained submunitions that autonomously target tanks, and visual + radar target recognition was a feature in both the Pershing 2 and Tomahawk missiles.
Multipoint sensor fusion contains track-data the autonomy calculates a level of confidence to decide friend from foe in the battlespace? The hole in that idea is the fog of war situation where the vehicle is unable to receive the world update to confidently interpret the moving parts onboard sensors perceive.