Your understanding of current AI behavior is flawed. In missions, for example, the script does not focus on the mission puller - it is the first ship to enter the pocket (if the ships have that trigger), or the first ship/drone to engage a target in a specific group (if that trigger is coded).
After initial aggro, aggro shifts are based on a variety of threat-determination mechanics, including relative ship sizes (small-class NPCs typically prefer to attack small-class targets when possible, for example), damage output, and use of supporting modules (ECM, remote reps, etc).
Could the AI use some improvements? Sure - but that’s the constant struggle with NPCs in video games. Coding a more responsive and intelligent NPC is incredibly difficult. It’s something CCP has continually worked on over the years, with many patches tweaking threat assessment behaviors and adding new abilities. And every time they update the NPCs, players ‘solve’ the content.
Asking CCP to develop cutting-edge AI on a game developer budget is unrealistic- the EVE NPCs will never be able to outthink Watson.