I’m not sure the problems brought up are solved by the proposed solution. Here are the problems highlighted by the OP as I understand them:
1. High difficulty
2. Unfair dice rolls
3. Slow combat
4. Steep learning curve.
1. High difficulty would be best solved through difficulty settings. Companion AI could either increase or decrease the difficulty, based on how it is implemented. What it would definitely do is remove player agency.
2. Lack of player agency is at the core of the problem with unfair dice rolls. (Player tried to do a thing, but game wrested control from them and said thing didn’t get done.) Companion AI would still have to roll dice, failure would be no less frequent. I believe failed rolls would be all the more disheartening if they were the result of automated decisions, because the player had no say at all in the matter.
3. Players already have no say outside their turn. Because enemy turns as so slow and plentiful, combat feels sluggish. However the pace picks up when the player is given control back. As others have pointed out, making companions automated would make their turns slow and strip 75% of the player’s actual game-time, both of which increase the boring, slow-feeling wait time between turns.
4. Automating companions would alleviate the mental load on newer players. However, it wouldn’t help them learn any of the tools they need to understand what’s going on. Get a player to high ground, they beat an encounter. Teach them to get to high ground, they beat the game.
All in all, I think the OP would be happier with a few changes to difficulty and learning curve than a system that handles a good chunk of the core gameplay loop in their stead.