The true problem with autobattles is the Medkits - you can heal all the damage in the world with a Medkit during battle. Fix that and people would be twice as cautious to resort to autobattle, even if no one dies, but rather several agents become severely wounded.
Another problem is that soldiers' skills develop way too slowly (max 1 point per battle per stat). Losing a soldier hurts. If they developed faster, much faster, getting a single soldier killed wouldn't be such a big deal.
I think that once a soldier dies, the game should scan all the savegames in the save folder and remove the soldier from existence permanently. No more reloading after deaths. If someone wants to make and keep manual backups of savegames, so be it. In fact, it would advisable to make a manual backup after 4 hours of playing.
There's a reason autobattle was put there in the first place. Who was it that coded it? Who supported it? Ask those guys why it's there.
A game can become a chore. We've all been there. Having autobattle is a good thing. But autobattle should have its drawbacks, such as less loot, less soldier development, more likely civilian casualties (already there, I believe), more displeased nation where the mission takes place, etc. etc.
We DO all agree that current autobattle mechanic is... too random, right? Right?!
Perhaps opting for autobattle would be more interesting if there was a prediction of the outcome, calculated by the game? The way it would work is that the game plays the autobattle a hundred times over and then makes a statistical prediction based on the results. The prediction would read something like "High probability of more than one casualty" or "It's more than likely that all of your soldiers will die if no combat supervision is provided". It could present the best case scenario and the worst case, such as "Best case: Light casualties. Unlikely deaths. Worst case: Several soldier casualties." Or simply print out the raw data in numbers: 42.8% chance of losing 1 soldier. 1.8% chance of losing all 8.