You have to treat LLMs like humans under time pressure. If you give a person too many tasks in a short amount of time, they will probably also skip and/or rush

You have to treat LLMs like humans under time pressure. If you give a person too many tasks in a short amount of time, they will probably also skip and/or rush some tasks. It's important to clearly identify which parts ought to be deterministic. It's difficult to get both human-level intelligence and machine-level precision at the same time. You still need deterministic tools to do the precision work. That said, I think LLMs definitely have a place and can provide a degree of flexibility to processes which was not possible before.