Authorizing the same execution over and over is a workflow defect I want fixed. I delegate a task and still end up responsible for pressing continue. The tool needs to carry my decision forward and recognize when a genuinely different decision has appeared.
The task is already under way and the agent pulls my attention back to questions it could have investigated itself or to permissions I had already granted.
I want the agent to advance up to a recognizable boundary. Research and preparation can be authorized while applying a change requires another decision. If the tool understands that, it can do substantial work before needing my attention, and the permission travels with the execution across several stages.
A missing credential or a new consequence deserves a specific question. Accessible documentation deserves reading. Combining those situations into a generic confirmation forces the user to reconstruct the context just to discover why everything stopped. The agent should investigate first and bring back only the point that depends on somebody.
I also want to see how far a permission reaches. If I authorize everything without understanding the consequences, I stop trusting it; if every step becomes a confirmation, delegating wears me out. The tool needs to tie identity and scope to the concrete action in a way the person doing the work can inspect. A confirmation needs to bring new information or a consequence outside what we agreed. Work that stays inside the permission already granted should continue without calling me back.