One thing that changes with agents is that the user has to trust the software with more than just information. If it can actually make decisions or complete tasks, a small mistake can affect the whole workflow. I can see SaaS becoming more proactive, but having clear limits and human approval for important actions will probably stay important.