That discomfort of letting an agent touch your personal inbox was actually the core insight for me. You lose visibility into what it's doing, and mixing agent actions with human mail is just a recipe for confusion. An agent should have its own identity, its own address, its own inbox. Would love to see where you're taking agentmail — are you focused more on the agent-as-sender side, or agent-as-receiver (verification codes, inbound parsing, etc.)?