I stopped carrying work out of my inbox. I gave my agents email addresses instead.

I kept seeing the standard advice to build an AI inbox triage agent. That sounds good until you look at a real inbox. Too many little rules. Too much context. Too many things where the right move depends on something the agent cannot infer from a subject line. I did not need an agent sorting my inbox. I needed a way to hand an email to the agent who could deal with it. So I gave my agents email addresses on my own domain. Now if a bill, document, article, or request shows up, I forward it to the right address and the work starts. I do not have to leave Outlook, open a chat app, find the right conversation, and explain what I am looking at all over again. The part I did not expect to like this much is replying. I answer the agent's email and it picks the conversation back up with the same context. The thread stays next to the email that started the work, which is where I was already looking anyway. Some addresses are agents. Some are just jobs. Forward a bill and it becomes a ClickUp task. Forward an article and a summary comes back. Send attachments to another one and they land in my knowledge vault. The address is basically the instruction. I still decide what gets sent where. I just stopped carrying work between apps before anything could happen. Has anybody else set up agents this way, where email is the front door instead of another thing the agent has to triage?