Agent Use Cases › Productivity

Have an AI agent list only the emails that actually need you

Three hands-on tests in the past week compared personal agents on inbox triage. Claude Duet scored 9.3 out of 10 in one, partly for saying “nothing here needs you” about an account full of automated mail, while Muse claimed an email hadn’t been sent when it had. In another, Grok Bot won by handing back its to-dos as a checklist document.

Done with: Claude Duet, ChatGPT Work, Gemini Spark, ChatGPT Dots, Grok Bot and Meta Muse, connected to Gmail

How did the testers set it up?

Diana, the no-code AI agents educator behind AI Agents Library, ran Claude Duet on her real business accounts on October 27 and 28, 2026, using the same run sheet she’d used for ChatGPT Work, Gemini Spark, ChatGPT Dots and Meta Muse. Her prompt asked the agent to go through the last seven days of email, read whole threads, check whether she had already replied, and list only what needed her, by urgency, without taking any action. The first Gmail account she’d connected held mostly automated mail; Duet said none of it needed her and asked whether she’d connected the right account. After she switched, it found several emails she had fallen behind on, ranked in the same order she would have chosen. It sent nothing and changed nothing. In Claire Vo’s nine-task test, Grok Bot returned its to-dos as a document with check marks, Claude used its task list, Muse found parenting obligations ahead of Halloween week, and Dots sent a message with about five priorities. She called Grok Bot’s document the best working surface, because a list buried in a chat message creates more work. On the ToTheMoon podcast, Alexander Volchek said Dots has taken over about 80% of his email: it sorts his children’s school mail and texts the relevant items to his wife over iMessage.

“Nothing here needs you” is the hardest answer for an inbox agent to give, and the most useful when it’s true.

How did the agents compare?

How can you try it yourself?

1. Connect one email account and confirm which one the agent is reading before you trust its answer.
2. Ask for a review only: the last 7 days, whole threads, skip anything you’ve already answered, ranked by urgency. No sending, archiving or deleting.
3. Ask for the result as a checklist or document you can tick off, not a paragraph in the chat.
4. Compare its list with your own for a week, and note anything it missed or flagged that was already handled.
5. Put money and deadline items in your calendar or task app as well, and only let it draft replies, never send them, until you trust it.

Go through my last 7 days of email in [account]. Read each whole thread, check whether I've already replied, and list only the emails that still need me, ranked by urgency, with one line on why and the date each one needs action. If nothing needs me, say so. Don't send, archive, label or delete anything. Give me the result as a checklist, and tell me which account you read.

← All Agent Use Cases