Letting AI triage my to-do list, and answer the questions I leave in it

I use Things 3 as my task manager, connected to Claude through MCP. Once that connection exists, AI can do a lot more than add tasks for me.

A grooming run, twice a day

I capture tasks fast and messy: half a thought typed on my phone, a link with no context, a title that only makes sense today. Twice a day, an agent goes through everything new and cleans it up:

• Fixes vague titles so each one starts with an action I can take
• Applies tags for context, like whether it's a phone call, an errand, or quick
• Files each task into the right area and project
• Flags anything that needs my review, instead of guessing

It works from a written guide to how I organize things: what each area covers, what each tag means, how titles should read, and how to decide between two places a task could go. That guide is what makes the results consistent. Without it, the agent would organize my list the way it thinks lists should be organized, not the way I work.

The part I didn't expect to value as much

When I capture a task, I sometimes leave an open question in the notes, like "what's a fair price for this?" or "which of these makes more sense?" During the grooming run, the agent sees the question and answers it right there in the note, before I've even opened the task.

A real example: I was selling a guitar. I captured the task and left a note wondering what it should go for. By the time I got to it, the agent had looked up comparable listings and written a price recommendation into the note.

No research detour, no switching contexts. The task was ready to act on. That's the shift: from a list that holds my work to a system that does some of the thinking while I'm doing other things.

The agent only writes into the note. It doesn't list the guitar or contact anyone. The decision stays with me.

Where this pattern fits in a business

Strip away the to-do list and the pattern is: a stream of incoming items gets cleaned up, sorted, and sent to the right place, and each one arrives with the research already done. Most businesses have a stream like that. Some ideas:

• A shared inbox (info@, sales@, support@): each email gets a clear summary, a category, an urgency, and the right owner, with anything unclear flagged for a person.
• Quote requests: each one arrives with the customer's history and what you charged for similar jobs pulled up next to it, so pricing takes minutes instead of a search.
• Support tickets: each ticket arrives with the customer's recent orders and any past tickets attached, and a suggested answer from your own policies.
• Online reviews: each new review gets sorted by topic and urgency, with a reply drafted for a person to approve. My App Store review monitor works this way.
• Internal requests: IT, facilities, or purchasing requests get routed to the right person with the details filled in, instead of bouncing around.

In every one of these, a person still makes the call. The automation does the sorting and the legwork, so the person's time goes to the decision.

The same agents later got a way to rank everything on the list, too: prioritizing a 238-task to-do list with AI.