An AI to do list: your agent adds, you decide

An AI to do list sounds like a list that thinks for you. Most people do not actually want that. They want the list to stay short and quiet, and they want help with the boring parts: typing things in, finding what is still open, tidying up. The quiet version is simple. Your agent adds and reads. You decide what matters and what is done.

What people mean by an AI to do list

Ask around and you get two different pictures. One is an app with AI built in: it sorts your day, suggests what to do next, and moves things around when you fall behind. The other is a plain list that your own assistant can reach, so you can say "put that on my list" in a conversation and it is there.

The first kind can be useful if you like being managed. The second kind is for people who already talk to an assistant like Claude and just want the things that come up to land somewhere real. This guide is about the second kind. The list stays a list. The intelligence stays in the agent you already use.

The quiet version

Start from a list that does very little. You type a todo in one row. You tap a checkbox and it leaves the list. You star the few that are urgent. That is the whole idea behind a minimalist to do list, and it matters more once an agent is involved, not less. A list with fewer moving parts is a list an agent cannot make a mess of.

Then connect your agent. Now the typing and the reading can happen in a conversation instead of on your phone. You still open the list when you want to. Nothing pings you. The agent only acts when you ask it something.

What it looks like in practice

You are talking to your assistant about the week. You say:

"Add booking the car service, renewing my passport, and buying a birthday card for Sara."

Three todos appear on your list:

• Book the car service
• Renew passport
• Birthday card for Sara

Later you ask:

"What is still open, and which of those did I star?"

The agent reads the list and tells you. On Friday you say "the card is done" and it checks that one off. It does not check anything else off because it guessed you were probably finished. It does not delete the passport todo because it looks old. It does what you said, and only that.

What to hand over

• Capture. Anything you mention in passing that you would otherwise forget to type in.
• Splitting. "Plan the garage clear-out" can become three or four concrete todos you can actually tick.
• Notes under a row. The phone number, the size, the reference number, written under the todo it belongs to.
• Reading back. "What is open?" and "what did I finish this week?" without scrolling.
• Tidying on request. Merging two todos that say the same thing, when you ask for it.

What not to hand over

• Deciding what is done. Only you know if the call actually happened. The agent checks off what you say is done.
• Deciding what matters. A star is your call. An agent that stars everything has told you nothing.
• Deleting on its own. Deleting is for duplicates and mistakes, and only when you ask. A finished todo is checked off, not deleted.
• Nagging. The list sends no reminders. If you want to be reminded, that is a calendar's job, not the list's.

Why a simple list works better with an agent

Big task apps have due dates, projects, priorities, labels and views. Each of those is another thing an agent can set wrong, and another thing you have to check. A simple to do list app has a title, a checkbox, a star and a note. There is very little to get wrong, and everything the agent did is visible at a glance.

It also keeps the list yours. You can open it on your phone in a second, see exactly what is there, and change anything. The agent is a way in, not the owner.

How to start

1. Connect your agent to the list. In Claude that is a connector you add once; the steps are in add to your to do list from Claude.
2. Try one capture. Say three things you need to do and check that three todos appeared.
3. Try one read. Ask what is open. Compare it with the list on your phone.
4. Try one check-off. Say something is done and watch it leave the open list.
5. Set the rules out loud. Tell the agent: short titles, check off only what I say, delete only when I ask.

That is enough. If after a week you find yourself typing less and forgetting less, the AI to do list is working. If you find yourself checking what the agent did, make the list simpler, not the agent smarter.

Where to go from here

Very Simple Todo is a simple to do list app built for exactly this: one row to add a todo, a checkbox, a star, a note under any row, a home-screen widget, and no reminders. Your agent connects in Claude or any other MCP client, reads and adds to the list, checks off what you say is done and deletes only when you ask. It costs $0.99 a month, and the first month is free.