Decide once
"Do my taxes" sat on my to-do list for eight months.
It wasn't hiding. It was in the notes app on my phone, right where I put it, between a grocery item and an idea I'd had in the shower. That was the problem. Every line on that list looked the same. No categories, no priorities, no dates. A task only got done when I happened to scroll past it on a day I happened to have the energy. Everything else waited until I rediscovered it, usually at the worst possible moment.
I don't think I lacked discipline. I lacked a system that made the decision for me before the day started making it instead.
The principle: decide once
Most of what drains me in the morning isn't the work. It's the deciding. What do I do first? What's due? Did I forget something? Every one of those questions is a small tax, and they compound before I've had coffee.
So the whole design rests on one split:
- A routine is decided once and runs every day in the same order. Let the dog out. Feed the cat. Treadmill. Shower. Breakfast. I don't re-decide any of it.
- A task is decided daily and done once. Book an appointment. Call the electrician. Finish the paperwork.
They live in separate places on purpose. When routine steps sit in the same list as real tasks, the list gets long, the important things get buried, and you're back to scrolling.
Where it lives
I started in Todoist. I hit the free plan's project limit the same evening. That turned out to be a gift, because it pushed me to put everything in a database I own: a Supabase project that already holds data for my other agents.
There are three tables:
- Tasks: every task, grocery item and agent idea, each with a project, a priority from 1 to 4, and an optional due date.
- Routine steps: the morning routine, in order.
- Autoship items: things I buy over and over, flagged to be delivered instead of fetched.
Day to day, I don't touch the database at all. I talk to Claude, which reads and writes the tables for me. "What's up for today." "Add pick up dog food." "That's done." "Kill the gutter thing." "What do I need at Sam's?" The interface is a conversation.
The rules that make it work
The software is the easy part. The rules are what keep the list honest.
Dates mean something. A task only gets a due date if it really has to happen that day: someone is waiting, there's a penalty, or it expires. Most tasks have no date at all. When a dated task slips, it keeps its old date and shows up as overdue until I do it, move it, or kill it. No quiet rescheduling.
Killing a task is a decision, not a failure. There are three statuses: open, done, and killed. Killed means I looked at it and chose not to do it. That's different from forgetting, and the list should know the difference.
Tasks are written as the next physical action. "Find the trust document and read it," not "review the legal paperwork at some point." If I can't picture my hands doing it, it isn't a task yet.
Priority is about drift, not urgency. P1 is urgent and important. P2 is important but not urgent, which is exactly the category that rots on a notes-app list. P2 is where the real value hides, so P2 gets special treatment.
Eat the frog
Every night I pick one P2 task for the next morning. That's the frog. After the morning routine and school drop-off, before the workday starts, I get thirty minutes for it.
One task. Not a list. The important-but-not-urgent work never wins a fight against email, so it doesn't get a fight. It gets a slot.
Groceries don't get a vote in the morning
Groceries are in the same table, but they follow their own rules. They're never dated, they're always low priority, and they never show up in the daily list. Each one carries a store. When I'm standing in the store, I ask for that store's list, and anything I don't buy stays for next time.
And when the same item keeps showing up, that's the signal it shouldn't be a grocery item at all. It should be on autoship. The goal isn't a better shopping list. It's fewer trips.
The agent: one email at 5:30
Here's the part I'm building now. Everything above works when I ask for it. The agent makes it show up without asking.
It's an n8n workflow with four steps:
- Wake up at 5:30 AM, my time, every day.
- Ask the database for the brief. A function in Supabase already assembles it: the morning routine in order, then today's tasks and anything overdue, sorted by priority.
- Shape it for a phone. Routine first, then the frog, then work and personal tasks in separate blocks.
- Send one email to my Outlook inbox, readable in under a minute.
No AI in the loop for this part, and that's deliberate. Writing a list doesn't need judgment; it needs to be correct and on time every single morning. The intelligence lives upstream, in how the tasks were captured, prioritized and dated. The morning email just delivers decisions I already made.
What I'm learning
I've never in my life kept a morning routine. Not once. The routine I'm running now was built backward from the one fixed point in my morning, my daughter leaving for school, and it collapses gracefully on a bad day to the three things that can't be skipped: the dog, the cat, and getting her out the door.
That's the real design lesson. A system that only works on good days isn't a system. It's a mood.
I have a lot of agent ideas. There's a whole project in the task list just to park them. This is the one I'm seeing through, start to finish, because it's the one that decides what I do with the rest.
Decide what you want. Build the system that makes it real.