From brain dump to schedule: capacity-capped weekly planning with AI

Step 1: Dump everything, don't tidy

Open a note and write every task in your head, on sticky notes, and in your inbox. Half sentences are fine: "call Dave re: van," "quote for Elm St," "dentist?" Don't sort, don't group. Getting it out of your head is the first relief.

If tasks mention customers by name, swap the names for something like [CUSTOMER 1]. Here's how to remove private details before you paste.

Step 2: Add what can't move

Under the list, write what's fixed this week:

• Appointments and meetings, with days and times.
• Real deadlines.
• Days or half-days you're not working.

Without these, the plan will look neat and won't fit your actual week.

Step 3: Ask for the plan

Paste both and send:

Here's my to-do pile and my fixed commitments this week.
Make a Monday-to-Friday plan. Max 3 tasks a day, real
deadlines first. Put small tasks under 10 minutes in one
batch per day. List what I could delegate or drop. Ask me
before dropping anything that involves a customer. If a
task is too big for one day, split it and give the first
step.

Step 4: Move what feels wrong

Read the plan. If Tuesday looks impossible, say so: "Tuesday is a site visit, move two things." If it put something hard on Friday afternoon, move it. The AI doesn't know how your week feels. You do.

Then put the plan where you'll see it: your calendar, a paper note on the desk, or the top of your notes app.

Step 5: Redo it on Friday

On Friday, paste what got done and what didn't, plus anything new, and ask for next week's plan. Things that slipped two weeks in a row are worth a second look. Maybe they need help, or maybe they don't need doing.

New to all this? Here's a first week of using AI at work, one small job a day.

Under the hood

Language models are decent at the parts of planning that are about language: grouping duplicates, spotting dependencies ("quote" before "order parts"), estimating rough size, and splitting a vague task into a first step. They're weak at the parts that need context they don't have: your energy, travel time, and which customer will call twice if ignored.

That's why the prompt pins down:

• Hard constraints the model can't infer.
• A capacity cap, since models overpack days by default.
• A confirmation gate on drops involving customers.

The weekly re-plan closes the loop. Completed and slipped items give the next plan better information than any single prompt can.

Short close

Dump the whole pile, add what can't move, ask for three things a day with deadlines first, and move what feels wrong. The pile won't get smaller. It just gets an order, and that's usually enough to start.