Sunday night, the task list holds 23 items across two apps and a notebook. 4 of them have rolled over so many times the app now shows a count.
The usual move is to pick the easiest 3 and call it a plan. By Wednesday the plan is wrong, and the list is longer than it was on Sunday.
The noise: wish-list planning
That ritual has a name: wish-list planning. A wish-list plan collects everything you want to do, then asks future you to make the cuts under pressure.
It has no ranking, so the rank goes to whatever shouts loudest. Deadlines shout. Messages shout. The task that has rolled over 6 times shouts loudest of all, and it’s often the least important thing on the page.
The tools made the list easier to keep. They didn’t make the deciding any easier. A task syncs to your phone, laptop, and watch in seconds. Deciding still takes a person and 30 quiet minutes, and those minutes rarely show up on a Tuesday.
The common fixes all add input. A new app sorts the wishes into new folders. Another productivity video adds another framework to sort them with.
Neither one makes the plan defend itself. A plan that never had to defend itself loses to the first surprise.
The model: 3 decisions and 1 argument
A week plan is 3 decisions: which 3 outcomes get your best hours, which smaller tasks get fixed time boxes, and what waits until next week. Everything else on the list is decoration until one of those decisions names it. If you keep one sentence from this piece, keep that one.
The missing step is the argument, and project teams already have a tool for it. Gary Klein, a psychologist who studies how experts make decisions, calls it the premortem. Before a project starts, the team assumes it has already failed and lists the reasons why.
Klein wrote it up in Harvard Business Review in 2007. He cites a 1989 study where people who treated a failure as certain came up with about 30% more reasons for it than people who treated it as possible. More reasons on Sunday means fewer surprises on Wednesday.
A premortem needs a room of skeptics. Your Sunday night has one person in it, and that person wrote the plan.
This is where AI earns its seat. It’s a skeptic that’s free at 9pm, has no stake in your feelings, and works in minutes.
There’s a catch. Chatbots lean toward agreeing with you. Anthropic researchers documented the pattern in a 2023 paper on sycophancy in language models.
Ask a chatbot to plan your week and it writes a tidy fantasy: 6 blocks of deep work, admin squeezed into 20 minutes, no look at the calendar. Ask it to argue against a plan you wrote, and it finds the task that depends on someone who hasn’t agreed yet. It finds the outcome that is secretly 4 outcomes, and the day with 6 hours of work under 5 hours of meetings.
The AI argues. You still decide which points win.
The loop wraps that argument in 5 steps: dump, triage to 3 outcomes, argue, assign time boxes, and close on Friday. Sunday costs 30 minutes. Friday costs 5. The Friday close feeds next Sunday’s dump, so by week 4 you start from 3 weeks of evidence instead of a fresh wish list.
Three outcomes is a default, not a law. It’s roughly what a week of good mornings can hold once meetings take their share. If 3 keeps finishing early, try 4.
If you run your quarters on a longer plan, such as Brian Moran’s The 12 Week Year, make each outcome name what it moves this quarter. An outcome that can’t name one goes to the wait list.
The practice: the 5 prompts
Run the prompts in one chat thread, in order. Keep the thread and the doc, because together they’re the memory of the system. Copy the prompts as written and change only the bracketed parts.
Prompt 1: the dump (5 minutes)
“ Here is my raw brain dump for this week: [paste everything, unsorted] Cluster it into four lists: outcomes (work that produces a result by Friday), admin, personal, and noise. Flag every item with no next action attached. Do not schedule anything. Do not ask questions. “
The dump is where rollover tasks get caught. An item with no next action is a wish, and the flag makes that visible before it costs another week.
Prompt 2: the triage (10 minutes)
“` From the clusters, help me pick 3 outcomes for this week. Ask me these questions one at a time:
- Which outcome, if finished by Friday, makes the other work easier?
- Which outcome has a real date or commitment attached?
- Which one have I delayed for more than 2 weeks?
After my answers, propose the 3 outcomes. Place every supporting task under one of them. List tasks that support none of them under "wait list", each with a one-line reason. “`
Answer each question in one line. Paragraph answers are how a 30-minute triage turns into a diary entry.
The wait list matters here too. Tasks that support no outcome get dated or deleted, and that’s the only step in the ritual that makes the list shorter.
Prompt 3: the premortem (5 minutes)
“ Argue against this week plan: [paste the 3 outcomes and their tasks] Hunt for: a 4th outcome hiding in the task list, tasks with no time estimate, work that depends on a person who has not agreed yet, and days where scheduled work exceeds available hours. Rank the 5 weakest points. Do not compliment the plan. “
"Do not compliment the plan" is the line that does the work. Ask a model to "review my plan" and its default is to approve it. "Argue against" and "do not compliment" give it permission to disagree, and that’s what turns the same model into a useful second voice.
Prompt 4: the assign (5 minutes)
“ Turn the plan into a Monday to Friday schedule: [paste the corrected outcomes and tasks] Rules: hard work before 11am, admin after lunch, every estimate gets a 30% buffer, and no day carries more than 4 hours of planned work beyond meetings. Output a table: day, block, task, outcome it serves. “
The table goes straight into the calendar, outcome column included. A random Wednesday meeting now has to name what it pushes out.
The buffer is borrowed from contractors, who put a contingency line in every quote because jobs run over. Kahneman and Tversky called the habit of skipping it the planning fallacy. A plan with no slack is a wish list with formatting.
Prompt 5: the Friday close (5 minutes, Friday)
“ Friday close. Here is what shipped and what slipped: [paste] Label each slip: planning error (the plan was wrong) or execution error (the plan was fine). Name the one rule that would have prevented the biggest slip. Draft my next brain dump by rolling over only what still matters, with one line on why each dropped item is dropped. “
The close is the loop’s memory, and the planning-or-execution label is what improves next week’s plan. A planning error means Sunday’s decisions were wrong. An execution error means the week won, and the wait list takes the slip without guilt.
When the plan breaks by Wednesday
Some weeks it will. Rerun prompt 3 on whatever is left and move one outcome to the wait list. The repair costs 5 minutes.
A midweek re-argument is cheap. A midweek collapse, which is what the wish list produced every week, is not.
Reality check Expect the first 2 Sundays to run past 30 minutes. Most of the overrun is in triage, when answers turn into paragraphs, and one line per answer brings it back.
The loop records lost deep-work blocks on Friday. It doesn’t stop a bad day from eating them. The wait list also grows in heavy weeks, so give it a 10-minute purge once a month or old items rot there quietly.
The Friday close is the cheapest 5 minutes in the system, and it’s the first thing a busy week skips. Skip it, and next Sunday starts cold.
You can stop here. The model is 3 decisions and 1 argument, and the 5 prompts are the whole system.
Win the argument on Sunday, and Monday just does the work.