Ask a high performer for their one keystone habit and you'll hear this answer suspiciously often: a weekly review. Not because it's glamorous — it's the least glamorous habit in existence — but because it's the control loop. Without one, you're open-loop: running on momentum and whoever emailed last. With one, errors get corrected weekly instead of discovered quarterly.
The problem with most weekly review advice is that it describes a 2-hour monastic ceremony nobody sustains past week three. This is the 30-minute version that survives contact with actual life. Same time weekly — Friday 4pm (close the week, walk into the weekend clean) or Sunday evening (set the week, sleep better) both work. Calendar it like a client meeting, because that's what it is: a meeting with the only client who's with you your whole career.
The three lists (your operating state)
The review maintains exactly three artifacts — anything more complex dies by February:
- PROJECTS — everything multi-step you've committed to, each with its next physical action written beside it. A project without a next action is a worry, not a project.
- WAITING — everything you've handed to someone else, with the date you handed it over. This list is where dropped balls become visible before they hit the floor.
- SOMEDAY — everything you've deliberately decided not to do right now. This list is a pressure valve: ambitions go here to stop haunting your task list without being abandoned.
The six steps
Step 1 — Collect (5 min)
Sweep every inbox into one view: email flags, chat saves, meeting notes, the capture pad from your deep-work blocks, phone notes, desk scraps. You're not processing yet — just corralling. Loose commitments scattered across six surfaces are the source of that low-grade "I'm forgetting something" hum. This step deletes the hum.
Step 2 — Process to zero decisions (7 min)
For each collected item, one of five moves, fast: Do (under two minutes — now), Delegate (to a person or your AI stack — then log it on WAITING), Date (it's an event — calendar it), List (multi-step — onto PROJECTS with a next action), or Delete (be honest; most of the pile is this). The goal isn't an empty inbox as decor — it's zero undecided items.
Step 3 — Audit the three lists (6 min)
- PROJECTS: does every line have a real next action? Any project untouched for three straight reviews gets demoted to SOMEDAY or killed out loud — zombie projects cost attention every time your eye passes them.
- WAITING: anything aging past its reasonable date gets a nudge scheduled right now, not remembered later.
- SOMEDAY: thirty-second skim. Occasionally something's time arrives. Promote it deliberately or leave it sleeping.
Step 4 — Review the week that happened (4 min)
Scan last week's calendar and your deep-work tally with three questions: What actually moved things forward? What consumed time without producing anything (name the pattern, not the villain)? What does last week want me to change, architecturally — a meeting to kill, a block to defend harder, a task class to automate? One structural fix per week compounds into a different career by December.
Step 5 — Design the week ahead (6 min)
This is the step that makes the other five pay:
- Pick the Big Three — the three outcomes that would make next week a win even if everything else burned. Not thirty. Three.
- Book them into real calendar blocks first, in your prime hours, before the week's noise arrives to claim the space. A priority without a timeslot is a wish.
- Glance at the week's fixed commitments and pre-solve the collisions now (the Thursday deadline vs. the Wednesday travel), while solving is cheap.
Step 6 — Close the loop (2 min)
Six lines in a running doc. In three months this log becomes the most honest performance dataset you own — and the STACK CHECK line doubles as the review loop for your AI stack.
Failure modes (and the patch)
- The skipped week. Life happens. The rule: never skip twice. One missed review is a blip; two is a lapsed system, and restart costs triple.
- The 90-minute creep. If reviews keep running long, your capture is leaking (too much archaeology in Step 1) — fix collection during the week, not the review.
- The guilt session. The review is an engineering meeting, not a courtroom. If Step 4 turns into self-prosecution, you'll unconsciously start avoiding the whole ritual. Patterns, fixes, forward.
- The pretty-system trap. Migrating to a new app every month is procrastination in productivity cosplay. Three lists and a calendar. The tool is not the system; the meeting is the system.
The end state: you walk into Monday with the week already shaped — priorities booked, loops closed, nothing humming in the background. The compounding effect is quiet but ruthless: fifty-two design sessions a year, while everyone around you gets designed by their inbox. Thirty minutes. Same time. Non-negotiable.
The monthly and quarterly layers
The weekly review keeps the machine running; two lightweight upper layers keep it pointed somewhere. Monthly (add 15 minutes to the first review of the month): read four weekly log entries in a row and answer three questions — did my Big Threes actually ladder up to anything? What pattern shows in four STOLEN BY lines? Which single system change (not task) would make next month different? Then check the numbers that only make sense monthly: deep-work hours trend, and your Money OS tune-up if it's quarter-end. Quarterly (a real hour, calendar it): re-choose the 2–3 outcomes that define the next ninety days, prune the PROJECTS list against them ruthlessly — anything serving no quarterly outcome goes to SOMEDAY — and rewrite your context docs (WHO.md changes more often than you think). The weekly review executes the week; the quarterly review decides what the weeks are for. Skipping the upper layers is how people run fifty-two perfect weeks in a circle.
A real Friday, annotated
4:00 — inboxes swept; 14 items corralled, including two voice memos and a scribbled sticky. 4:05 — processing: six deleted, three two-minute replies done on the spot, two delegated to the AI stack (logged on WAITING), three become next actions. 4:12 — lists: one zombie project demoted after three untouched weeks (a genuine relief, which is the tell it was overdue), one WAITING item nudged with a scheduled Monday message. 4:18 — last week's autopsy: Big Three went 2-for-3; the miss traces to Tuesday's meeting pileup — structural fix: decline the standing meeting that produces no actions, effective next week. 4:22 — next week designed: Big Three chosen, three 90-minute blocks booked before the calendar fills, Thursday collision pre-solved. 4:28 — six-line log written. 4:30 — laptop closed, week actually over, head quiet. That last clause is the product. Everything else is the factory.