Build the sharpest AI setup you can. Feed it your calendar, your notes, your goals, your last fifty calls. Then leave it alone for four weeks and watch it hand you last month's plan with full confidence.
The model didn't get dumber. Your context did. A second brain captures a moment, and the moment moves. Without something that points the system at what matters this week, you end up with a very smart assistant arguing for priorities you dropped in May.
That something is the loop. In an AIOS, the AI operating system you own, it's the fifth and final layer. The one most people skip, and the one that decides whether the other four stay alive.
A system without a loop is a snapshot.
Snapshots rot. The week moves and the file doesn't.
Your memory layer is a record of what was true when you wrote it. Brilliant on day one. By week three it's quietly lying. The project you killed is still listed as active. The client who went cold is still "warm." The deadline that slipped is still next Tuesday.
Here's the trap. The richer your context, the worse stale context hurts. An empty ChatGPT knows nothing about you and admits it. A loaded system that's three weeks out of date sounds authoritative and is wrong. It will reason beautifully from facts that expired.
I've watched this happen with founders I work with. They spend a weekend setting up the perfect notes, the perfect dashboard, the perfect prompts. Two weeks later they stop opening it, because it stopped matching reality and nobody told it to update. The setup wasn't the problem. The missing loop was.
A second brain you never update isn't a brain. It's an expensive diary.
Steal the weekly review from GTD.
David Allen solved this in 2001. The fix is a standing appointment, not a better app.
If the weekly loop feels familiar, good. It's lifted almost wholesale from Getting Things Done. Allen built his whole method around one recurring habit he calls the Weekly Review, and he's blunt that it's the make-or-break step. Skip it and the rest of the system decays into a pile of stale lists. He recommends running it every 7 to 10 days, without exception.
The structure is two moves, and they map straight onto an AIOS.
First, get clear: empty every inbox and get the loose stuff out of your head and into the system. Allen's own checklist calls this step "empty your head," processing any uncaptured projects, actions, and waiting-fors so nothing is hiding in your skull. For an AIOS that means feeding in the week: the calls, the decisions, the new commitments, the things that changed.
Second, get current: walk your project lists, your calendar, your waiting-for list, and bring each one back to what's actually true. Mark what's done. Cut what's dead. Move what slipped.
That's it. Allen wrote it for paper and folders. The mechanism doesn't care. Run the same loop against your AI's context every week, and the context stays honest. Stop running it, and you're back to arguing with a confident, outdated machine.
The weekly review was never about productivity theater. It was about keeping a system in sync with a life that won't hold still.
“The model isn't what goes stale. Your context is. And context is the one thing only you can refresh.”
What the loop does for the AI, and what mine looks like.
It tells the system what matters now, so Monday doesn't start from a blank page.
Here's mine, concretely. Every Sunday I run a routine I call EXIT7. It walks the week across seven areas of my life, body, mind, family, network, business, money, environment, and asks one question per area: did this week move me toward what I said mattered, or away from it? The answers get written into a single file the AI reads at the start of every session. I keep that file under about 14 KB on purpose. The moment it sprawls, the AI starts treating noise as signal.
Then the loop closes the other way. The system reads the refreshed file and re-aims itself. Next session it isn't guessing from a month-old plan. It's working from this week's real priorities, this week's real friction.
Last Sunday it drafted my weekly review for me. I read it, changed two lines, and that was the whole job. First time in a long time I didn't start from a blank page. The thing wrote the first draft of my own week back to me, and it was mostly right, because I'd been feeding the loop.
That's the payoff people miss. The loop isn't admin you do for the system. After a few weeks it starts doing the review with you, then mostly for you. The morning brief, the Sunday recap, the "here's what's drifting" nudge: none of that works without a current context, and context only stays current if something refreshes it on a schedule.
This is also why the loop sits on top of the other four layers instead of beside them. Your constitution sets the rules. Your memory holds the facts. The loop is what keeps the facts from going off. Without it the whole stack quietly drifts out of date, and you're back to a blank prompt and a shrug.
One file, once a week. Cheaper than the model. Worth more.
If you've built up an AI setup that was great for a fortnight and then went quiet, the missing piece is almost always this one. Start with a weekly habit, even a rough one: ten minutes on Sunday, the week written down, the context refreshed. Build the rest later.
I'm putting together a small group of founders building their own AIOS, loop included. Join the waiting list. And if you haven't yet, read how the memory layer works first: What is a second brain, the thing this loop keeps alive.
- The 11 Steps of the Weekly Review, Getting Things Done (David Allen Company). Official description of the GTD Weekly Review, the get clear / get current / get creative structure, and the every-7-to-10-days cadence.
- GTD Weekly Review Checklist (PDF), Getting Things Done (David Allen Company). The official checklist, including the "Empty Your Head" and "review project and action lists" steps referenced above.