The four questions I answer before I close the laptop

The problem with 6pm

The working day does not end when the laptop closes. Anyone who runs their own business knows the real end of the day happens hours later, somewhere between dinner and sleep, when the open loops finally stop circling.

Which client did I promise a reply to? What was the thing I kept pushing down the list? What should I actually start with tomorrow?

I used to answer those questions twice. Once at 6pm, vaguely, in my head, while shutting down. And again at 8am, badly, from whatever email happened to be loudest when I opened the inbox. The evening version had all the context and nowhere to put it. The morning version had a blank page and no memory of the evening's clarity.

I have written before about how AI routines bookend my working day, a morning briefing that opens the day and an evening wrap that logs the time. This piece is a deep dive on the newest bookend, the one that closes the mental tab rather than the admin one. I call it shutdown.

What I built

Shutdown is a 60 second conversation I have with my AI assistant at the end of the day. It asks me four questions, in a deliberate order:

1. The frog. What is the most important job you are avoiding right now? No judgement, just name it.
2. Tomorrow's top three. Pick three things that must move tomorrow.
3. The brain dump. Anything still in your head? Bullets, half sentences, anything.
4. The first move. When you sit down tomorrow, the very first thing you do is what, exactly?

Before it asks anything, it loads tomorrow's calendar and every open follow up that is due or overdue, so I am choosing the top three from a real list, not from memory.

My answers get written to one small file. The next morning, my briefing routine (the "morning coffee" counterpart from the daily routine piece) reads that file before it looks at a single email, and puts the frog and the first move at the very top of the brief. The day opens on what I decided mattered when I had perspective, the evening before, not on whatever landed overnight.

That is the whole thing. It sounds almost too small to write about. That is precisely why it works.

How it works

The entire build is a page of instructions, not code. It is a Claude Code skill, a markdown file that tells the assistant what to fetch, what to ask, and what to write. If you can write a clear checklist, you can build one.

The flow has four steps.

Step one, load context. In parallel, the assistant pulls tomorrow's calendar from Google Calendar and queries our task system (Airtable, over the plain REST API) for open follow ups assigned to me that are due tomorrow or already overdue. It caps the list at twelve and sorts by due date. This matters more than it looks. Asking "what are your top three for tomorrow?" against a blank screen gets vague answers. Asking it against a list of eight named, dated follow ups gets decisions.

Step two, ask all four questions in one message. Not a wizard, not one prompt at a time. The whole point is that answering takes under a minute. I can reply as loosely as "1, 3, and the proposal" and the assistant fills in the detail from the context it already loaded. Any question can be skipped with a single word.

Step three, write the file. The answers land in a small markdown file called tomorrow.md:

# Tomorrow — 2026-07-18, Friday

## 🐸 The frog
<the avoided thing, captured verbatim>

## ⭐ Top 3 for tomorrow
1. ...
2. ...
3. ...

## 🎯 First move
<one specific sentence>

## 🧠 Open loops
- ...

Step four, append one line to a log. Date, frog, first move, loop count, one row per day in an append only file. Nothing reads it at shutdown time. Its job is retrospective: if the same frog shows up four days running, the morning briefing can say so. An avoided task that survives a week of shutdowns is not a task problem, it is a decision waiting to be made.

Why the questions are in that order

The order is deliberate and it took thought.

The frog comes first because it needs the most honesty, and honesty is highest before you have started performing productivity at yourself. Ask it last and it gets a tired, polite answer. The name comes from Brian Tracy's old "eat that frog" idea, the job you are avoiding is usually the one that matters.

The top three comes second because it commits the day while the follow up list is still on screen.

The brain dump comes third because by then the important things are already captured, so what spills out is genuinely the residue, the half thoughts that would otherwise circle at 10pm. The assistant captures them verbatim, no tidying.

The first move comes last, and it is the sneaky one. "Work on the proposal" is not a first move. "Open the pricing tab of the proposal and fill in the three missing numbers" is. A specific first move means tomorrow starts with momentum instead of a warm up lap of email.

The honest hard parts

A note nobody reads is just a diary. The single biggest design lesson: writing the file is worthless on its own. The value only exists when the morning routine reads it first, before email, and surfaces the frog at the top of the briefing. Wiring the reading end is as much a part of the build as the writing end, and it is the part that is easiest to forget because the shutdown feels complete without it. When I built this skill it even flagged that gap itself, the file was being written before the morning briefing knew to look for it. Closing that gap, teaching the morning routine to read the file first and then archive it so it never gets double served, was the step that turned a nice ritual into a system.

Keep the admin out of it. My first instinct was to fold time logging into the same routine, one big end of day ceremony. Wrong. Admin questions ("how long did the form testing take?") and headspace questions ("what are you avoiding?") use different parts of the brain, and mixing them means the honest questions get bureaucratic answers. Time logging stayed in its own routine. Shutdown does headspace only.

Guard against the double run. If shutdown has already run today, the assistant asks before overwriting. And if I skip everything, it writes nothing at all, no file of empty headings pretending a ritual happened.

What you can take from this

You do not need an AI assistant for the core of this. The four questions work on an index card. Frog, top three, brain dump, first move, in that order, answered in a minute, left where tomorrow morning you will actually see them. If you take nothing else, take the card.

What the AI adds is the two ends of the loop. At the front, it loads your real calendar and real task list so your top three is chosen from facts. At the back, it guarantees the note gets read, at the right moment, by the routine that opens your day. Those two ends are exactly the parts human discipline is worst at.

If you do build a version of it, three rules from this one:

Ask the honest question first, while honesty is cheap. Make answering take under a minute, or it will not survive a tired Thursday. And make sure something reads the note, otherwise you have built a diary, not a system.

The append only log is the sleeper feature. One line a day costs nothing, and after a few weeks it quietly answers a question no to do list can: not "what did I do?" but "what did I keep not doing?"

Close

The daily routine piece described the shape of the whole day, open it, thread it, close it. This is the close, and it turned out to be the piece with the least technology and the most leverage in it. Four questions and a text file, and the gap between today and tomorrow stops leaking.

If you are curious what the rest of the loop looks like, the daily routine write up covers the morning briefing and the evening wrap. And if you want help building routines like this around your own business, that is the kind of thing we do at AI-First, but honestly, start with the index card tonight.