The file that tells me to stop building

The Workflow

Every morning starts the same way now. I open one file before I open anything else. Not an app, not a dashboard. A single page of plain text that sits at the front door of my whole business. At the top is the only rule that isn’t allowed to move: every day, do something that grows the email list. Publish, or build the thing that publishes. Everything under that rule exists to answer one question: what do I do right now.

Step 1 — Open the one file first. Before email, before Slack, before the tempting new idea, I open [0000] - Daily Instructions.md. It’s the front door. If I skip it, the day picks its own priorities, and the day has bad taste.

Step 2 — Two minutes to orient. I say “good morning” to my AI. It reads my logs, my standing commitments, and my calendar, then hands back a short list: today’s date, the three or four things that actually have to happen, and one suggested first move. I don’t plan the day. The day is already planned. I just have to read it.

Step 3 — Work down the list in order. It’s ordered, and I don’t get to pick the fun one first. Most days the whole job is one piece of content out the door and one commitment cleared. That’s a good day. I spent a year believing a good day meant building something impressive. It’s doing the boring thing again, on the morning I didn’t feel like it.

Step 4 — Ask the one question when the pull hits. When the day starts to feel like too much and I catch myself wanting to “quickly set up” something, I stop and ask: is this publishing or selling, or is it setup? Publishing or selling, go. Setup on a day that isn’t my build day, don’t.

Step 5 — Tinker after, not instead. Once the have-tos are done, I build whatever I want. The file doesn’t ban tinkering. It just makes me earn it.

That’s the entire system. A page of text and a question. I kept waiting for the part where it had to get more complicated. It never came.

What Broke

The same week I leaned on that file, I built the exact thing it warns me about.

I spent two days on what I started calling the dashboard to end all dashboards. A command center for a one-person business. Live chat windows wired to my AI tools, a goals list, a journal, and a bar across the top that pulsed like an oscilloscope every time the AI was working. It looked incredible. I was proud of it.

Then, instead of asking the AI to make it better, I asked it whether it should exist at all. Be honest, not nice. The answer took about thirty seconds. The chat window did the same job as a tool I already had, worse. The goals and the journal lived in two other places already. The part that wrote to my notes duplicated something I already ran, and that overlap could quietly corrupt the one asset I can’t afford to lose. The pulsing bar was, and I am quoting the review here, dopamine. Two days to rebuild three tools I already owned, wrapped in a light show.

So I froze it. I wrote a short note at the top of the project that says, in plain words, do not add features to this. It still runs. I just don’t touch it.

The lesson landed flatter than I wanted it to: the tool that tells you no is worth more than the one that builds whatever you ask.

Time Ledger

• Time saved: Two minutes each morning to know exactly what has to happen, instead of the twenty or thirty I used to lose deciding, re-reading yesterday, and talking myself into the wrong task first.
• Time added: Two full days building the dashboard the file was supposed to stop me from building. Net value shipped to a subscriber or customer from those two days: zero.
• Net: The file itself cost one afternoon and pays that afternoon back every week. The dashboard was the real bill, and I paid it the same week I wrote the rule against it.

The number that matters isn’t the two minutes or the two days. It’s that the two-minute version is the reason anything got published this month, and the two-day version is the reason I almost didn’t.

The Prompt File

Two things make this work, and you can steal both.

The first isn’t a prompt for your AI. It’s a rule for you, sitting at the top of your own version of the file:

Before I start anything: is this publishing or selling, or is it setup? If it’s setup and it isn’t my dedicated build day, it’s probably avoidance. Do the next thing on the list instead.

The second is the prompt I run whenever I’ve built something and I’m not sure it should live. Don’t ask your AI to improve the thing. Ask it whether to kill it.

PROMPT FILE

You are a blunt operations advisor, not a cheerleader. Tell me whether
something I built earns its place, or whether I talked myself into it.

Inputs:
WHAT_I_BUILT: [describe the thing in plain terms]
MY_REAL_CONSTRAINT: [the one thing that matters right now, e.g. grow my email list]
MY_BUILD_DAY: [the one day a week I'm allowed to build, e.g. Saturday]

Do this:
1. Judge it against the constraint. Does WHAT_I_BUILT move MY_REAL_CONSTRAINT
   forward directly? If it only feels productive, say so.
2. Check for duplication. List anything I already own that does part of this
   job. If it rebuilds a tool I have, name the tool.
3. Name it. "Keeper" or "distraction" in one word first, then two sentences of why.
4. Point me back. If it's a distraction, tell me the actual have-to I'm
   avoiding by building this.

Be honest, not encouraging. If it's a distraction, say so plainly. I would
rather hear it now than after another two days.

A good AI teammate will build whatever you ask. A better one tells you to go write the post instead.

Manager’s View

Ask yourself, or the busiest person on your team: how many full days last month did you work hard, heads down, all day, and ship nothing a customer would ever see?

Most people can’t answer it, because we track effort, not output. A day spent building an internal tool, tidying a system, or reorganizing a workflow feels identical from the inside to a day spent shipping. Both end with you tired and the screen full. Only one of them moved something a customer will ever touch. The gap between those two kinds of days is where a solo operator’s month quietly disappears.

To actually ask it, don’t lean on memory, because memory grades on effort. Open last month’s calendar and your real output side by side. Count the days that produced something a customer saw: a published thing, a sent thing, a sold thing. Then count the days that only produced setup. If the second number is bigger, the problem isn’t that you’re lazy. It’s that building feels like progress, and nobody put a rule in front of you to catch it.

Field Notes

• The dashboard has exactly one piece of documentation: a FREEZE.md I wrote at the top of the project that says, in four lines, do not add features to this. It’s the only doc I’ve ever written for an audience of one, me, next week, when I’ve forgotten why.
• The most useful thing I made this year is a page of plain text. The version with the live oscilloscope and the voice interface sits frozen two folders over. I still open it sometimes, like checking on a car I’m not allowed to drive.
• The best move my AI made all week wasn’t writing or building. It was about thirty seconds of “this doesn’t earn its place, go write the post instead.” I had to tell it to talk to me like that. It won’t volunteer the honesty you didn’t ask for.