7

Systems, not sessions

The endgame: writing your own constitution, so quality stops depending on your memory and starts living in the structure.

A confession about how I worked for far too long: every morning, I re-trained my AI from scratch. Same preferences, same warnings, same context, retyped or half-remembered. "Remember, always check before confirming. Remember, I prefer short drafts. Remember what we decided about how to handle money things."

Some mornings I remembered everything and the work was excellent. Some mornings I was tired, skipped half of it, and the work was mediocre in exactly the proportion I had skipped. It took me an embarrassingly long time to see what that meant: the quality of my entire system was a function of my mood at 8am. The standards existed only in my head, and my head, like the machine's context window, is a sliding window too.

The fix had been staring at me from Chapter 5 all along. My standards were facts living in a bad place. So I gave them a home. I sat down and wrote what I now think of as a constitution: a plain document of durable rules. The verification rule from Chapter 3. The are-you-sure rule from Chapter 4. Where facts live, from Chapter 5. How to debug, from Chapter 6. Plus the small personal things: tone, format, what never to do. And I set things up so the machine reads that document at the start of every single session, whether or not I remember anything at all.

The change was immediate and strangely emotional. The 8am version of me and the exhausted midnight version of me suddenly got the same quality of work, because neither of them was the source of the standards anymore. The document was. When the machine got something wrong in a new way, I did not just fix it, I asked the Chapter 6 question, found the missing rule, and added a line. The constitution grew the way skin grows: at the exact places where the world had scraped it.

And then the deeper thing happened, the one I want you to aim at. Months in, I noticed the system needed me less. Not because it had gotten smarter. Because every lesson I had ever taught it was still there, compounding, while my involvement was no longer the load-bearing wall. I checked in less and trusted more, and the trust was earned, because it was built on rules plus verification, not on hope.

Here is the measure I now use for everything I build, and I mean everything: is it needing me less every month? If yes, it is a system. If it needs me more, or the same, it is a hobby wearing a system's clothes. The whole point of this technology, the entire reason it is worth the trouble of these chapters, is to convert your judgment into something that runs without draining you. Sessions evaporate. Systems accumulate.

Stop here. Actually sit with this before you scroll on.

What do you re-explain every time, and why is it not written down?

Try this nowUnder 30 minutes
  1. Open the Shelf and copy the starter constitution. It is five rules, each one earned in an earlier chapter.
  2. Rewrite each rule in your own words. Word-for-word copies do not stick. Rules you phrased yourself do.
  3. Add two personal rules. Think of the last two times an AI annoyed you. Each annoyance is a rule you have not written yet.
  4. Put it where the machine sees it every time. Most AI tools have a place for standing instructions or a profile. If yours does not, paste the constitution as your first message, every session, until it does.
  5. Live with it for a week, then edit. The first version is a draft. The editing, rule by rule, scrape by scrape, is the actual practice.

Two questions before you go

Answer, then say whether you were sure or guessing. Being honest about which is the skill being trained.