Grown, Not Generated

A law from 1975

Grown, Not Generated

Every complex thing that works descended from a simple thing that worked. AI just made it free to skip the ancestor. It did not make the ancestor optional.

Gall’s law

Every complex system that works has a simple ancestor that worked.

The design can be written in an afternoon. The working cannot.

John Gall, a pediatrician in Ann Arbor, Michigan, in General Systemantics (1975), a small, half-joking book about how systems fail. The other half of the law: a complex system designed from scratch never works, and cannot be patched until it does. You start over from something simple that runs.

John Rector · johnrector.me October 26, 2026 ≈ 7 min read

The pediatrician’s law

In 1975 a pediatrician named John Gall published a slim book about why big systems so rarely do what they were built to do. Most of it reads like satire. One line in it has outlived the joke: a complex system that works is always found to have grown out of a simple system that worked. Its twin is harsher. A complex system designed from scratch never works, and you cannot patch it into working. You have to go back and start with something simple that runs.

Gall was careful about the direction. He did not promise that simple things work; plenty do not. The guarantee runs the other way. Find anything complex that works — a city, a language, the internet, the spreadsheet with the DO NOT DELETE tab — and you will find a simpler working ancestor underneath it, and another under that.

Why it is true by structure

Every system is a bundle of guesses about the world. What customers will type into the box. Which step people will skip. What happens the day the supplier is late, the day the manager is out, the day two orders arrive with the same name. A design is nothing but guesses. The world is the only thing that answers them, and it only answers when the thing is running.

A simple system asks the world a few questions at a time. When it breaks, you know which guess was wrong, because there were only a few to choose from. You fix it, and that answer is now built in. The next generation adds a handful of new guesses on top of a body of answered ones. That is all growth is: questions asked a few at a time, with the answers kept.

A complex system designed from scratch asks every question on day one. The failures arrive together and tangle. Three wrong guesses produce one symptom, and nobody can tell which guess to fix. That is why Gall said it cannot be patched. Every patch is one more unanswered guess, added to a pile that is already failing. It is dead reckoning with no landmark ever coming into view.

A design is a list of questions. Working is the list of answers. Only the world writes the answers.

The bodyguard is gone

For most people, cost enforced Gall’s law without anyone naming it. A small business could not afford to design the whole system, so it grew one: a notebook, then a spreadsheet, then a second tab. Only large organizations could afford to break the law, and they paid for it on go-live days.

That bodyguard just left. Tonight anyone can generate the complete thing — the full app, the forty-page handbook, the twelve-agent workflow, the policy manual for every situation. Complete now costs the same as simple. Which means everyone can make the big organization’s mistake, on a weekend, alone.

Here is the part most people miss. Gall’s law was never about the price of building. It was about the number of unanswered questions. AI made the design free and left the answering exactly as expensive as before, because answers are not generated. They arrive. In my own language: a design is pure Expectation, and only an Actual can teach. Actuals come one event at a time, and no amount of compute brings tomorrow’s customer in early.

So two things happen at once. A flood of generated systems — handsome, complete, never run. And, in the hands of people who understand the law, generations that used to take a quarter shrinking to an afternoon, because the same machine that writes the design also makes each trial nearly free. AI does not let you skip the ancestor. It lets you raise twenty of them in a week.

Figure 01

Two ways to arrive at something complex

Grown · a few questions at a time

1. One rule, one user answered
2. Plus the exception that broke it answered
3. Plus a second kind of user answered
4. Plus the report someone asked for answered
5. Complex, and working answered

Generated · every question at once

1. Every rule unanswered
2. Every exception anyone imagined unanswered
3. Every kind of user unanswered
4. Every report unanswered
5. Day one: all of it meets the world together tangled

Five predictions

1. The launch disappears.

   Go-live day goes the way of the grand opening for a website. Software arrives as a seed that works the afternoon you ask for it and grows every day after. “Version 1.0” starts to sound like “a roll of film.”

2. The best AI will refuse to build everything at once.

   Ask for the whole system and a good assistant will hand you the smallest version that runs, and tell you why. It will feel like stubbornness until it becomes the mark of quality. The assistants that obey you literally will produce the most beautiful wreckage.

3. Generated complexity becomes the new debt.

   A wave of complete, never-run systems is coming: handbooks, apps, agent teams written in a weekend. They will not be fixed; Gall said they cannot be. The cleanup job of the decade is finding the smallest part that actually runs and growing from there.

4. Lineage becomes the asset.

   When anyone can generate the design, the design is worth nothing. What cannot be generated is a history of working — every generation that met the world and survived it. Buyers of businesses, software and methods will ask one question first: what did this grow from? The growth log will be worth more than the spec.

5. The ugly working thing gets promoted.

   The spreadsheet with the DO NOT DELETE tab stops being the thing a project replaces and becomes the seed it grows from. Migrations run backward: instead of building the new system and moving people onto it, the AI grows the old working thing until it is the new one.

Living it

I watched this happen in my own work yesterday. When my AI built the first working version of my SIM feed — the personal-AI feed I have been writing about — it filled it with ten small, tactical cards. I rejected them within the hour as too short-sighted. That was not the project failing. That was generation one meeting the world, and the world was me. The second generation carried eight job-level cards, and it was better because the first one existed to be wrong.

The writing grew the same way. The Feed Turns Around, then Try It Is the Business Model, then a build guide I corrected twice, then Tab, Enter, then Add One Egg. Nobody, me included, could have written that series from an outline. Every part was an answer to what the last part got wrong. I did not plan a series. I grew one.

The claim ledger

Already true

Gall wrote the law half a century ago, and every working complex thing you rely on has a lineage you could trace. AI can already generate a complete app, handbook or workflow in an evening. And you have lived a go-live day, which is why the opening scene needed no explaining.

What has to happen

Assistants have to learn to offer the smallest working version first, and people have to learn to want it. The loop has to close: the AI must see where the little system broke — the record of what actually happened — not just hear the next request.

Where I am probably wrong

Simulation. If machines get good enough at simulating people that most of a system’s questions can be answered before a real person ever touches it, Gall’s law weakens. The generations would happen inside the simulator, and the design would arrive already grown. I think the questions that matter most — what a tired person actually does when a screen asks them something — stay outside any simulator for a long time. If I am wrong, the launch comes back, and this piece ages badly.

Go back to 10:40 on go-live day. The spreadsheet is still open on the supervisor’s screen. It was never impressive. It has no steering committee and no training week. It just never stopped working, and that makes it the most valuable thing in the building. When you hand the problem to your AI, hand it that.

Start with the thing that works.

Background

• John Gall, General Systemantics (General Systemantics Press, Ann Arbor, 1975); republished as Systemantics (1977) and later as The Systems Bible (2002). Overview at Wikipedia.
• John Rector, “Time Since Last Fix” · “The Second-Cheapest Way to Find Out”
• John Rector, “Every App Is an Org Chart” · “The Machine Has Old Taste”
• John Rector, “The Feed Turns Around” and the series that grew from it.