The AI revolution does not start with a decision from the top. It starts where no approval process is needed.
3-07 looked at how Japan's SIer industry turns over. This chapter says where that turn actually begins. I will lay out, in order, why I see it that way.
What AI is good at is work whose answer can be checked
Anything with fixed rules, where a machine can check the answer, is something AI becomes good at. Go was like that. So is the work of Word and Excel, and so are core business systems. The form is fixed, and a machine can judge whether the result is right.
That is not where the difficulty is. *The difficulty is deciding what the answer should be, and keeping the thing running.* Deciding what to build. Keeping what was built usable on the floor by fixing it, continuously. Neither of those has a correct answer a machine can check.
So it goes like this. The inside of software becomes AI's world. People move outside it — out to where there is no answer, and decide the answer there.
The solo developer becomes a builder
A solo developer who does nothing but software gets replaced by AI, because that is the domain where answers can be checked. Instead, become a builder. Work in a community.
There is a gap right now, and it is a clear one. *Almost nobody can put OSS into an organization and carry it through to running it.* SIers will not do it. There is no money in it: work with no license margin and thin billable hours cannot be taken on inside that structure (3-04).
So builders have to do it. And for a solo developer, this is revenue. Install it. Maintain it. Fix the forms. Build the screens. None of it can be done without going onto the floor.
It is only putting the floor back to what it should be
I do not want to be misread, so let me put it plainly. I am not saying build something new. *AI can do what the SIer and the IT department did. So all that is left is putting the floor back to what it should be* — making it possible to develop software on the floor. That is all.
The headcount then moves like this. The internal IT department is not needed, because AI arrived. Meanwhile the floor has its own specific problems, and AI does not solve all of them. So the floor needs more people, not fewer. Even in today's system development, people from the floor are required. Nothing works unless the floor gains people.
Change starts in the development and planning departments
Inside a company, change the constitution starting from the development and planning departments. The IT department lets go of its authority. Cut the IT department's people and budget; increase the people who build in development and planning, the people on the floor, and their budget.
One thing needs to be explicit. What moves is the budget, not the people. Moving IT department staff onto the floor does not work. The dependence on Microsoft is too deeply learned. How tools get chosen, and the assumption that things get outsourced, stay in the shape they were learned in.
Say "cut headcount" and the company is left with nothing
IT companies say "bring in AI and cut headcount." The moment that is the phrasing, the outcome is fixed.
For people on the floor, using AI in hiding becomes the better move. Using it openly means handing over the material for eliminating your own job. Using it in hiding means the time saved is yours. It is a rational choice.
And what is used in hiding does not stay with the company. The code that was written, the procedure that was learned, the fix for the thing that got stuck — all of it stays in one person's hands and disappears. *A company that said "cut headcount" gets nothing out of bringing in AI.*
Say "the floor gains people" and it reverses. AI gets used in the open, and it stays with the company.
Six things stay with the company
Used in the open, what stays? Six things.
- The code — in git, and daily operation runs on the code alone
- Business knowledge put into words — what was in someone's head, in readable form
- The checking mechanism — a form in which a machine can judge whether it is right
- A record of what got stuck and how it was fixed — so the next person does not stop there
- People who can build — someone inside who can fix it without going outside
- The data — on your own side, and portable
These six survive cancelling the AI subscription. Put the other way: any way of using AI that does not leave these six behind never becomes the company's property, however much is spent.
If the company will not allow it openly, use it in the department
If the company will not allow it openly, do not wait. Use it in the department. Used in a department, the six stay inside that department. This is where the AI revolution begins.
No company-wide agreement is needed, and no company-wide policy. What is needed is the department's own discretion.
All it takes is a line and a hundred-odd dollars a month
Two things are needed. An internet line, and a hundred-odd dollars a month in subscription fees (2-02). No approval process, no server budget, no per-seat licenses. Once it is in steady operation, $40 a month is enough, as 2-02 sets out.
So the audience is clear. Anywhere this can be done without an approval process.
- Small and mid-sized companies, and sole traders
- Departments inside large companies that have real discretion
- The periphery of large companies: subsidiaries, regional sites, new businesses, research
None of them has a reason to wait for a sign-off. They can start the day they decide to.
Adopted from the top, it always fails
Top-down adoption always fails. The reason is in the sequence.
Top-down adoption goes through the IT department first. Then through the SIer. There, the story becomes "bring in AI and cut headcount." Once that happens, the floor uses it in hiding. The company is left with nothing.
Only what starts from below spreads while leaving the six behind.
Executives have no time to look at small problems
The same thing happens inside a company as in a government.
What an executive at a large company thinks about with AI narrows to efficiency. Not out of ignorance. There is no time to look as far down as the small problems. With no time, the only thing that can be decided is a number: cut this percent, save this much.
The people who can spend time on small problems are on the floor. And *small problems are exactly what AI can solve* — fixed forms, repeated procedures, one-off exceptions. So AI belongs to the floor.
Held at the center it becomes repression; held below it solves problems
One level up, the same shape appears.
If the side at the center — the government — holds AI, it is used for surveillance, censorship, and propaganda. That is how I see it. Looking at the United States and Russia now, governments are not using it any other way. The digital sovereignty argument in 3-03 was about control from outside the country; the same structure exists inside it.
So AI is held below, on your own side. Run the model on your own machine and keep the data from leaving (2-16). This is not a matter of technical taste. It is a matter of who decides what it is used for.
The money stops going to Microsoft
The flow of money changes too. What was being spent heavily on Microsoft 365 becomes unnecessary, and more convenient at the same time. The destination moves from Microsoft to Anthropic and OpenAI. *This is one of the essentials of the AI revolution.*
How it is paid changes as well. From everyone paying every month, to a builder paying for what development and maintenance actually use. It stops stacking up per head.
On 2026-09-16 this took concrete form. Document and slide creation moved inside Claude (2-07). The reporting treats it as a move against Microsoft (Axios, checked 2026-09-20).
So AI produces a revolution
Put it together.
AI produces a revolution not through a decision from the top and not through a large budget. It does so because *anywhere that can act without an approval process starts with a line and a hundred-odd dollars a month, and what works spreads.*
There are countless places for it to start: small companies, sole traders, departments inside large companies, subsidiaries, regional sites. And *there is no hand that can stop it.* Nobody can stop a thing that needs nobody's permission.
Examples
Ukraine — it spread from the front line and the workshops
Ukraine was invaded by Russia in February 2022. It is the side defending its own country. Inside that defense, things built from below spread to the whole army.
- FPV drones. Front-line units, volunteers, and small workshops built them from commercially available parts and repaired them in the field. In kitchens and basement workshops, at roughly $300 per drone. A maker says there are at least 15 similar workshops in Kyiv alone (Inkstick)
- The scale. Production passed two million a year in 2024. The defense minister has said the forces are on track to receive three million in 2025 (Kyiv Independent)
- Delta. The system for sharing the battlefield picture was started in 2016 by the volunteer group Aerorozvidka. It became operational in a limited configuration in August 2022, moved to the defense ministry's technology centre in 2023, and was formally adopted in August 2024 (Delta)
Top-down procurement does not produce that speed. Deciding takes time, and by the time it is decided the adversary has changed. All of the above was checked on 2026-09-20.
What I did myself
I do not want to leave this at the level of large examples, so here is what is in my own hands.
- flet-printing — added the printing that Flet did not have
(
aiseed-dev/flet-printing) - aiseed office — forms that only opened in Word now open the same way on Linux
(
aiseed-dev/officework, 2-07) - Running Windows in a virtual machine inside Debian — done on 2026-09-20. Nothing got stuck
None of it went through an approval process or a budget. A line and a monthly subscription were enough.
Summary
- What AI is good at is work whose answer can be checked — the difficulty is deciding the answer and keeping it running
- It is only putting the floor back to what it should be — no IT department, more people on the floor
- Say "cut headcount" and it gets used in hiding — the company is left with nothing
- Six things stay — code, business knowledge in words, the checking mechanism, the record of what got stuck, people who can build, the data
- A line and a hundred-odd dollars a month — it starts where no approval process is needed
- Top-down adoption always fails — it goes through the IT department and the SIer and becomes headcount reduction
- AI belongs to the floor — executives have no time for small problems, and small problems are what AI solves
There are countless places for it to start, and no hand that can stop it. The next chapter looks at why this transition does not reverse, as a structure.