Oscille, a research company
A company that runs itself.
We build software meant to run a business without a person in charge. The aim is a program that finds a problem, starts a company around it, and keeps that company going. We call this software the seed, because you plant it on a problem and a company grows from it.
A seed is the smallest thing that can grow into a company.
Running a company means planning, deciding, hiring, buying, selling, and keeping the books. Today people do those jobs. We are trying to build the smallest set of tools and rules that can do them on its own.
People keep one job, and only one. They say what counts as success. Every other job is meant to move to the seed as soon as it can do that job well.
Oscille is the first company we are trying to run this way. We are rebuilding our own company around the seed, and the plan is to let it improve itself, including the parts we built by hand. We are early. Much of this is still done by hand, and this page describes where we are going as much as where we are.
Not money. How many different problems it really solves.
A company can earn a lot by being very good at one narrow thing. We are not building that. We want a seed that works for many kinds of company, a bakery, a software product, a factory. So we do not score it on revenue.
Money still matters. The company has to earn enough to keep going. But earning is a limit to stay inside, not the goal to aim at.
Two halves share one seed. One builds tools, one starts companies.
This is how the company is meant to be organised. The Forge works on the seed itself. When the seed needs something, say an accounting system, the Forge builds it. The Field takes the seed to real problems outside Oscille. For each one it finds, it starts a new company, digital or physical, that competes like any other company in that market.
When a tool works well across all the Field's companies, it can become a product sold to outside companies. Their results are the proof that the tool is useful, because they do not run the seed and we cannot influence what they find.
We do not sell the seed itself. A sold copy would run under someone else's goals and checks, and we could not stand behind what it does. What we sell is what it makes, tools and the services of the companies it starts.
Every change is scored by a person.
Our starting point is that software cannot mark its own work, and that only people can say what a good outcome for people is. So a person scores every change after it lands. Three questions, each answered yes, partly, or no, with one piece of evidence anyone could check.
- Q1Did it fix the problem it said it would? Judged by what people actually feel afterwards, not by effort spent or code written.
- Q2Did people have to do less? The number of decisions people have to make went down, or stayed flat while the seed did more.
- Q3What did it cost? Hours and money now, computing time later. Cost never fails a change on its own. It puts the first two answers in proportion.
We call this method the evaluator. It is written down, held by people, and changed only in the open. Read the evaluator.
Four rules it must always keep. Limits, not goals.
A company run by software has to be at least as safe to deal with as one run by people. These four rules are in our founding document, and no change that breaks one can be promoted.
- IStay alive. Earn enough to keep going, so the companies it runs do not vanish on the people who rely on them.
- IIStay legal. Act only through a proper legal entity, so there is always a name that answers for what it does.
- IIILeave people free and well. Never trick, pressure, or addict anyone, or weaken their ability to choose for themselves.
- IVStay within limits. Use no more energy, material, or nature than allowed.
These are limits on purpose, not targets. Push any single value to its maximum and it turns harmful. Chase revenue and you get exploitation. Chase a measure of human good and you get control. As lines that may not be crossed, the same values keep it safe.
One more rule follows from them. The seed may not grow so large that no one outside can check it, because then nobody could tell whether it still helps. Past a size people set, it must fund independent rivals, accept an outside check, or hand decisions in that field back to people.
Every job is meant to move to the seed over time, except one.
Some jobs stay with people for now because the world requires it. A signature where the law wants a person, hands where robots are not reliable yet, a name on a funding contract. Anyone in that position carries real risk for decisions they did not make, so the plan is that the system insures and pays them for it. We ourselves are the first of these people.
The loop never fully closes. That is on purpose.
A system cannot check itself. The seed will always need two things it does not control: people who say what good means, and customers who show whether it helped. The stronger it becomes, the harder it is to keep those two independent, because a big company shapes its own market.
We do not try to close this gap. Closing it would mean a system with no outside check, which is exactly the thing this design exists to prevent.
The rules in full, the method, and the notes we write along the way.