30 years of open source

In August 1991 a student in Helsinki posted a message to a discussion group with a line that has aged better than almost anything else written that decade: he was building a free operating system, just a hobby, nothing big or professional. He was twenty-one. The project was not expected to matter.

I like this story for an unfashionable reason. It is not a story about genius. It is a story about somebody who wanted a thing to exist, could not buy it, and decided that was not a good enough reason to stop. That instinct is the one I have built a business around, so I am hardly objective about it.

What was actually being solved

To understand why anyone bothered, you need the state of things at the time. Serious operating systems were commercial products from large companies, and they came with licence terms and prices that put them out of reach of a student. The academic alternative available for study was deliberately small and limited, precisely because it was a teaching tool rather than a working system.

So the gap was specific: no free, working kernel that ran on the ordinary hardware people actually owned. Not a philosophical gap — a practical one. And the answer was not a manifesto, it was code, released for anyone to take, use and improve, under terms that required improvements to be shared back.

That licensing choice mattered more than the code. Plenty of clever projects from the same era are gone. The ones that survived generally had a rule that made contributing rational: if you improve this, your improvement stays available to everyone, including the next person who improves it further. That is not idealism. It is a mechanism for stopping the commons from being quietly fenced off.

The part nobody predicted

Version 1.0 arrived in 1994. Through the second half of that decade the project stopped being a curiosity and started being infrastructure — quietly, in server rooms, where nobody outside the trade was looking. The desktop, which was supposed to be the prize, never really came. Instead something stranger happened: the kernel went everywhere else.

Today it runs the machines behind most of the web. It runs inside Android, which means it runs inside the majority of the phones on the planet. It runs on every one of the fastest supercomputers in the world — not most of them, all of them. It runs in televisions, cars, routers, cameras, cash machines and industrial controllers. There is a reasonable chance you own five or six devices running it and have never once thought about it, which is the highest compliment an operating system can receive.

The people who lost this argument in the 1990s were not stupid. They were reasoning from a sensible premise — that serious software needs a company behind it, with paid staff, quality control and someone accountable. What they got wrong was assuming those things require ownership of the code. It turned out you can have paid staff, quality control and accountability while the code itself belongs to nobody. Most of the kernel's work today is done by people employed by companies. They are paid to improve something none of their employers owns.

Why a story from 1991 is on a shop's website

Because it is the reason my business is possible, and I would rather say that plainly than pretend otherwise.

When I take a computer that a vendor has declared finished and put it back into service, I am not performing a clever trick. I am using three and a half decades of accumulated work by people who never met me and will never see a złoty from what I do. The drivers that make a 2016 machine talk to a modern network are there because someone maintained them. The file system that keeps a client's photographs intact is there because a team of people argued about it in public for years until it was right.

That is a debt, and it is worth naming. The whole refurbishment trade — everyone extending the life of hardware rather than replacing it — exists on top of software that somebody gave away. Without it, a machine outside its vendor's support window really would be finished, and the only honest advice would be to buy a new one.

Three moments when it could have failed

Hindsight makes everything look inevitable, so it is worth marking the places where it was not.

The first was the licence. Early on the terms were different and more restrictive, and they changed to a copyleft licence within the first year. Had that not happened, companies would have had no confidence to build on it and contributors would have had no guarantee their work stayed available. A great deal turned on a decision made by one person who could have gone the other way.

The second was the arrival of money. When large companies started employing kernel developers, the obvious risk was capture — that whoever paid the most would steer the project toward their own interests. It did not happen cleanly in either direction, and the reason is structural rather than moral: no contributor can remove anyone else's rights to the code, so buying influence buys you a seat at the argument, not ownership of the outcome.

The third is ongoing and unresolved. Enormous amounts of modern infrastructure rest on components maintained by very small numbers of people, sometimes one, usually unpaid. That is not a historical curiosity — it is the current state, and it is the most fragile thing about the whole arrangement. I will come back to it in the article about contributors, because it deserves more than a paragraph.

What the history actually teaches

Let me resist the temptation to make this tidier than it was. The story is not a straight line of triumph. There were long stretches where the desktop experience was genuinely bad, where hardware support was a lottery, where the community's internal arguments were vicious and public. Some of that has not gone away. Anyone telling you the last three decades were a smooth ascent is selling something.

What did hold, consistently, was one property: the work was visible and it could not be taken away. Companies that built on it came and went. The code stayed, because the licence made it impossible for any single party to close it. Durability, it turns out, is less about quality than about ownership structure.

And the beginning still matters to me, three decades later, for the reason I started with. Somebody wanted something that did not exist, was told in effect that it could not be done by an individual, and did it anyway — badly at first, publicly, with help. I have never found a better argument against the phrase "it can't be done" than the fact that most of the internet now runs on a student's hobby project.