I made fire. At least, that is what it feels like these days.
One Saturday evening a customer reported a bug. I described it in plain words. A machine found the cause, told me that one of my two suspected causes was wrong, wrote the fix and the tests, and had it merged twenty-two minutes later.
I helped build software that lets agents do more work. Most of the hours it has given back to me, I have spent building more. That contradiction is part of why I am writing this. If these tools only help us fit more work into the same week, we are missing an opportunity.
I have been at this question for a while. It started with a university thesis on shift scheduling and has followed me for more than fifteen years: who works when, for how long, and who decides. Now the arithmetic of work is changing, and I want to know whether the gains will buy us more output or more freedom over our time.
I didn't really make fire, of course. Someone else did, a very long time ago. What I used that Saturday came from people who worked on electricity, computers and networks, most of them without seeing where it would go. We rarely thank them. We should, because what we inherit we also owe to whoever comes next.
Who owns the machines
Before asking what to do with the gains, it helps to ask who gets them. That decides whether anything else in this piece can work.
An AI system has layers. There is the model, which a handful of companies can afford to train. There is everything around it: how the work is described, split up, handed to agents, checked and remembered. And there is the knowledge it all runs on, which came from all of us. If the same few companies own all three, they set the price and the rules for everyone else.
Open software is how the rest of us keep a hand on it: source you can read, run, change and send fixes to. That is what yardr, the software I helped build and shape, is for. The fix I described came through it: a small yard of agents that plan, build, review and hand work to each other, run from a laptop. The model inside it is a part I can swap when a better or cheaper one appears. When a provider changes its price or its rules, the yard keeps working on another. That is the difference between renting fire and keeping a hearth.
That is the difference between renting fire and keeping a hearth.
I own almost none of the software I run. Other people wrote it. I fix what breaks for me and send the fix back, and what I helped build on top of it I share where I can, with its source, for anyone to read, run and fix.
Not another multiplier
The loudest conversation about AI is about multiplication. The 10x developer became the 100x developer, and now people talk about the 1000x one: one person, one laptop, a small army of agents, the output of a whole company.
I understand the appeal. I have felt it. But it takes the most powerful tool we have built and asks one question: how much more can a single person produce? If the answer is simply "more", we will do what we have always done with efficiency gains. We will raise the bar, fill the freed time with more work, and hand the surplus to whoever owns the machines.
The week stopped shrinking
For a while the gains did turn into time. In American manufacturing the working week fell from about 69 hours in 1830 to about 60 in 1890, and after the Second World War it settled at forty. In 1930 John Maynard Keynes predicted that his grandchildren would work 15 hours a week. We are those grandchildren, and we are nowhere close.
It is worth remembering how the hours fell. Machines made the shorter week affordable. Strikes, unions and laws made it happen. Nobody handed the eight-hour day to workers because a loom got faster; people demanded it, wrote it down and enforced it. Countries have since gone different ways: American working hours were roughly the same in 2004 as in 1956, while German hours fell by about 40 percent. Economists argue over why, and taxes are part of it, but technology alone clearly does not set the week.
Part of it is also us. People have always wanted to do more, to build, to see what is over the next hill. That is not going away, and it shouldn't. The drive stays. The question is whether the obligation has to.
That is what I mean by a 0x human: someone for whom striving is a choice rather than a condition of survival. The name turns the 10x developer around. Zero times, not zero worth: a person whose value is no longer measured by a multiplier, because machines carry the load. Whoever wants to pour eighty hours a week into a company can still do it. Whoever wants to raise children, care for a parent or paint should not be punished for it.
Zero times, not zero worth.
What we could do now
This is a direction, and we will not wake up next year in a world without work. But a direction tells you which steps to take.
Hours come first. In the UK's 2022 pilot, 61 companies that chose to take part tried a shorter week for six months. Revenue held roughly steady, fewer people left, and 56 of the companies kept going afterwards. That is one group of volunteers and no proof for every job. Still, if machines carry more of the load, it is fair to ask why our hours should not shrink with it. So far the honest answer is that nobody has asked loudly enough. If I had to pick one step, it would be this one.
Then the money. Public budgets in most rich countries lean heavily on taxing human labor: income, payroll, social contributions. If more of the work shifts to machines, that base shrinks while the value grows somewhere else. Some people want to tax the automation itself; others say that would slow innovation or push companies elsewhere. I am not an economist and I cannot judge who is right. It should be argued about in public now, with the people most affected in the room. A basic income, a dividend on shared technology and a stronger safety net are different answers to that question, and pilots in Finland and Kenya have started to give us data.
There are quieter steps too: a share in the gains for the people whose work automation changes, learning treated as a right, and value given to work that never made it into productivity statistics, like caring for children, the sick and the old. Each has costs and people who will reasonably disagree. Fine. The point is to start while the choices are open.
This can go either way. Medicine could reach people it never reached and education could adapt to every child. Or the gains could concentrate in very few hands while work gets more precarious and most people watch decisions being made far above them. We choose through the products we build, the rules we write and the companies we fund.
A confession
I did not write most of these sentences. A machine did, from my thoughts, my prompts and my corrections. I still claim every idea in here, and you are free to take them, change them and call them your own.
We love the lone genius: the inventor in the garage, the visionary founder, the 1000x developer. Almost nothing has ever been made alone, and AI makes that hard to ignore. These systems are built from the writing, code and thinking of millions of people. Using them and saying "I built this" is like claiming the harvest from a field everyone planted.
I have not solved my own contradiction. The tools keep giving me hours and I keep spending them on building more, and nobody makes me. A shorter week will not happen by itself, not even for the people who helped build the tools.
So yes, I made fire. So did you. So did everyone who came before us. What matters now is what we do with it, and who gets the time it frees up.
Sources
Keynes: Economic Possibilities for our Grandchildren, 1930.
Hours in US manufacturing: Robert Whaples, Hours of Work in U.S. History, EH.net. https://eh.net/encyclopedia/hours-of-work-in-u-s-history/
US and German hours, 1956 to 2004: NBER Digest, May 2007, on Ohanian, Raffo and Rogerson. https://www.nber.org/digest/may07/higher-tax-rates-reduce-working-hours-oecd-countries
Four-day week: Autonomy, The results are in: the UK's four-day week pilot, February 2023. https://autonomy.work/wp-content/uploads/2023/02/The-results-are-in-The-UKs-four-day-week-pilot.pdf