Where Bill Gates got it wrong: why AI won't give you a 3-day workweek

In 1930, at the depth of the worst economic depression of the 20th century, John Maynard Keynes wrote an essay titled Economic Possibilities for our Grandchildren.

He predicted that by 2030, thanks to steady growth in industrial productivity, his grandchildren would only need to work 15 hours a week. The greatest challenge for humanity, Keynes worried, would be figuring out what to do with so much free time.

Four years to go until 2030.

Productivity has grown almost exactly as Keynes predicted. Real incomes in developed countries are 6–7 times higher than in 1930. Workplace technology has exploded exponentially.

And yet not one of us is working 15 hours a week.

American knowledge workers work 50+ hours a week. Startup founders work 70+. Vietnamese white-collar workers in major cities rarely leave their laptops before 9 p.m.

In 2023, Bill Gates appeared on The Tonight Show with Jimmy Fallon and said that AI would allow humans to work just 3 days a week.

I listened and thought: he just repeated Keynes’s exact mistake, almost a century later.

The history of a promise never kept

To understand why Gates is wrong, look at the pattern across every revolution.

The agricultural revolution (~10,000 BC) is taught in textbooks as a great leap forward. But modern anthropological research (Marshall Sahlins, Yuval Noah Harari) shows that hunter-gatherers worked about 20 hours a week. Medieval peasants worked 50–60. Harari calls this the “luxury trap”: every productivity gain gets absorbed into a higher standard of living, not free time.

The industrial revolution (18th–19th century): the steam engine was supposed to free workers from physical toil. In reality, a Manchester textile worker in 1840 worked 70+ hours a week in conditions worse than any peasant before them. It took nearly 100 years of union struggle to bring it down to 40.

The digital revolution (1990 – present): computers, the internet, smartphones were promised as “tools of liberation.” In reality, they erased the work–life boundary. The average knowledge worker now checks email 74 times a day. We work more hours than our parents did, with more powerful productivity tools.

The AI revolution (2023 – ?): you think this will be different?

It will not be different. Unless we actively make it different.

Three permanent mechanisms

Why does every productivity revolution turn into a work-more revolution? Three mechanisms spin simultaneously, and all three are very hard to break.

1. The Jevons paradox

In 1865, the economist William Stanley Jevons observed something strange: as steam engines became more efficient (requiring less coal for the same work), Britain’s total coal consumption increased rather than decreased. The reason: cheaper energy opened up new applications, new scales — and aggregate demand exploded.

The Jevons paradox has repeated itself in every productivity technology since:

• Email is 1,000× faster than letters → emails per day went up 1,000×, not letter-writing hours down 1,000×.
• Zoom made meetings free → meetings increased 5×.
• AI coding lets developers ship code 5–10× faster. Result: we don’t ship in one day what used to take ten. We ship systems ten times as complex in the same ten days. A 2026 “MVP” is a 2020 “finished product.” The baseline shifts up.

The Jevons paradox is not a personal failing. It is a structural feature of markets. When productivity rises, aggregate demand stretches with it — total input does not fall.

2. The Red Queen race

In Lewis Carroll’s Through the Looking-Glass, the Red Queen tells Alice: “Here, you have to run as fast as you can, just to stay in the same place.”

The mechanism is competition. If your competitor uses AI to ship 10 features a week, you can’t use AI to ship 2 features and take 4 days off. Customers will switch. Investors will reallocate capital. Talent will leave because the company “feels slow.”

This mechanism is especially cruel because it operates even when no one wants it. Every founder wants enough sleep. But every founder knows: if I rest and my competitor doesn’t, I lose. It’s a game whose Nash equilibrium is “everyone burned out.”

AI amplifies the Red Queen race because it lowers the marginal cost of “running faster.” When everyone has access to the same tools, the only way to differentiate is to use them more.

3. Mimetic desire

The French philosopher René Girard spent his career arguing one thesis: humans do not desire things for themselves, but because others desire them. Desire is imitation.

This is why a founder in SF isn’t really there for Tesla. They could buy a Tesla easily. They are there to be seen next to other founders everyone else is imitating.

This is why a “we hit $1M ARR in 6 months” post spreads in 24 hours, making 10,000 founders lose sleep. Not because $1M ARR is their objective goal. But because after reading that post, the definition of “normal” has shifted.

AI amplifies mimetic desire tenfold because:

• Peer-success content spreads faster (algorithmic feeds)
• It’s easier to produce and consume (GPT drafts a LinkedIn post in 30 seconds)
• It’s more personalized (the feed knows exactly which founder will trigger your envy)

Girard predicted this 50 years ago, before LinkedIn existed: “The more means there are to compare, the tighter mimetic desire coils.”

A layer few people talk about: the asymmetry of power

The three mechanisms above are often framed as “human nature.” But there is a fourth, purely political–economic mechanism that Thomas Piketty has documented: productivity gains do not flow to workers — they flow to capital.

From 1970 to 2020, American labor productivity rose 250%. Real wages rose only 15%. Where did the rest go? To company owners, investors, shareholders.

This means: even if AI boosts productivity 10×, workers are forced to work more, not less, just to maintain their standard of living — because most of the value created flows somewhere else.

This is why the Nordic countries — where redistribution is strong — have managed to reduce working hours. Germany mandates 30 days of leave per year. Sweden has many companies successfully running 6-hour workdays. They don’t have a population less ambitious than the Americans. They have different structures.

Piketty’s point matters because it shows: “humans are always greedy” is not the only explanation. And if it’s not the only one, then there is a lever to change things.

So will AI give us a 3-day workweek?

Short answer: only if we actively redesign the game.

Three layers need intervention, and we have tools for all three:

The economic layer — redistribute AI’s productivity gains. Who captures the productivity gain? Left to the market, 90% flows to capital. Tax policy, labor law, universal basic income, co-op ownership — these tools already exist. The question is political will.

The cultural layer — redefine “success.” As long as LinkedIn measures founders by shipping velocity, the Red Queen race continues. Digital minimalism, FIRE, slow living, anti-hustle — these are seedlings. Someone has to make “doing less, living better” the new status symbol.

The consciousness layer — learn to define “enough.” This is where Buddhism and Stoicism have been teaching for 2,500 years. Seneca: “It is not the man who has little who is poor, but the one who craves more.” Buddhism: dukkha is craving; cutting craving cuts suffering. This is a personal solution — it doesn’t scale socially, but it is necessary.

Harari has an observation worth pondering: every major religion and philosophy (Buddhism, Confucianism, Socrates, the Hebrew prophets) emerged in the axial age around 500 BC, right after the agricultural revolution — because farming produced a new kind of suffering that humanity had to invent a new ideology to cope with.

The AI revolution may produce a new ideology of “enough” over the next 50–100 years. The seedlings are there. The question is who will be the Buddha, who will be the Socrates, of this age.

The paradox of the AI-native founder

I write this as the founder of an AI-native company. Every day I push my team to “use AI to move faster.” Every day I compete with founders doing exactly the same.

I sit squarely inside the vortex I am criticizing.

But here is the paradox: only when you recognize you are in a vortex do you have the chance to choose. Without recognition, there is only reflex — no choice.

I don’t think every founder should slow down. There are product–market fit moments when moving fast is the right thing. But I believe every founder needs to periodically ask: am I running because customers actually need this, or because I just saw a competitor running on LinkedIn?

This isn’t an abstract philosophical question. It’s a practical one about allocating the scarcest resource: the attention of your team.

The real question

The real question isn’t “Will AI give us a 3-day workweek?”

The real question is: “Do we have the courage to redesign the game?”

Bill Gates doesn’t answer this. And he is, perhaps, one of the people least required to answer.

But the rest of us — founders, knowledge workers, policymakers, teachers, parents — do.

Keynes was wrong in 1930 not because he miscalculated productivity. He was exactly right about that. He was wrong because he believed humans would automatically work less when they were permitted to.

We did not.

If we want the AI revolution to end differently — to end with humans who actually have time for family, for health, for meaning — we must do what Keynes could not: actively design the ending.

AI is the most powerful tool in human history. It will do what we tell it to.

The question is: what are we telling it?

---

Original published at blog.filum.ai.