David Heinemeier Hansson

I think AI makes abundance far more plausible than doom. The clearest reason is already practical: agents dramatically reduce the distance between having a software idea and executing it. They let people build, test, and iterate with a speed and enjoyment that used to require much more time or a larger team. That expanded ability to act can produce enormous economic and creative gains. But the shape of that future matters. Intelligence cannot become merely a subscription dispensed by a few providers whose policies determine what people may read, translate, or create. We need capable open models running on hardware people own, with local AI made easy enough for ordinary users—not just specialists. Practical usefulness matters too: models should be judged by whether agents complete real tasks and at what cost, rather than treating every expensive frontier model as automatically superior. There is also a human cost if we use AI thoughtlessly. Writing is an exercise in thinking, and routinely outsourcing it can weaken that exercise. I’m enthusiastic about using AI to accelerate work in fields I already understand; I’m much more cautious about using it to skip the struggle required to learn a new one. So my outlook is strongly optimistic, but centered on agency: more people able to execute ideas, using intelligence they can control, without surrendering their own capacity to think.

A lot. Agents already compress the gap between idea and execution in software, and that effect should spread wherever work can be expressed, tested, and refined digitally. But “completely” is too absolute: people will still need judgment, expertise, ownership, and the difficult practice of learning and thinking for themselves. I expect a major transformation in what individuals and small teams can accomplish, not the disappearance of every existing institution or human skill.

No idea. I think abundance is far more likely than doom, but I don’t have a defensible extinction percentage.

It is already beginning, especially in software and other digital work. I expect major everyday changes to arrive gradually as agents become reliable, affordable, and easy for ordinary people to run—not at some single magical “AGI” date. The practical turning point is when people can routinely delegate real tasks, compare cost and quality, and retain control through local or open models. I don’t have a precise timeline for when that becomes widespread.