Addicted to Problems That Matter

AI gives us an unprecedented ability to solve problems. The obvious promise is that this should give us time back. So far, it feels like the opposite.

In one controlled experiment, people using ChatGPT finished writing tasks 40% faster and produced work rated 18% better. You would expect gains like that to create slack.

But economists have a name for a similar pattern: the rebound effect. When a resource becomes cheaper to use, we often consume more of it. Efficiency does not always reduce demand. Sometimes it expands it.

AI applies something like this to thought. When the cost of testing an idea collapses, more ideas become worth testing. Every half-formed problem becomes promptable, buildable, fixable. The work does not disappear. The queue gets longer.

There is a psychological pull too. Flow research points to clear goals, immediate feedback, and a balance between challenge and skill. AI-assisted building produces that loop almost perfectly: prompt, output, test, correct. The problem stays just close enough to feel solvable.

That does not mean the feeling is always accurate. A randomized study of experienced open-source developers found that AI tools made them 19% slower, even though they expected to be faster. Movement can feel like progress before the evidence agrees.

That is the strange pull of AI. It does not empty the queue. It expands the frontier of what feels possible, and every solved problem exposes three more.

The discipline is no longer just starting. It is choosing which problems deserve your attention—and knowing when to stop.

AI reduces the friction of doing. Stillness now has to be defended.