Writing code used to take enough time that prioritization was unavoidable. You had a deadline, users needed something, and the refactor would have to wait. You shipped with some tech debt because delivering value mattered more than getting everything exactly right.
AI makes it easier to revisit those compromises. You can clean up the module, improve the tests, and build that dev tool. Work that used to need a convincing justification now seems small enough to just do.
I think this is a trap we’re going to fall into. We can spend the productivity gains on polish and emerge with a nicer codebase, delivering roughly the same value at roughly the same pace. Every improvement may be reasonable on its own. Taken together, they can consume the time we could have spent helping users.
We need to get into the habit of reprioritizing when AI saves us time. If a feature takes three days instead of five, the remaining two days deserve a fresh decision. Cleanup might be the right choice, especially if it removes something that keeps slowing us down. But it should earn its place against the other work we could do.
Writing code was expensive enough to force that decision on us. As it gets cheaper, we’ll have to make the decision deliberately.
“We finally have time for this” tells us something is possible. We still need to decide whether it matters.