There is simply no general recipe for this. Sometimes I put my little tools and libraries in order and then I'm very productive with them and looking back it seems to have been the key enabler to the actual thing getting done. Other times I go dirty mode and just hardcode constants, copy code files under time pressure and looking back it is clear that getting to the same result with the clean approach would have taken months and the benefit for later tasks would be unclear.

I know some are tired of AI discourse, but I found AI can help to sharpen the tools but at the same time I find that my scope grows such that dealing with the tools takes just as much time but the tools have more features "just in case" and support platforms or use cases that I won't often need, but it feels easy enough to just do, but then as I said it still takes long in total.

It's all mostly an emotional procrastination issue underneath it. Though that can go both ways. Sometimes you procrastinate on thinking about the overall architecture in favor of just making messy edits because thinking about the overall problem is more taxing than just doing a small iteration, sometimes you procrastinate on getting the thing done in favor of working on more tightly scoped neat little tools that are easier to keep in mind than the vague and sprawling overall goals.

What looks like "wasting time on procrastination" may be actually "spending time on thinking". Thinking takes time.

Making messy edits is a bet on previous code quality. If you have paid off enough technical debt, you can take another "technical loan" and expect the rest of the owl to still function despite the mess being introduced. If things are already messy, there's always a risk to make the fess incomprehensible and failing in mysterious seemingly unrelated ways, with the only way to fix it being to git reset --hard to the prior point, and do a more right thing. But the time would have been wasted already.

My approach is usually to timebox it, and cut the losses if an acceptable solution is not yet in sight when the time runs out.

[dead]