I can trace a straight line through my hard drive from one abandoned repository to the next, each one a monument to a Tuesday night when an idea felt so obviously correct that starting seemed like the only reasonable response. Almost none of them shipped. All of them felt, at the time, like the one that would.
For a long time I called this curiosity. It’s a flattering word, and not entirely wrong, but it left out the less comfortable half of the story: starting a new project is also a very effective way to avoid finishing an old one, because a new idea has no flaws yet. It hasn’t met reality. It’s still perfect.
Marginalia The exciting part of a project and the valuable part of a project are almost never the same eight weeks.
An old project, by contrast, is full of decisions you now regret, edge cases you didn’t anticipate, and the specific, unglamorous work of telling people it exists. None of that is thrilling. All of it is the actual job.
Slow as a constraint, not a virtue
I don’t think slowness is inherently good — plenty of things deserve to be built fast and thrown away. But I’ve started treating “can I resist starting something new this month” as a more honest measure of discipline than any output metric, because it’s the one number that’s hardest to fake.
Shipping the boring 20% is the only part of the work a new idea can’t compete with.
The fix that’s actually worked for me is unglamorous: a hard rule that any new idea has to survive being written down and revisited a week later before it earns a repository. Almost none survive contact with a week of distance. The ones that do tend to be quieter, less thrilling, and far more likely to still interest me in month four.
Aside Building in public helps here too — a half-finished thing with an audience watching is a much stronger commitment device than a private folder no one will ever ask about.
What finishing actually requires
Not more ideas. I have never once run out of ideas. What I’ve run out of, repeatedly, is the willingness to sit with a project once it stops being new — which is, uncomfortably, exactly the point where most of the value gets created.