I was up until four in the morning this week, and the change I actually shipped took about ten minutes.

The implementation was ten minutes. The understanding underneath it was hours. And instead of banking those hours of understanding somewhere I trusted and going to sleep, I stayed up to spend the last ten minutes right then — which turns out to be the whole bug in miniature.

Let me separate the part that was right from the part that wasn't, because that's the entire point. I work on a distributed team, split across continents; the people depending on this work start their day as mine ends, so the only window where we can move together sits late in my evening. Taking that window isn't a failure of boundaries — it's the job working as designed, and I'd make that call again. To unblock them responsibly I had to actually understand the thing, so I researched: read, compared, traced how it really worked, checked my conclusions. That's real work, and it's most of the work.

The bug is what happened after the window closed and no one was waiting in real time. By then the understanding was done and the change was a few lines — ten minutes of implementation resting on hours of comprehension. And I did those ten minutes at 3am anyway, at the cost of my sleep, when they'd have been ten minutes at any hour the next day. Not because I lack discipline, and not because anyone needed it then. Because the system had no way for me to put the understanding down and trust I could pick it back up. I was holding the whole problem in my head, and I knew that if I stopped, I'd have to re-climb those hours to get back. So I stayed — not because the work demanded it, but because stopping felt like dropping it.

The fear that keeps you up is a system smell

That impulse to keep going is the actual bug, and it took me a while to see it as one rather than as a virtue. I'd framed it as diligence — I'm being thorough, I'm finishing what I started. But strip the flattering language off and it's narrower: I didn't trust the system to hold my place. I didn't believe I could write down where I'd gotten to, walk away, sleep, and resume tomorrow at the exact spot without re-paying the understanding cost. The fear wasn't irrational — given a weak handoff, it was correct.

And that is not a willpower problem. You can't fix it by resolving to have better boundaries, or telling yourself sleep matters — I know it matters, I tell myself constantly, and at 3am the fear of losing the thread wins anyway. It wins because it's pointing at something true: if my capture is bad, then stopping really does cost me the context. The fear is a rational response to an untrustworthy system. The way out isn't to override the fear. It's to make the fear wrong — to build a capture I can actually trust, so that stopping is genuinely free.

What "trustworthy capture" has to earn

The bar is higher than a scribbled note, and I think that's why I don't cross it under pressure. For me to close the laptop at a reasonable hour and believe I've lost nothing, the record has to hold three things, not one: where I got to (the state), what I concluded and why (so I don't re-derive it), and what the very next action is (so tomorrow-me doesn't have to reconstruct the plan before starting). A note that only captures the findings isn't enough — if it doesn't also capture the next step, then picking it up still requires the expensive part, the re-orientation, and my 3am self knows that, which is exactly why it won't stop.

When the capture holds all three, the calculus flips. Stopping no longer means dropping the thread — it means parking it somewhere it'll be exactly as warm tomorrow. The ninety percent I spent understanding becomes an asset that persists, instead of a fire I have to keep tending so it doesn't go out. And the ten-minute implementation gets done tomorrow, in ten minutes, by a version of me that slept.

Fix the system from the one event

The instinct after a night like that is to extract a lesson about me — be more disciplined, stop earlier, respect the sleep. I've learned those lessons many times and they don't hold, because they ask willpower to beat a real incentive. The better move is to treat the single night as a system defect: make the handoff trustworthy enough that walking away is the obviously safe choice, not the risky one. Then stopping stops being a test of discipline and becomes just what the system makes easy.

And there's a layer above the personal one. A distributed team is a deliberate, valuable design — you trade colocation for follow-the-sun coverage and a wider talent pool. But that trade has a seam, and a seam has to be absorbed somewhere. When the mechanism for absorbing it is missing, it defaults to whoever quietly stays late to keep things moving — not because anyone decided that, but because an unnamed cost lands on whoever picks it up first. The fix is the same one, one level up: make the handoff an artifact the system owns, so the seam is absorbed by a process instead of by a person. That's not a grievance; it's the same engineering move that turns any implicit cost into an explicit, designed one.

Because the goal was never to finish at 4am. It's for tomorrow-me to be rested and handed a context warm enough to resume in ten minutes — and for the team to keep moving without that depending on anyone burning the night. A handoff worth trusting gives both, and asks nothing of willpower at 3am, which is the only hour the question ever actually gets asked.