A lot of my best work seems to happen in the hour or two after I’m supposed to have stopped.
The feeling got strong enough that I went and counted. Over the past year, by commits, the busiest hour of my workday was 5 to 6 p.m., with about one in every eight weekday commits. Roughly another sixth came after six.
That number doesn’t prove much. A commit marks when something got finished, not how long I spent thinking about it, and some of those commits were written by an agent anyway. But it lined up with the feeling, so I stopped writing the feeling off.
So, one more hour? I usually say yes.
What’s good about the evening
The obvious answer is that it’s quiet. Fewer messages, no meetings, nobody waiting on me.
But I think the bigger thing is what the quiet lets me do, which is work on something alone with AI.
During the day, most of what I do goes through other people. At a big company, even a small change means explaining the background, figuring out who needs to sign off, waiting for replies, and explaining it again to whoever missed the first round. Nobody’s doing anything wrong. That’s just what it costs to get a lot of people on the same page. It’s still slow, and a big chunk of my energy goes into that part.
In the evening it’s just me, the problem, and an AI agent. I say what I want, see something in a few minutes, fix what’s off, and try again. Bad ideas die fast, before I get attached to them. Good ones I can build on right away, no meeting required.
For the kind of work I do, that’s just faster. And the feedback is really strong. Every small thing that works makes me want to start the next one, and an hour like that sometimes gets further than a whole day of waiting.
Talking to people still matters. Whether something should be built at all is a conversation, and an agent can’t decide that for me. But with people, I wait until everyone’s free. With an agent, I can try an idea the moment I have it.
I might be giving the evening too much credit
I’m a little suspicious of myself here, though.
When a problem finally clicks at seven, it’s easy to forget the afternoon I spent ruling out the wrong approach. That felt like going in circles, and the evening felt like progress. But the evening probably moved fast because the afternoon had already done the groundwork. Everything was loaded in my head and the next step was obvious.
Also, that extra hour is often the only stretch of my day nobody interrupts. I’m comparing it to a daytime that’s been chopped into little pieces. Of course it looks great.
So maybe there’s nothing special about the evening. Maybe the right conditions just tend to show up after work.
Wanting to keep going, or not being able to stop
All that said, I really do like that hour. Sticking with a problem because I want to see how it turns out is one of my favorite feelings, and I don’t feel like making excuses for it.
What gives me pause is that two very different thoughts can keep me at the desk. One is “I want to see what happens next.” The other is “today doesn’t count until I have something to show for it.” From the outside they look the same. They feel very different to live through, and a lot of the time I can only tell which one it was afterward.
Calling it “high ROI” doesn’t help me tell them apart. The return is easy to see: something that was broken now works. What else that hour could have been is impossible to count, because it never happened.
I like the idea that doing a little extra every day adds up. I just want it to stay something I choose, and never turn into a rule that every free hour has to be filled.
Could I move it to daytime?
If the evening works because the loop is short and nobody interrupts me, the time on the clock shouldn’t matter much. I should be able to move some of it earlier.
Something like a block during the day that’s just me, the problem, and the agent. I’d need to know where to start before I sit down, and ask people my questions the day before, so I’m not waiting on anyone once it starts.
The smallest thing I can do tonight: before I stop, leave myself a note for tomorrow. Where I am, what I was about to try. Then I don’t have to finish everything tonight just to avoid losing my train of thought.
Some nights I’ll still want to keep going, and that’s fine. I mostly want to find out whether, with time like this during the day, I’d still find the evening so hard to let go of.
Monday morning, then. Notifications off, last night’s note waiting for me.
我一天里干得最顺的时候，常常是本来该收工以后的那一两个小时。
这个感觉越来越明显，我就去数了一下。过去一年，按 commit 算，工作日里最忙的一个小时，就是下午五点到六点，差不多每八个 commit 就有一个在这儿。六点以后，还有大约六分之一。
这个数字说明不了太多。commit 只记事情做完的那一刻，前面想了多久，它不知道。而且有些 commit 本来就是 agent 写的。不过它跟我的感觉对得上，我也就不觉得那是错觉了。
所以，再干一小时？我一般都会说好。
晚上到底好在哪
最先想到的当然是安静。消息少了，没有会，也没人等着我回话。
但我觉得更关键的，是安静下来以后，我终于能一个人跟 AI 把事情做下去。
白天我手上的事，大部分都要经过别人。在大公司里，哪怕一个小改动，也得先把来龙去脉讲清楚，搞明白要谁点头，等回复，再给错过上一轮的人重新讲一遍。谁也没做错什么，人一多，协调本来就是这个成本。可它真的慢，我有好大一块力气，都花在让大家对齐上了。
晚上就简单了，就剩我、问题，和一个 AI agent。我说想要什么，几分钟就能看到结果，哪儿不对马上改，再试一次。不行的想法很快就知道不行，还没来得及舍不得。行的想法当场就能接着往上搭，不用先约个会讨论。
对我做的这些事来说，这样干就是更快。而且正反馈特别强：每跑通一个小东西，就更想接着做下一个。这样的一个小时，有时候比我白天等一整天走得还远。
跟人聊当然还是重要的。有些事该不该做，得跟人商量，这个 agent 替不了我。只是跟人沟通，得等大家都有空；跟 agent，我想到一个点子，马上就能试。
功劳可能记错了账
不过我也有点怀疑自己。
一个问题晚上七点终于想通了，我很容易忘掉下午花了多久，才排除掉一条走不通的路。那时候感觉像在原地打转，晚上才觉得有进展。可晚上走得快，很可能就是因为下午已经把路铺好了：上下文都在脑子里，下一步也很清楚。
还有一点。多出来的那一小时，常常是我一天里唯一一段没被打断的时间。拿它跟被切得稀碎的白天比，它当然显得厉害。
所以也许晚上没什么特别的，只是那几个条件，碰巧总在下班以后才凑齐。
想继续，还是停不下来
话说回来，我是真喜欢那一小时。因为想知道结果，就跟一个问题一直耗下去，这种感觉很好，我也不想为它找什么借口。
只是我发现，让我坐在桌子前不走的，可能是两种完全不同的念头。一种是“我想看看接下来会怎样”，另一种是“今天还没做出点东西，不能算数”。从外面看一模一样，自己过起来差别挺大的。而且我常常要过后才分得清，当时是哪一种。
把这段时间叫作“高 ROI”，也帮不了我分清。回报看得见：一个跑不通的东西，现在跑通了。可这一小时本来还能拿来干什么，那些没发生的事，是算不出来的。
我挺喜欢“每天多做一点点，慢慢会攒起来”这个想法。只希望它一直是我自己愿意的，别慢慢变成一条规矩：有空就得填上。
能不能挪到白天
如果晚上好用，靠的是回路短、没人打断，那跟几点钟关系应该不大，我应该能挪一点到白天。
比如白天专门留一段，只有我、问题和 agent。坐下之前就知道从哪儿开始，要问别人的事，前一天就问好，开始以后就不用等谁。
今晚就能做的最小一步：停下来之前，给明天的自己留几句话，写清楚做到哪儿了、下一步打算试什么。这样就不用为了怕思路断，非得今晚一口气做完。
有些晚上我肯定还是想接着干，那也挺好。我就是想知道，要是白天也有这样的时间，晚上还会不会这么舍不得停。
周一上午试试看吧。通知关了，前一晚的便条在等着。
Comments留言