"Useful writing tells people something true and important that they didn't already know, and tells them as unequivocally as possible."
— Paul Graham《How to Write Usefully》(2020)
It could perhaps be argued that, in certain contexts, microservices may sometimes offer some advantages.Microservices help when your teams ship faster apart than together. Otherwise they cost more than they're worth.前者全是缓冲垫，读完不知道你到底主张什么；后者给出一个可被反驳的强主张——这才有用。
"Writing about something, even something you know well, usually shows you that you didn't know it as well as you thought."
— Paul Graham《Putting Ideas into Words》(2022)
I basically understand how OAuth refresh works.Writing the doc, I couldn't explain token rotation — which proved I didn't actually understand it.「大致懂」经不起写作的考验；写不出来，就是没真懂。
适用场景 + 常见错误
✓ 任何你「自以为想清楚了」的复杂问题：架构决策、战略、复盘——先写一版逼自己暴露漏洞
✓ 把写作当思考工具，而非汇报负担：初稿是写给自己想的，不是给别人看的
✗ 误区：迟迟不动笔，等「完全想清楚」——那一刻永不会到来，是写让它到来
✗ 误区：让 AI 替你把草稿写顺——顺的是文字，你却跳过了那场最值钱的思考
本周习作 + 思考题
挑一个你「觉得已经懂」的技术或决策，不查资料，逼自己写 300 字解释给外行听。记下卡住的地方——那就是你其实没懂的地方。思考题：如果 AI 能替你写出流畅的解释，你会失去什么？
"The less energy readers expend on your prose, the more they'll have left for your ideas."
— Paul Graham《Write Simply》(2021)
原理解读
复杂的句子和大词常被当成聪明的标志，PG 说那恰恰是写作者偷懒、或想唬人的信号。他的逻辑是经济学的：读者的注意力是有限预算，每个生僻词、每个绕弯的从句都在征税；税收得越多，留给你真正想传递的 idea 就越少。这和奥威尔「能删就删」同源，而 PG 又在《Writing, Briefly》里把整套写作浓缩成一句近乎全部的建议——删掉一切不必要的。短不是目的，短是把读者的认知预算还给思想的手段。但注意：简单 ≠ 简陋。复杂的思想完全能用简单的句子承载，难的是写作者愿不愿意替读者把复杂先消化掉。
修改示范
We endeavored to ascertain whether the implementation would be efficacious prior to its deployment.We tested it before shipping.大词（endeavored / ascertain / efficacious）每个都在向读者征税；换成普通词，意思一点没少。
It is the contention of the author that this approach is suboptimal.I think this approach is wrong, and here's why.「It is the contention of the author that」是没人会对朋友说的话；换成「I think」，立刻有了人味和担当。
比任何时候都更成立。AI 能替你产出文字，但思考的收益恰恰来自「你自己被句子逼着想通」的那个过程——这部分无法外包。如果让 AI 写完直接交，你得到了文档，却跳过了写作本该带给你的认知升级。PG 的洞见反而成了试金石：在 AI 时代，写作的价值从「产出文字」彻底转向「锻炼思考」，而后者只有亲自写才拿得到。