First, Let’s Put the Numbers on the Table
Over the past 31 days, I sampled a dozen or so repositories I have on hand: 274 commits. The most active one, lynxhot, had commits on 21 out of those 31 days. A blog that was only set up on August 3 racked up 100 commits in 8 days. On top of that, there are still more than a dozen research reports sitting on my hard drive—from Tesla’s open-source approach to car building to layered trash bags, from a content analysis of 371 Zhihu answers to the whereabouts of a graduate student who dropped out.
Peak construction hours were between 10 p.m. and 1 a.m. A few nights went straight through to three or four in the morning.
Don’t rush to call it hustle culture. More than 90% of the code was written by AI. My role was closer to that of a contractor overseeing a dozen construction sites at once: drawing up plans, setting rules, inspecting work, ordering rework, and occasionally putting out fires.
The First Half of the Month: AI Made Starting Things Addictive
At 2:47 a.m. on July 9, the first machine-pushed article landed in my WeChat Official Account draft box. That night, I worked from midnight to four in the morning and got the entire flow running for the first time: “write article → render cover card → push to draft box.” Before daybreak, it had solidified into a reusable process. By 9:30 that same morning, the AI morning-and-evening curated four-piece set had become routine—the pipeline went on duty the very day it was born and has not missed a day since.
That feeling is addictive. In the past, starting a project meant weighing it for half a month. Now it takes one night. So over the course of a month, I opened more and more pits: a blog, a web editor, a TG bot, a real estate platform, furniture placement previews, a paid salon… Each one could run. Each one looked like it had a shot.
And that was exactly the problem.
On August 1, I Set a Rule for Myself
There were too many construction sites, and only one contractor. Each project ate up a bit of attention every day, diluting real progress until it became invisible. Add to that a more practical constraint: cash flow.
That day, I wrote myself a rule called “project concentration discipline”: No new projects. If a new idea wants to pass the gate, it has to be one in, one out—if I want to start something new, I have to bury something old first.
Holding a Funeral for a Project Is Much Harder Than Breaking Ground
Burying a project is more counterhuman than I expected.
LynxMarven, with 93 unit tests and 8 end-to-end tests, all green, clean and elegant code—frozen. Lynxhouse, with 5,152 rows of data across 23 cities, online for barely a week—data archived, local version removed, recovery path written into documentation. There was also a paid salon, with all the private-repo ammunition ready to go—sealed and waiting.
A funeral has to be done properly: write down why it was frozen, how to unfreeze it, and which assets can be reused later. A well-frozen project is hibernation. A poorly frozen one is an unfinished mess. The difference is entirely in the documentation.
Research reports are the same. For the layered trash bag idea, I spent a night researching it and ended with two words: don’t do it. The value of those two words was this—they stopped me from sinking another month into a dead end.
The Last Night, 70 Minutes
On the evening of August 9, I pulled together a blog content pipeline that had been growing in scattered pieces for a month: six repositories, four outputs, five parallel sessions under construction at the same time, merged into one LynxPipe in 70 minutes. There was even an incident along the way—the code from one parallel session, not yet committed in time, was completely overwritten by a migration operation from another session. In the end, I rewrote it according to the requirements and, while I was at it, carved the lesson into the memory bank.
A month of scattered growth, converging on the final night. I think that order was right: first let each piece get beaten up and stabilized in real use, then talk about merging. If you start by designing the perfect architecture, more often than not you’re laying foundations for imaginary requirements.
What I Really Want to Say
AI has driven the cost of “starting” close to zero. A demo in one day, a prototype in three—two years ago that was mythology. Now it’s my routine.
But the cost is buried deep: when starting is no longer a decision, finishing becomes the decision. Every day, ten new pits are tempting you. But your attention, your cash flow, and those three or four late-night hours have not increased at all.
So what I practiced over and over this month was not starting, but ending: freezing, sealing, archiving, merging, writing handover documents. It doesn’t sound as sexy as “I launched another new product,” but after reviewing the month, I actually have fewer loose ends in my hands—and every remaining thread is one that can truly fight.
AI eliminates the barrier to starting, not the weight of choosing. The latter now rests entirely on humans.
That is probably the most important commit of this month.