I’m beginning to think this principle is misunderstood by the grinder/hustler crowd. It’s not some ineffable virtue of “building” or “shipping.” 🤔

Many non-building/shipping activities might in fact comprise the same behaviors as building/shipping. I think “shipping” output form factors are maximal context compounding containers, like bank accounts. if you have some money (effort) and some bank accounts (output form factors) with different interest rates (context compounding rates), you’d obviously pick the highest rate one, unless the risk levels are different, in which case you’d pick the highest risk-weighted return rate.

It’s not you “learning,” it’s the thing itself accreting context. And it can do that faster (but at more risk) in production than in testing, and faster in testing than in simulation, and in simulation than in planning analysis.

It’s a subtle difference (or used to be until claude md made it on-the-nose). The thing itself accreting context vs you “learning.” Often looping through production gets a thing working even if I’m none the wiser at the end. Pure trial and error is the clearest case. You try 10 times. The 10th time works and you repeat it and it works reliably but you never entirely learn why, even unconsciously. You didn’t in fact learn. The shippable wrapped itself in context.

I’ve literally never opened up a claude md file, let alone hand-edit. Sometimes I ask claude code how big the file is and tell it to break off some pieces into other md files. But I don’t open it. It’s nice to know it’s human-readable but I don’t read it.