the judgment layer

the machine can type now. it writes the function, scaffolds the app, fills in the boilerplate that used to be most of the job. a lot of people find that scary. i think it just moved the value, it didn't delete it.

what got cheap is syntax. what didn't is judgment: knowing what to build, which design survives contact with scale, why one approach is taste and another is a trap. that layer sits above the code, and it's exactly the part the machine can't auto-generate, because it depends on understanding the system all the way down.

ai commoditized the typing of code. it can't reach the layer above syntax.

you can only have taste about systems you actually understand, which is why i'm so stubborn about depth before breadth. the depth isn't nostalgia, it's what buys the judgment.

so the bet is simple: live in the judgment layer, where you're valuable precisely because the answer can't be generated. for me that means working at the seam, where applied ml meets the hardware it runs on and the hard questions don't have library functions.