// writing

• not even wrong

  2026-07-10

  One worker's AI writes the document, another worker's AI reads it. The company's knowledge becomes a pile of artifacts nobody holds, and it feels like productivity because producing is fast and evaluating is invisible. Nobody has measured the loop where it closes.

• eating its own tail

  2026-07-04

  The models write the web and train on the web. The tails die first, every escape route runs through fresh human data, and the pipeline supplying it is already machine-laced. Leave an LLM an open decision and watch what it picks. The average, every time.

• three kinds of nothing

  2026-07-03

  Every new tool ships with a ledger of gains; the mind-states it displaces never get invoiced. The feed bought the idle gap, and the bill is still arriving. The LLM is running the same purchase on the stuck stretch and the absorbed hours, the state that made the work worth doing.

• no one in particular

  2026-06-08

  AI prose converges on one register because the LLM has read everything and been nowhere. A voice is the opposite, the residue of one particular, forgetful life, and it is the part the average can never reach.

• a tool for thinking

  2026-06-06

  Writing is a tool for thinking, not the thinking itself. It holds what won't fit in your head and saves where you got to. Hand the working-through to an LLM and you get the result without doing the thinking, and without a state you can pick back up.

• the reps you stop taking

  2026-06-01

  What quietly goes when you let a language model do the thinking, why it's the part of you worth protecting, and how to keep it sharp, in mind and body.

• a compression of a compression

  2026-05-31

  Why putting a thought into words flattens it, what that means for language models trained on the flattened version, and the gap that scale doesn't close.

• the brain drives the tool

  2026-05-22

  What AI actually is, what it's genuinely good for, and the part of the work no language model can do for you.