Work Doesn't Get Documented Anymore. It Gets Encoded. (But Read the Footnote.)

Erica Brescia's post on the new founder mode is worth five minutes. One line in particular earns its keep.

"Work doesn't get documented. It gets encoded."

That's not a productivity observation. It's a structural one. It names something most AI-at-work commentary misses entirely: the difference between a company where employees use AI tools, and a company where institutional knowledge lives inside the systems themselves. That framing does real work. Other people say "automate your workflows." This says where the knowledge goes when you do.

The Railway numbers are striking. Hold them loosely.

Two FTEs in support, 38 million deployments a month. That's the headline. But Railway is developer infrastructure — structured failure patterns, technical users, enumerable problem types. It's almost the ideal conditions for AI support automation. The moment Brescia moves from "here's what Railway did" to "founders should ask what their company would look like if AI was available everywhere," the argument quietly overreaches.

She does acknowledge it. Once. Buried: "not all founders should be writing code like Jake… it depends on your sector, stage, and where you spike."

That sentence deserved its own paragraph. It's doing the heaviest lifting in the piece and gets a single line before the conclusion rolls on as if it weren't there.

The bifurcation she describes — AI-native operating model versus pre-AI org with AI bolted on — is real and will matter. The encoded company is coming. Just not on the same timeline, or in the same shape, for a law firm, a hospital, or a consultancy as it is for a dev tools startup with a technical founder and a structured support queue.

The insight is worth stealing. The universal application isn't earned.