Assertion Versus Estimate
The Steward's name is chosen through an act of compression: the developer holds the system's purpose in mind, identifies the function's role within that purpose, and compresses the identification into a label that communicates architectural intent to the next reader. The Steward's name is an assertion.
The model's name is predicted: the next likely token given the surrounding context. The model's name is an estimate.
The codebase built through friction is a codebase in which assertions outnumber estimates, and the human developer can still tell the difference. The codebase built through frictionless generation is a codebase of estimates — statistically plausible syntax that encodes no human purpose because no human purpose was present at the moment of generation.
The Durability Test
Biological Intent determines whether a codebase survives change. When the architecture requires refactoring, when a security vulnerability demands root-cause analysis, when requirements shift in a direction the model has never seen, the codebase with Biological Intent contains the residue of human understanding in its names, its structure, its architectural boundaries. The codebase without it contains only statistical averages. The averages cannot explain themselves. The intent can.
Field Note: "The code runs either way. The difference is whether anyone can explain why it was built this way when the time comes to build it differently."