CLAUSE IVTHE THESIS

Most “AI dev tools” today are pipeline thinking[1]: PM → Engineer → QA → Ship. Linear. Brittle. No oversight. No rework. Hope it works.

Oath is governance thinking[2]: hierarchical authority, independent oversight, review gates, rework loops.

The first AI that ships entire products — without a human in the loop — will arrive in the next three to five years. Whoever ships it will bake governance into the model's architecture — not into a prompt that says “you are the PM.”

This page is my claim on that future.

[1] see: AutoGPT, BabyAGI, AgentGPT, MetaGPT (default config).

[2] see: § I–VI above.

CLAUSE IITHE METHOD

$ man oath

  user                                            oath
  ────                                            ────

  1. describe what you want   ───────────────►   receive intent
                                                       │
                                                       ▼
                                                 decompose to spec
                                                       │
                                                 ┌─────┼─────┐
                                                 ▼     ▼     ▼
                                                arch  code  test
                                                 │     │     │       parallel
                                                 └─────┼─────┘       execution
                                                       ▼
                                                 audit gate ────┐
                                                       │        │ reject
                                                       │ pass   ▼
                                                       │     rework
                                                       │        │
                                                       ▼        │
  2. receive ship receipt   ◄─────────────────  deploy ◄────────┘

// rework loops until promise == artifact. there is no "good enough."