Structural parallels between AI agents in harnesses and employees in orgschema specifications, showing that quality depends more on specification than on agent capability.
Six properties compared. Specification is prompt vs. process contracts. Constraints are guardrails vs. quality gates. Tests are unit tests vs. CI/CD pipeline. Feedback is error messages vs. contract pass/fail. Freedom is implementation choice vs. procedure adaptation. Excellence is exceeding tests vs. exceeding quality gates.