AI And Semantic Edits

AI is most useful when it works against the same typed semantic model as a human editor.

Proposals, not source replacement

An AI agent proposes operations such as:

• add a field or function;
• rename or move a definition;
• decompose one value into several values;
• extract or inline behavior;
• resolve a typed obligation;
• add a test, invariant, or operational objective.

The proposal names stable semantic identities. It does not replace arbitrary source files and hope that textual references still correspond.

The compiler remains the authority

The type system, effect rules, authorization model, verifier, and deployment compiler decide whether a proposed edit can be accepted.

Incomplete proposals remain inspectable edit sessions. Accepted changes become atomic semantic commits with reviewable intent.

Better context

Because the codebase stores dependencies, identities, revision history, and unresolved obligations, an agent can ask precise questions and explain its impact. The same model supports human refactoring, automated maintenance, and AI-assisted development without separate safety rules.