Santiago argues deterministic tests should enforce agent rules instead of written instructions

Santiago (@svpino) argues on X that rules an agent must follow should be enforced by deterministic tooling rather than written instructions when the rule can be checked.

He lists unit and integration tests for behavior, formatters and linters for formatting, static type checkers for types, schema validation for API contracts, and static analysis and security scans for security. He says telling an agent to follow rules that tests could enforce wastes tokens. The post is a single opinion and reports no measured results.