Meng Shao shares @poteto's four-layer framework for trusting agent work

Meng Shao reposts and translates a framework from @poteto on how to trust output from AI agents.

The framework ranks four layers of trust, with improvements to the codebase at the top and human review at the bottom. It argues that encoding correct behavior in architecture, types, and invariants works better than relying on written rules or manual review.