Five articles, one argument: AI agents don't need smarter models, they need a trust layer — identity, mandates, provenance, liability. Here's where it must go.