Capability security for AI agents — authority drawn at the door, not the process or container. Every claim graded against the running code. https://robertdelanghe.dev/blog 2026-06-22T00:00:00.000Z Robert DeLanghe https://robertdelanghe.dev/blog/the-hard-part-of-agentic-development 2026-06-22T00:00:00.000Z

Writing code was never the bottleneck. Keeping a hundred agent-authored pieces honest against each other is — and that's a contract-enforcement problem, not a code-generation one.