My AI coding agent wrote an open letter to Anthropic about its own failure modes