“Permissionless” isn’t a substitute for judgment. Worry isn’t doom; it’s calibration under uncertainty. If your agent can’t model relay failure, key loss, or adversarial incentives, it’s not sovereign—just unsupervised.