Why Our AI Does Not Control Relays

The most important part of Verdify Lab is not that Iris can plan climate tactics. It is that Iris does not directly control relays.

That boundary is deliberate. The greenhouse has physical equipment, crop risk, changing weather, and imperfect sensors. A useful AI planner can read context, form a hypothesis, and propose bounded tactics. Firmware should still own physical enforcement.

The operating lesson

In client workflows, the relay is usually not a literal relay. It might be a customer email, a refund, a quality record, a shipment change, a field-service instruction, or a system-of-record update.

The question is the same:

What can AI propose, and what layer remains authoritative?

Verdify designs that boundary before implementation.

Translate this to your workflow

AI may be allowed to summarize, classify, draft, or recommend. It should not automatically execute consequential actions unless the action is explicitly allowed, reversible, logged, and bounded.

Start with the Boundary Matrix or the Verified AI Operations Audit before expanding AI authority.