The approval gate: why it, and not the AI, is the product

Everyone sells AI. Few can say what happens when the AI is wrong. Our answer: nothing that cannot be undone, because the AI does not decide.

For every new ticket Kolli writes a decision brief. Short, one phone screen: what it costs us, what we can charge, the risk, the size of the change, who asked and whether they may, who is hit if it goes wrong, how to roll it back, whether it creates a permanent burden, what we promise to every other customer by saying yes. Then a recommendation. Then a person decides.

The decision is the ticket's state. Not a chat message a program could misread. Recorded, timestamped, attributed, and impossible to skip. Every action that touches a system re-checks that state before acting. An instruction hidden in a customer's email therefore achieves nothing: nobody asks the AI's opinion.

The same holds outward. The reply and the closing summary stay drafts until a person sends them. There is no back door in the system. We tried to get round it, deliberately, as a test. It cannot be done.

That is why we say the gate is the product. The AI becomes usable because something stops it.