The agent proposed. The store said no. Then a person decided.

On October 1 we gave an AI agent a spending limit of $5,000 and sent it shopping in a demonstration store. It found a sofa for $1,899 and sent the cart to the shopper, who paid. Then it tried a sectional for $4,980. With $1,899 already spent, the store refused the cart before the shopper ever saw it.

The technology did its job, and that's the part people notice. The part worth noticing is what happened next: nothing moved until a person raised the limit. Once the shopper set it to $7,500, the same cart went to her, and she paid. The agent found, compared and proposed. It never decided anything that bound anyone.

Every organization that lets agents buy from it, or buy for it, is about to meet that moment over and over. A cart over the limit. An agent nobody recognizes. A purchase someone wants reversed. The wall can refuse on its own. What it can't do is decide who is allowed to say yes. Somebody has to own the limit.

In most companies today, nobody does. It falls somewhere between sales, risk, IT and legal, which in practice means it falls to whoever is on call that day. That isn't a technology gap. It's a decision-rights gap: the Organization question, showing up in a Technology costume.

So the order matters. Decide who owns the yes, write it down, and then let the wall enforce it. Build the wall first and you get a very precise mechanism enforcing a decision no one actually made.