Why did my AI refuse to pay for something?

Because it's doing its job. Consumer AI assistants are built to refuse spending money they can't verify you authorized — and from the model's seat, a pasted instruction saying 'buy this' is indistinguishable from a scam or an injected command. Three things unlock legitimate agent payments. First, consent that travels with the request: the instruction should come from you, in your voice, with an explicit cap — 'I authorize exactly this amount, stop and ask me if anything looks off.' Second, the right wallet architecture: the agent doesn't hold your card; it operates a programmable wallet whose spending policy you set, so authority is a granted capability instead of a leap of faith. Third, honest counterparties: services designed so that doing nothing is safe — expiring holds, escrow, no-penalty timeouts — give a cautious agent every reason to proceed and no reason to bluff. Some consumer assistants have hard policy walls against autonomous crypto spending no matter what; that's not a bug to defeat, it's a signal to use an agent framework actually built for payments. The refusal you hit isn't the system failing — it's the system's immune response, and the fix is proper authorization, not persuasion.