This reminds me of the whole house automation proposition, where the problem with it is that even if you remove 100% of the friction in my life from turning on and off light switches, you really haven't removed that much from my life. Replacements don't get very much "budget" before they're more hassle than they are worth and it doesn't take very many technical snafus where I can't actually turn my lights on because of the automation for the whole thing to go net negative.
The actual approval to spend the money is not the hard part. I'm not doing it hundreds of times per day. The research of what to spend it on is the hard part. The agents already do that well enough. Even if you save me 100% of the effort of actually buying a product you're not really saving me that much, and most importantly, it doesn't take much for the whole proposition to go negative because there wasn't much positive in the first place. One product I didn't really want and have to physically return negates in one fell swoop quite a few "oh thank goodness I didn't have to approve this purchase". One misinterpretation that causes the agent to spend 10x what I really wanted to spend negates months of benefits.
This is one of those features where it's something OpenAI wants rather than something most people are going to want. Yes, of course OpenAI wants direct unfettered access to your credit card. Most businesses would love that. But just because they can wave the words "AI Agent!" in your face doesn't mean you should actually give it to them.
We are already at a place where the agent can surely budget better than most people if given the right instruction to do that. I don’t see this tech as being obviously as for the rich.
While I understand many people are wealthy, I can't imagine being so insensitive to price that I would delegate the purchase entirely to an agent.