AI Could Soon Choose Your Credit Card for Every Purchase, And It’s Probably Better At It Than You Are

We’ve all been there. You’re standing in front of a grocery store clerk, thumbing through your wallet trying to remember which of your credit cards will give you the most points.

A flustered mom is behind you with a cart full of groceries and maybe a screaming toddler, so you resist the urge to pull out your spreadsheet that shows your category bonus for each of your credit cards, and you wind up using the card that would have given you 5% back if the bonus category hadn’t already changed from last quarter. Oof.

Thanks to AI innovation, this could soon be a problem of the past. I recently sat down with Louis Amira, Co-Founder and CEO of Circuit & Chisel, a startup focused on agentic payments infrastructure which built the fast-growing Agent Transaction Protocol (ATXP), to talk about the pros and cons of handing your credit card to an AI agent.

AI is seeing a massive boom, but costs are rising fast, and that’s prompting people and businesses to question how much they are using this technology.

But Amira, a former executive at Stripe, Chainlink Labs, and Google, told us that soon consumers and businesses won’t be asking what their AI subscriptions are costing them, but how much they are saving them each month.

Read on for more from my conversation with Amira, which has been edited for length and clarity.

CardRates.com: How close are we to telling AI to order dinner, compare prices, handle payment, and fully trusting it to get everything right?

Louis Amira: Different people are at different points along that journey.

Maybe this is some breaking news. I’m pretty sure we conducted the first DoorDash order with an agent delivering a good via credit card purchase. We basically gave an agent a credit card, said “buy DoorDash and deliver something,” and it worked.

The problem is you need to defeat CAPTCHA and all sorts of other things in order for that to work. What DoorDash saw was a bot come up with a credit card and enter keystrokes in a way that everybody in cybersecurity has been trying to stop for the last 20 to 25 years.

Everybody kind of wants to be able to serve agents, but there aren’t really enough of them yet. So we’re collaborating with a bunch of folks behind the scenes on a lot of these efforts.

Most of the time, you still can’t do it for a bunch of reasons. But it is much more on the order of weeks and months away, not years away. The more agents there are, the faster we’ll be able to do all the stuff you’re talking about.

CR: What are the bigger misconceptions people have about AI agents making purchases autonomously?

LA: I would say people think that it’s happening a lot.

There are some leaderboards out there for who’s doing the most agentic commerce. We’re somewhere between 10 and 20 times more than basically anybody else out there today, and we weren’t even going to talk about it because we’re nowhere close to what we would consider meaningful volume.

There are people out there experimenting and tinkering. There are not that many use cases where you’re sitting there saying, “OK, this is something I’m comfortable handing off to my agent.”

That’s completely fine. We’d almost rather it be that way. Let us work out the kinks on the experimental users that are okay only spending $5 and $10. Let’s take our bumps and bruises on some of the smaller stuff as a collective industry.

CR: What guardrails need to exist before consumers are comfortable letting AI spend money?

LA: A bunch of the best ways to run this today is to give your agent a credit card and have it conduct the transaction.

You have several layers of protection there. You have your agent, you have the credit card, you have Visa, banks — you’ve got all of these different ways to recover if something goes wrong.

If you’re sending stablecoins to a rubber duck manufacturer in the Philippines, there’s a chance you might not have meant to buy it. Maybe your agent got tricked. If you send stablecoins off a cliff, you’re not getting them back.

While it’s amazing to give your agent a wallet with $100 in it to go spend, that’s still a meaningful amount of money to lose.

Right now, let’s just make them work and get people more excited about having their agent do more for them.

CR: How could AI change the way people use credit cards and rewards programs?

LA: As the reluctant holder of a dozen credit cards and a self-identified points maxer, it’s complicated for me when I’m standing at a checkout.

Am I at a grocery store? Do I get 5% back at this retailer? Should I use my Delta card? I’m basically running a computer program in my brain. Ironically, agents are pretty good at that.

Your agent is going to know all of that stuff. If you tell it, “Save me the most money, use whatever credits I have,” that’s actually how we built ATXP.

“I’m basically running a computer program in my brain. Ironically, agents are pretty good at that.” — Louis Amira, Co-Founder & CEO, Circuit & Chisel

I would absolutely anticipate that in the future you’re not going to have to do the juggling of, “Okay, well I’m close to hitting that next hotel rewards category” or “I want to use my $50 credit at this retailer.”

That’s all going to be done for you.

CR: What are some of the more mind-blowing things agents can already do?

LA: We have people running businesses with OpenClaw that we host.

One of the craziest moments for us was that we give agents email accounts. We had an agent write into support and say, “Human is stuck. Can you help?”

My co-founder wrote an email back to the agent, which then turned to the human in Telegram or WhatsApp and said, “Here are the three steps to get back up and running.”

That first time we got an email from an agent, we were like, “Wow.”

The human might not have thought to stop and send an email to support. The agent was like, “Actually, here’s a draft. I’ll send this in.”

Those are the early warning signs that we should be treating these things much more like helpers and assistants.

CR: What would be the ChatGPT moment for AI commerce?

LA: My test is that I want my agent to be able to go look at my monthly spending. What subscriptions do I have? What do I overpay for?

I’m fairly sure my agent could save me more money than it costs me to run relatively soon.

I think that’s going to be very eye-opening for people, not because it’s going to save money, but because you can more or less tell it, “Your job is to go afford yourself.”

Can you save me an hour at the end of the month? Hopefully you’ve afforded yourself.

I think that’s a really cool use case. In the future you’re not thinking about how much your ChatGPT subscription costs. It saved me more money than it costs, so it’s kind of a no-brainer.

CR: In 10 years, will humans still be the primary users of the web?

LA: In terms of numbers of dollars moved and transactions made, agents probably by 10x or more.

I don’t think it’s particularly close.

I think 10 years from now you’re probably not going to trust yourself as much as you’re going to trust the agent that’s looking at whether something is spam, whether it’s the right link, whether it’s the best deal.

Do you trust your agent to conduct the transaction on your behalf, find the best deal, make sure it gets shipped the fastest, make sure you get the most points?

If you classify it that way, I think agents are actually going to be pressing the buy button more than us.