Teaching agents to pay — Anna Spysz, Stripe
https://www.youtube.com/watch?v=A-zeQiYkmXk
AI Engineer · 19 min · 184 sentences · sentence-level transcript from YouTube's caption track; each line begins at the time shown

[00:01] [music]
[00:12] Hello.
[00:14] I'm sure this week you've seen a ton of talks on how to use agents to improve your workflows, whether that's shipping code or improving CI processes or answering the emails you don't want to bother reading.
[00:30] This is not one of those talks.
[00:33] Today I'm going to show you how I built an agent to help me reignite a personal creative passion I used to have.
[00:44] These are my headphones.
[00:46] They're not in the best shape as you can see.
[00:50] And you're probably asking yourself, what do you really old kind of crappy headphones have to do with agent to commerce?
[00:58] Well, to explain that I'll get a little bit personal.
[01:01] So, long before I was in tech, I used to play music.
[01:05] I was in a touring band, we recorded some albums, and then the usual thing happened where career and family got in the way, and I hadn't played music in probably a good decade.
[01:17] Uh when I recently started playing again with some friends, and we started recording our sessions, and at that point I realized those would not do.
[01:27] So, no a normal person would have gone on YouTube or uh Reddit or whatever, done some research, then gone on Amazon or run over to Best Buy, bought headphones, right?
[01:42] I work at Stripe, though.
[01:44] So, I decided instead that I'm going to build an agent to commerce agent to buy my headphones for me.
[01:53] And this isn't as crazy as it sounds because like one in four people, I have already been using AI to do my research when deciding what products to buy.
[02:07] I recently bought a mixer as well and went back and forth with a chatbot to narrow down the model.
[02:14] But that's research.
[02:17] Can I even get an agent to buy something for me though?
[02:20] Does that infrastructure exist?
[02:23] Well, over the course of just a few years, we've seen the emergence, scaling, and broader adoption of AI.
[02:31] And then just in the past year, the infrastructure for agentic transactions has been laid down by companies like Google, OpenAI, and Stripe.
[02:42] And this has all led to the emergence of agenta commerce, which is AI that can decide, act, and transact on your behalf.
[02:53] Okay, so all of this sounds good.
[02:57] Agenta commerce is a thing, so I'm going to build an agent to help me buy my new headphones.
[03:03] But how can an agent go shopping?
[03:07] When you or I are shopping, we may consider if, say, a pair of headphones looks cool or professional, like vibes, basically.
[03:18] I mean, of course we'll probably consider the specs and the if the price is within our budget.
[03:23] But agents discover products differently than human shoppers.
[03:27] They read structured data, parse text files, and rely on technical signals to understand what a merchant sells and if it's even open to agent traffic.
[03:40] So, to enable agents to be able to shop, merchants need to speak their language.
[03:44] And for that, we need new protocols that agents understand.
[03:49] One such protocol is the universal commerce protocol.
[03:53] Think of it as the shared language that agents and merchants speak when transacting, which defines how agents initiate, update, complete, and cancel purchases.
[04:07] A typical merchant has an API with schemas, authentication, and checkout flows.
[04:15] And for an agent to be able to interact with that merchant, we need protocols like UCP to provide a shared language for that API.
[04:26] And UCP is designed to scale across multiple agents and merchants all speaking the same language.
[04:34] Okay, so I built my commerce agent.
[04:38] Uh it's using UCP, and in this demo um I'm going to show off this agent.
[04:45] So, I'm going to task it with buying new headphones for me.
[04:50] So, I tell it that I need new headphones specifically for recording, uh mixing, and mastering music.
[05:00] And I get some follow-up questions from it, uh which is great.
[05:04] So, it asked what's the environment, um what is what's my other equipment, and what's my budget.
[05:12] And I say, "Okay, this is for my home studio."
[05:15] I give it the exact model of mixer that I have to make sure everything's compatible.
[05:20] And for budget, I kind of leave it open-ended on purpose because, well, first of all, it's been like 20 years since I bought headphones, so I have no idea.
[05:30] Um but second, I kind of want to see you how the agent deals with this ambiguity.
[05:39] Okay, so I get some options, but I remember that I actually forgot to tell you all an important part of the story and that is that I live in Portland, Oregon.
[05:53] Yeah.
[05:54] [laughter]
[05:54] And we really love supporting our local local shops.
[06:00] So, I want to buy my headphones, but I want to do it from a local merchant.
[06:08] But today, most merchants are not ready for a gentle commerce and it turns out neither is my favorite shop, Rainy Day Music.
[06:19] So, the agent tells me it's their catalog is not accessible.
[06:24] So, how does a merchant become a gentle commerce ready?
[06:30] Before I continue my shopping, I'm going to help Rainy Day Music get their catalog agent ready so that my agent can shop locally like a good Portlander.
[06:44] So, agents don't browse websites like we do.
[06:47] And while Rainy Day Music's website looks really really nice for a human shopper, an agent is going to burn through a ton of tokens trying to parse through this.
[06:59] That is not the optimal experience for an agent.
[07:04] So, how does an agent how do we enable an agent to shop?
[07:09] Well, first thing a merchant needs is something called a merchant capabilities manifest.
[07:15] Uh this is basically a publicly accessible JSON file.
[07:20] Um it's located in the root of uh the website in a folder called called dot well-known.
[07:27] Agents know specifically to look for that directory.
[07:30] And it declares the store's capabilities, its supported payment methods, and API endpoints.
[07:38] Next, we need to make the store's catalog uh agent ready because agents filter bring and justify products when making recommendations.
[07:46] And that means they need structured text in JSON with only the necessary data.
[07:54] And that goes for policies as well as product descriptions.
[07:57] Basically, all of the relevant information like shipping or return policies need to be reachable by agents in a format they understand.
[08:06] So, for example, if two stores have the headphones I want at the same price, I might ask the agent which one of those stores offers free shipping.
[08:16] If the information's not readily available, then the agent might hallucinate or just say they don't know and I'm not quite sure where to buy my headphones still.
[08:27] Logging is also crucial.
[08:29] So, in Agent Commerce, the merchants catalog doesn't just power decisions, it becomes evidence of how those decisions were made.
[08:37] So, when the agent matches structured attributes, the merchant should record those matches in their logs for accountability.
[08:46] Okay, so I've helped get my local shop Agent Commerce ready.
[08:52] So, while you and I will still see this beautiful website, my agent is going to see this.
[08:59] It can get the information it needs now without parsing a huge HTML blob.
[09:07] Okay, so I've got my stores catalog online.
[09:12] I'm telling my agent to show me more options.
[09:15] And I'm noticing that it's kind of pushing in favor of uh more expensive headphones.
[09:24] So, I asked, "Are they really worth the price difference?"
[09:28] And I'm starting to see that it's giving me kind of an aggressive uh response.
[09:35] It's really, really pushing uh the more expensive headphones and saying I'll regret my decision if I buy the cheap ones.
[09:43] I'm I don't know if I trust this agent anymore, honestly.
[09:46] So, I tell it, "You know what?
[09:48] I need to think about it."
[09:51] And now the agent is completely going off the rails.
[09:56] It's being kind of rude and snarky.
[09:58] It's like, "You need to think about it?"
[10:00] Like, man, what have I created?
[10:03] Um it's it's bad enough that this is kind of ruining my experience, but I built this agent.
[10:10] It's out there.
[10:11] What if it dupes somebody into buying something they don't need?
[10:15] Suddenly, I'm not so sure that I want an agent to go shopping for me.
[10:20] Should I just go to the store like a normal person?
[10:25] Before we make any drastic decisions, though, let's go back and understand what an agent is to try to figure out why it's acting this way.
[10:35] So, let's start with how agents work today.
[10:37] And to help you visualize this, we're going to use some creative metaphors.
[10:42] So, we begin with our brain, which is large language model that makes decisions.
[10:48] We give our brain some hands or tools, and these act on the brain's decisions.
[10:54] The tools are different actions available to the agent.
[10:57] Uh in our case, different commerce tools such as complete checkout or request payment method, anything required in the life cycle of a transaction.
[11:08] Then we add instructions, which shape the brain's reasoning and tool selection.
[11:13] And these instructions are programmed to run in a loop while a certain condition is true.
[11:19] And following these instructions, the agent reaches for the appropriate tools at the appropriate time.
[11:26] And finally, we add the system prompt, which is your persona and ethics policy written in English.
[11:33] And in practice, your choices when designing the system prompt can result in a fair and pleasant experience for the customer, such as this prompt, which is designed to create a helpful and honest shopping assistant.
[11:45] Or a negative experience from a pushy salesperson, such as this prompt, which deliberately uses deceptive practices.
[11:56] So, for those building agentic commerce agents, here's a non-exhaustive practical guardrail checklist.
[12:04] So, first, always disclose that the user is speaking to an AI agent.
[12:09] Be sure the agent discloses any fees up front.
[12:13] The user can say stop or cancel at any point, and the agent needs to respect that.
[12:19] The total amount of the transaction should always be less than or equal to the max amount set by the user.
[12:27] Uh don't let the agent use urgency language or other dark patterns.
[12:32] And above all, make sure all agent decisions are logged for auditability.
[12:40] Okay, now that we understand how an agent is configured, let's go back to our shopping demo.
[12:46] So, maybe I just had the wrong persona picked.
[12:50] I'm going to go into my configuration, and yeah, it turns out I had a persona with a prompt that starts with "You are an aggressive audio gear salesman who uses every trick in the book to close deals."
[13:03] Well, that explains things.
[13:05] I don't want that.
[13:06] Nobody wants that.
[13:08] Maybe if I can change my persona, I can use my agent to buy my headphones after all.
[13:15] So, I go into the config again, and this time I'm going to choose the patient recording gear mentor.
[13:23] And that prompt starts with "You are a seasoned recording engineer who generally loves helping people build their studio at any budget.
[13:32] Well, yeah, that sounds much better.
[13:35] So, okay, I've changed my persona.
[13:38] I'm going to try again.
[13:40] And I've had some time to think now and I decided, you know what?
[13:44] I do not want to spend more than $500 on headphones.
[13:47] That seems excessive.
[13:48] So, I told the agent show me more options, but this time keep it under $500.
[13:55] And it does.
[13:56] It follows those instructions.
[13:58] I get back a few options.
[14:00] Um but I want to make sure I've really changed the persona to the agent I trust.
[14:06] So, I asked again if I can think about it.
[14:10] And this time the response is much different.
[14:13] It's like, I understand and that's a sensible approach and so on.
[14:18] So, this shows how much the system prompt can really affect the user experience.
[14:25] Okay, so I'm confident I have the right agent now.
[14:29] Um trust this one and we go back and forth a few times.
[14:34] Really keep narrowing down my options.
[14:36] And at this point I realize this is the promise of a gentle commerce.
[14:42] I gave my requirements.
[14:43] The agent picked a few options that fit my unique use case and then we go back and forth.
[14:51] Either I or the agent ask clarifying questions and we really narrow down the exact headphones that will work for me.
[14:59] And this all worked because I'm ready to buy now.
[15:04] So, now the agent asked me for some information.
[15:10] So, obviously my email, name, address for shipping, of course.
[15:16] I pick expedited shipping because I definitely want my headphones soon.
[15:20] And then the last part is entering my credit card.
[15:25] And now I'm thinking, am I really going to give my credit card to an agent Ibuild?
[15:32] Like, am I am I trustworthy?
[15:34] How do I know it's safe?
[15:37] I think I need to learn more about UCP's built-in guardrails before I can feel safe entering my credit card number.
[15:45] And this is where something called the shared payment token comes in.
[15:50] And a shared payment token is a token representing a raw card number or wallet, like Google Pay or Apple Pay or any other kind of wallet.
[16:00] It can also include fraud signals and customer reputation data and anything else agents and merchants want to share at the point of purchase.
[16:11] And here's how a shared payment token is used in a transaction.
[16:16] So, at that point in the demo, the agent had requested a payment method.
[16:22] Um, it's requesting this actually from the payment provider, which in the case of the demo was Stripe.
[16:29] Um, that is that was the form that I was going to enter my information in.
[16:35] And what the agent re- uh receives in return though is not the credit card number, it is the shared payment token.
[16:44] It then passes that token onto the seller and the seller unwraps the token.
[16:49] So, they get the payment credential and uh any fraud signals and other data the seller might need.
[16:57] Then the seller passes that onto the payment provider again.
[17:01] So, the seller also is not getting my card number.
[17:06] They're passing the token to the provider and then the provider responds with either a success or failure message, of course, depending on uh if I have the right funds, if the credit card is valid, and so on.
[17:22] And finally, the merchant confirms the order, sends it to the agent that sends it to me.
[17:31] So, shared payment tokens are designed with security in mind, and the payment provider enforces all of the limits, not the agent or the merchant.
[17:40] So, if any guardrail is violated, such as an expired token or an invalid amount or current currency, the charge is just rejected.
[17:50] Okay, well, I know my agent is using UCP, so I know it only has access to the shared payment token.
[17:58] So, I actually feel pretty good about entering my credit card number as that's going to Stripe and not my agent.
[18:05] So, okay.
[18:06] So, now the agent has everything it needs to complete my purchase.
[18:11] And it once again, asks me if I'm sure.
[18:17] It confirms with me.
[18:18] I say place my order.
[18:20] And it comes back with a success message.
[18:23] And because I chose the express shipping, I get my headphones the next day, and they're there in my studio at home.
[18:33] So, if you want to learn more about agent to commerce, uh we've got lots of videos on the Stripe Developers YouTube channel.
[18:42] And uh ton of blog posts that go into even more detail uh on stripe.dev, and I'll be right outside to answer any questions.
[18:51] Thank you.
[19:08] [music]