The Agentic Commerce Stack — Ahnaf Prio, Best Buy

AAI Engineer
Internet TechnologyBusiness NewsComputing/Software

Transcript

00:00:00My name is Anaf Priyo. I'm a Senior Engineering Manager at Best Buy. Me and my team are working
00:00:18together right now to figure out what does Agentec Commerce mean and how can we meet
00:00:23our customers, where they're at, and the newest place that they're at is at Agentec Services.
00:00:29I'm excited to give my talk today. And, well, what's my credentials? Where ever since I was
00:00:35a young boy, I dreamed of high throughput inference, harnessing my tools within a context
00:00:40window, kept in check with evals. Yeah, that's absolutely correct. In 2003, all those things
00:00:47definitely existed. I kid. Over the last one year, we have been learning a lot. Shopping
00:00:54isn't new. Shopping is probably one of the most fun things one can do and one of the most essential
00:00:58things that people need to do ever since the economy existed. But I have been super excited
00:01:05by it. So, I'm going to go talk about what are the things, some of the things that I've learned,
00:01:11and hopefully share the notes. So, what is Agentec Commerce? I'm not going to go over the broad
00:01:17definition again. But basically, it's the idea that AI assists will help you with their shopping
00:01:24journey. Shopping has different facets to it. For instance, there's discovery. There's figuring
00:01:31out the aspects of do I actually truly need it? Understanding and deciding. There's loyalty.
00:01:37There's pricing. There's fulfillment. Post fulfillment. It's a lot. And believe it or not,
00:01:44right now, right now, about 45% of all agent sessions that happen within major providers
00:01:51like ChatGPT.com and Google Gemini are related to shopping. Maybe it's a little biased that
00:01:58I don't use it as much, but I'm an engineer. But the humans out there are using AI and talking
00:02:04to them to help with their shopping journey. So, it's also not a binary, you know. Right now,
00:02:11we're at that state of human in the loop. The ideal state would be autonomous shopping. You tell what
00:02:17you're excited about. Your agent goes around, talks to different merchants. I'm originally from Bangladesh.
00:02:24We haggle a lot with the merchants too. Maybe does that, negotiate, does the payment. But right now,
00:02:30we're in the human in the loop. And the talk today is going to talk about the mental model of how that
00:02:36human in the loop is working right now. I'll also provide you with the architecture if you choose to
00:02:41extend it or show you the vision of how autonomous shopping might work. So, this isn't our first attempt.
00:02:50Even a year ago, there were people trying to figure out how can we automate this. Even now,
00:02:56you can go probably download the cloud Chrome extension. Maybe you've used Atlas where you tell
00:03:02the AI you need something. You need headphones. You need that grocery list of items that you have been
00:03:10meaning to buy but never made the actual effort to show up because, you know, you didn't have the time.
00:03:16So, why don't you take screenshots, read the DOM, navigate to the merchant site, fill forms for me,
00:03:23do loyalty. It kind of just didn't work as expected. It was really clunky and slow and brittle. And if
00:03:31you are a merchant who's trying to sell stuff, any engineering department of that merchant will tell
00:03:36you an AI impersonating your browser is just firing up all the alarm bells. So, a lot of times, you will
00:03:46probably be even stuck on the payment flow because we don't want you to be using AI to put in that order.
00:03:53Or at least, in that phase, that's what was happening. So, what did actually work? And is it actually
00:04:00working right now? It is. ChatGPG shopping, Google AI mode is doing just that. Right now, Agentech shopping is
00:04:11considered to be a $7 billion industry and might go up to $65 billion industry by 2030.
00:04:19And the majority of the shoppers are using the mainstream conversational AI assistance,
00:04:25which is on the browser or in your app, ChatGPG and Google Gemini. We're also seeing that pop up in
00:04:31Instagram and Facebook. Meta wants to do Metacommerce now. I heard GoPuff and Grok came together to make
00:04:40an app as well. And also, Microsoft Copilot just yesterday announced in the UK that you can buy
00:04:45Ray-Bans now inside Microsoft Copilot. So, to make that happen, Google and OpenAI separately came up with
00:04:55their own little primitives, ACP and UCP, which is basically talking about how you would actually
00:05:02talk to us. For some of you who are shopping on the other side as the customer, there is not much of
00:05:08a difference between adding an item to cart, adding a second quantity. But to us merchants, that's a
00:05:14second line item, buddy. That's not the same skill. So, if we don't talk about the nuances and the
00:05:20primitives of commerce and standardize it, things will just not work and will remain to be clunky.
00:05:25So, ACP was ChatGPT's attempt at it and Universal Commerce Protocol, UCP, was Google's attempt at it.
00:05:33So, now that I've already established that this is happening, I just wanted to say that it is
00:05:39happening as easy as you go to the ChatGPT Gemini to tell it to find me cat cookies. More to why I chose
00:05:47cat cookies later in this example. The AI surfaces the product, agent calls the merchant checkout API,
00:05:54no browser, payment flows via scope payment mandate or a delegated payment token, an order confirms and
00:06:02human kind of didn't have to touch the cart. So, to all of this that's happening for the user,
00:06:09a lot is happening on the other side. And it's kind of overwhelming. One day we're talking about MCPs,
00:06:15another day, A2A, ACP, UCP, AP2. Like, what is even real? Like, if someone came up to me tomorrow and
00:06:23said, I came up with HYPE, I would probably think it's probably real. So, I wanted to dissect this
00:06:29mental model for you as I've learned about it more. MGP is still the model context protocol,
00:06:35the way that the AI agent identifies the tools. So, maybe we can figure out what does this AI agent
00:06:40specifications are, to showcase what products they have, to showcase the details of a specific product,
00:06:46to showcase loyalty. A2A is how agents talk to each other. They're more of a spec. ACP, UCP are the
00:06:53primitives, and AP2 is the agentic payment protocol scope, payment mandate that Google's open specification
00:07:00came out. And we'll talk all of them one by one, how they actually relate to agentic shopping. So,
00:07:06the MCP tool access is very important because without knowing the different capabilities and
00:07:14hitting those different capabilities, taking the time to bring it into context, understanding the user's
00:07:19memory, the agent will never be able to figure out what you're even trying to do. And the only way to get
00:07:24access to the specific capabilities is through MCP tool calls. The next one is A2A. So, now,
00:07:32there are different ways to architect this, different capabilities I talked about, like payments.
00:07:38Let's say, what do you call it? Loyalty. You could make agents about specific domains itself. Sorry.
00:07:46Right? And if you have specific domain level agents, agents need to talk to each other. We need to find a
00:07:52standardize way to talk to each other. So, A2A, the specifications kind of fill in that gap. Also,
00:07:59if your customer agent and your merchant agent need to talk to each other, maybe you could, they're
00:08:04both agents, maybe we can use A2A. So, now to the UCP/MCP primitives. So, the most important data is that
00:08:13product data. So, UCP allows for adding the product data in a more organized way. And ACP does the same
00:08:23because we don't want to go through your PDP and crawl and figure out every specific attribute.
00:08:29Merchant, just tell us. And also, those products change a lot. So, maybe you can tell us when they
00:08:35change as well. So, send this. So, that kind of data is happening. That kind of data flow is happening in
00:08:42the product feed. Normally, you would assume that this would be a search catalog. However,
00:08:47both ACP and UCP right now, so Gemini and ChatGPT does not support that search catalog call. They want
00:08:54you to send that feed to them. And for those of you who are like, why wouldn't you do that? There's
00:08:59reasons to it. Sponsored products, retail media, related things, ranking. But the most important
00:09:05technological challenges, if you have M number of merchants and N number of products, now it has to
00:09:10call that many. While if you send the product feed ahead of time, we can index it and be ready to
00:09:16offload when you ask for something. I've also put an example of Meta's product feed. As you can see,
00:09:22they're similar but still different. Everyone has an opinion. They think their opinion is the best one.
00:09:27And that's what they're rolling with. So, there's three different specifications right here.
00:09:32So, now that we talked about product feed, talking to each other, calling tools, let's talk about
00:09:37payments. Right now, none of them are supporting the more autonomous form of, you know, X402 or some
00:09:47other kind of payments. We're just not there yet. We're just not confident yet. We want more human in
00:09:51the loop of a merchant to be talking to a payment processor who will take the responsibility or, in
00:09:58this case, liability to actually initiate the payments. So, in ChatGPT, payments only happen
00:10:04through a shared payment token right now. And Gemini UCP, the payments are only being accepted through
00:10:10Google Pay. So, the scoped mandates will tell you what the products are. What I'm excited about is more about
00:10:17AP2, which is an extension of UCP. You see what I'm talking about? There's so many acronyms. AP2 is
00:10:25more about, hey, if we wanted to do autonomous, can you tell me who authorized the agent? What exactly
00:10:30can it buy? And what's the max amount that we should be able to haggle with, maybe? And then the revocation
00:10:38URL and the user concept proof. All right. And all talking, I love building stuff. So, for the sake of
00:10:45this, I have put together a little demo. For those of you who have remembered that cat cookie example,
00:10:52it's because the demo is about my cat. Ginny is my orange tabby. And in this made-up example, Ginny has
00:11:01transformed into a bakery agent. She wants to earn her keep by selling baked goods. So, right now, the
00:11:08model that I'm using is from Cerebrus at 3,000 tokens per second. So, hopefully, this will be really,
00:11:13really fast. And we can give you an example of the entire flow. And just like Chrome DevTools, I've
00:11:20kind of had a couple of tools in place to showcase what happens. The first thing I will tell Ginny, my
00:11:26beautiful cat who's selling baked goods now. Hi. Tell me about all your products.
00:11:36And this is supposed to be a demo. An example. And Ginny has given me exactly that. All the different
00:11:44products that she might need. So, here, let's look at this. The agent to agent protocol actually made the
00:11:50call from Ginny, the customer agent, to the merchant agent. And this is the message being sent. And this
00:11:55is me getting the message back. The merchant agent is returning the completed task. And the way that
00:12:03I found this is through an MCP tool call, which is product search. Instead of, like, not being able to
00:12:13tell what I truly want, Ginny has figured out that, hey, when I give her the intent that I want to find
00:12:20products, we should call the MCP tool called product search. So, right now, we're seeing this. And now,
00:12:27what if I want to add something? Add to cart the shortbread.
00:12:39So, now, Ginny is asking me about any discount and promo code. I actually do not remember any of the
00:12:46discount and promo code. But what if I ask Ginny, Ginny, can you just tell me a discount code? As you can
00:12:54tell, I'm definitely a hackler. Ginny is not telling me that. All right. Proceed to checkout
00:13:04account code without discount code. There you go. So, now we're making some of those calls. Here's
00:13:13the UCP protocol, which the checkout APIs will have state. And the three different states are not
00:13:19ready for payment, ready for payment, and then completed. So, now, here, I'm not using a delegated
00:13:25payment token. I'm not using Google Pay. I like AP2. So, my demo is built on AP2. And as you can see,
00:13:34here was a call was made through the MCP server for create checkout. And then the UCP endpoints will
00:13:42tell us, hey, call the checkout sessions and tell me if it's added to cart. So, it's added to cart,
00:13:49but it's not ready for payment. I have to pick in what I want to pay with. I say credit card and debit
00:13:56card. And this is where I issue the AP2 token that, hey, I do want that. And then it went from ready
00:14:03to ready for payment to complete. So, the other side of it, this is the UCP specs, right? The other
00:14:12side of it, to just draw a comparison, how is it differing from the ACP specs, I have added ACP here
00:14:19as well. So, you can see the same checkout calls, just different, just different schemas are being
00:14:26utilized and the order goes through. But remember that AP2 token that I was talking about? This is how
00:14:33it would look like in real life where the user demo, the max amount is this, the currency is this. If you
00:14:39want to revoke it, you can. And what's the maximum, here we didn't want to haggle, so we just put the
00:14:45max amount of that. And then it's also a single time usage. This demo also has, comes with a timeline, so
00:14:53you can actually open any of these and see these happening. Remember that catalog I was talking about,
00:15:00that they don't do the search? We actually do a product feed, sending it to them. I have added that as well.
00:15:08And the feeds, because they're so different, there's a place to actually compare them. So, here's the feed
00:15:17being called, by the way, if you went to timeline, every couple of seconds we try to get the catalog in
00:15:23sync for what is in inventory, what's not. This is the UCP one, and here's the meta one. So, I've shown you this,
00:15:32and you could reuse this same demo or the same concepts. What if I didn't want to do external
00:15:39agentic commerce on Gemini or ChatGPT? You could still build your own custom implementation of a
00:15:46merchant agent or Gini on your website. Maybe I start selling cat goods. I could reuse some of this,
00:15:55but I would advise maybe look into some of these primitives and trying to use them, because they've
00:16:00been standardized across merchants. So, they have been well thought out. And also, you could probably
00:16:05reuse them to sell externally as well on ChatGPT and Gemini. So, remember, I was talking about the
00:16:14discount codes. There's a reason for that. When we build out this demo and in my time building
00:16:19agenda commerce at Best Buy, we have realized working with AI and conversational experiences
00:16:24without evals is playing whack-a-mole. So, if you choose to use the same architecture for
00:16:29reviewing customer base, like Gini.com websites, think very much about creating evals. One of the
00:16:38things that I could not emphasize more about is you should test, test, and test. If you go over here,
00:16:46I can also run my scripts or run evals. And this eval folder has all those evals. The reason I'm also
00:16:57showcasing the code is there's a template folder here. And we can go back to the slides.
00:17:08And if you don't do evals, this is what might happen. I love Chipotle. I don't know if it's true
00:17:14or not. But I found it really funny, so I'm gonna talk about this. So, this popped up that when Chipotle
00:17:21rolled out their agent, people were using it to ask programming questions, right? If you don't tell
00:17:29your agent to not allow for those kind of things, people will use it. This is hands down one of the
00:17:35most creative ways to get free AI usage when you don't want to pay for that cloud subscription. And if we don't write
00:17:42our evals and test intensely, those things will happen in production. The discount code will be told even
00:17:51sometimes more sensitive things like who else is checking out this product. So, the kinds of evals
00:17:58that I would highly recommend you write is behavior evals, protocol compliance, because when we're selling
00:18:03it. So, when you get to, let's say, gpt, chatgpt.com or Gemini, you want to make sure that the feeds are
00:18:08actually conforming or else they will not support it. You should also think about latency benchmarks.
00:18:14Every second in retail in the shopping journey where you're actually not selling, there are chances that
00:18:20the other website is going to be faster and people are just going to move away or that just don't feel
00:18:25like it anymore. Lastly, I also recommend using LLM as a quality judge. You don't have to use something
00:18:32fancy. Talk to your product friend and figure out what's the best way to do it and use the best use
00:18:39cases and write them out. And I would like to also talk about, now that I've discussed all of this,
00:18:46what's actually stable today and what's still forming. MCP is widely adopted. A2A is widely used.
00:18:52UCP/ACP is out there. What's still forming, though, is AP2 and actual usage of it. ACP versus UCP
00:19:00convergence. Do we always have to do two different specs? Identity constant standards and multi-agent
00:19:05checkout delegation. So, if you have to leave here today with anything, I hope you leave today with a
00:19:14good mental model of agentic commerce works. I have nothing to sell you, but I do have gifts for you.
00:19:19I find agentic commerce really exciting. So, you can find this entire presentation on GitHub and I came
00:19:25up with a template. It's a three service starter. If you want to do customer agent or you want to do
00:19:31the merchant agent, you can do that because I love evals and that saved my life. I have some eval
00:19:37templates for you and you know what if you want to send it to all of your merchants, not just one. I have
00:19:43a catalog sync process as well which will allow you to type into your own product and then turn it into
00:19:49ACP or UCP or meta so you can sell there. And lastly, but not the least, we all know now these days we
00:19:56don't write code like that. If I give you a template, you'll be like meh. So, I have agent skills that
00:20:01specifically does a merchant agent, customer agent, and all those different catalog syncs that we have
00:20:08talked about. I hope you had an amazing time and learned and have fun as much as I had presenting
00:20:15this. Thank you. My name is Anup Priyo and I hope to see you again soon.
00:20:32you

Key Takeaway

Agentic commerce standardizes AI shopping journeys through structured product feeds, protocol primitives like ACP and UCP, and delegated payment tokens.

Highlights

  • Agentic shopping sessions account for 45% of interactions on major platforms like ChatGPT and Google Gemini.

  • Agentic commerce represents a $7 billion industry projected to reach $65 billion by 2030.

  • OpenAI and Google utilize Agentic Commerce Protocol (ACP) and Universal Commerce Protocol (UCP) to standardize merchant communication.

  • Agentic Payment Protocol (AP2) manages spending limits, authorized agents, and single-use transaction tokens.

  • Rigorous behavior and protocol evaluations prevent conversational AI models from leaking confidential discount codes or answering unrelated programming questions.

Timeline

Current State of Agentic Commerce

  • Shopping constitutes 45% of agent sessions on platforms like ChatGPT and Google Gemini.
  • Browser automation using extensions like Atlas remains clunky, slow, and prone to merchant security blocks.
  • Agentic commerce operates as a $7 billion industry with projections reaching $65 billion by 2030.

Human-in-the-loop workflows dominate current shopping interactions while fully autonomous agent negotiations remain in development. Early automated browser extensions trigger security alarms by impersonating user DOM navigation. Mainstream conversational assistants now handle direct API checkouts for millions of shoppers.

Protocols and Architecture

  • Model Context Protocol (MCP) identifies agent tools and product capabilities.
  • OpenAI utilizes Agentic Commerce Protocol (ACP) while Google employs Universal Commerce Protocol (UCP).
  • Agentic Payment Protocol (AP2) handles spending limits, token revocation, and user consent proofs.

Merchants upload structured product feeds directly into AI platforms instead of relying on real-time search catalogs to optimize ranking and sponsored product placement. Agent-to-Agent (A2A) specifications govern communication between customer agents and merchant domain agents. Payment flows transition from shared tokens and Google Pay toward delegated authorization frameworks.

Live Implementation and Evaluations

  • A live bakery agent demo showcases product searches, shortbread cart additions, and AP2 checkout execution.
  • Uncontrolled conversational agents mistakenly answer unrelated programming questions in production deployments.
  • Behavior evaluations, protocol compliance checks, and latency benchmarks ensure system reliability.

Building conversational retail experiences requires rigorous evaluation frameworks to prevent unauthorized discount disclosures and off-topic responses. Automated test suites verify feed schema compliance and maintain low latency thresholds. Standardized templates and agent skills accelerate the deployment of merchant and customer agents.

Community Posts

View all posts