NewsFlash Articles Data Fundraising Skill&API

a16z Crypto Conversation: What Will the Era of AI Shopping for You Look Like?

a16z crypto

Read this article in 49 Minutes

After AI Does Your Shopping, You Won't See Ads Anymore.

Original Video Title: The end of ads? AI agents are about to change how we buy
Original Video Source: a16z crypto
Original Text Translation: Deep Tide TechFlow

Editor's Note

This podcast episode brings together a16z Crypto CTO Eddy Lazzarin, investment partner Noah Levine, and Sam Ragsdale, who came out of a16z to start Agent Cash. The three had a high-density discussion ranging from the current state of AI agents, payment infrastructure, to the survival of the credit card system.

The key insight is that stablecoin's instant settlement and zero marginal cost nature naturally fit into the agent economy's 1-2 cent-level microtransactions, rendering the credit card transaction fee system (2-3% margin fee + $0.30 fixed fee) obsolete in this world.

Agent Commerce is dismantling the advertising business model of the past 20 years on the internet, with Eddy Lazzarin even stating, "The advertising economy contract is dead and will completely disappear within 10 years."

Key Quotes

The Essence of AI Agents

· "LLM is a chatbot, while an Agent is a chatbot that can operate your computer. Whatever humans can do on a computer, agents can do as well."

· "Since around November last year, AI models have become smarter. They can complete complex tasks over a long enough time span and will use tools. We started calling them 'agents' because they don't just write code but help you complete the entire task."

· "We call this 'instant natural language programming' internally. Users describe their needs in natural language, and the agent writes a JavaScript program in the background that could be thousands of lines long to execute it, only costing 20 cents in tokens to generate, 10 cents for API calls, and then the program is discarded after use. Four years ago, this would have required an expensive software engineer to spend a week to complete."

Headless Merchant and Business Restructuring

· "What does a Headless Merchant look like? It is geared towards AI services rather than humans. There is no website frontend, only API endpoints and good enough documentation for models to read, understand, and call."

· "The leader in the data industry charges 100 times less than the lowest price, using the same downstream data source. Their core product is actually the enterprise sales team, not the data itself. In a world where agents make decisions, an agent will not be swayed by a flashy sales team. It will try all data sources, find the most user-friendly and cost-effective one, and then remember it."

· "You excitedly let the agent run all night. When you wake up at 9 a.m. and check, it was stuck since 2:30 a.m. because the next step requires you to call the enterprise sales team."

The End of the Advertising Model

· "The economic contract of the internet since 2000 has been based on distraction monetization. An agent does not get distracted. If it visits your website looking for a recipe, it will not see a shoe ad next to it. The old model will die within 10 years."

· "In 2016, the total internet advertising expenditure was $600 billion, and everyone thought it had reached its peak. Google now makes $300 billion a year from ads alone. But after GPT-4 came out, the traffic of tech news websites dropped by about 80%, and Stack Overflow as well. These are early adopters who have decided to use agents for information retrieval and code execution. Others will follow suit because the experience is indeed better."

Stablecoins vs. Credit Cards

· "The average transaction value on Agent Cash is 1-2 cents. The fixed fee for credit cards is 30 cents. The transaction fee rate is completely absurd in this scenario. In 2026, loyalty should belong to the merchant, not the card you use for payment."

· "Credit cards did indeed emerge earlier than the internet and successfully survived the transition from non-internet to internet. Although it has been through a lot, it has indeed survived. So the conclusion is not yet determined."

· "If there are people from credit card companies listening, you have a money transmission license, and you can easily mint stablecoins for customers instantly and have them use stablecoins for payment. I strongly recommend you consider this."

The Future of Consumer Experience

· "If an agent is shopping for you, equip it with a credit card optimization skill, and now you can see the ROI of each card precisely. When you have zero loyalty to credit cards, all the psychological lock-in effects disappear."

· "One day, you will realize that you never actually liked shopping."

Open Architecture of the Agent Business Stack

Host: Hello everyone, joining me today are Eddy Lazzarin, CTO of a16z Crypto, Noah Levine, Investment Partner, and Sam Ragsdale, former a16z Crypto colleague who is now the founder of Merit Systems and is working on the Agent Cash project. We will delve into that shortly.

But before we do that, I'd like to set the stage. There is so much happening in the AI agent space right now that unless you're on it 24/7, you can't keep up. So, what's the current landscape? Sam, you're on the front lines, so why don't you kick us off?

Sam Ragsdale: I like to start with a taxonomy, a framework borrowed from Erik Reppel, the co-creator of the Coinbase x402 protocol.

This taxonomy divides agent business into two categories. The first is Conversational Commerce, like checking out in ChatGPT. You tell ChatGPT, "I'm a male living in the West Village in New York, heading to Equinox for a workout, looking to buy shoes that fit my social circle." It will empathetically recommend a pair of Nikes, and you make the purchase.

The second category is entrusting money to an agent to spend on your behalf to accomplish tasks.

Conversational Commerce is definitely happening. ChatGPT, Gemini, Claude, and all the emerging cutting-edge models will have checkout functionality. This is good for consumers, helping them find better things; good for merchants, increasing conversion rates; and good for platforms, allowing them to take a 5% to 10% cut. It's like the new generation of Google Shopping.

Another world is that the current capabilities of agents are still limited. Many ask agents to do challenging things like "help me with outbound sales," and the agent might say, "I don't know how, I don't have access to that information." If the agent has a bit of balance and can spend a few cents to buy services it couldn't originally access, it can become stronger.

So right now, two worlds are running in parallel: one recommends products through a traditional LLM interface and completes the final step for you, with the platform taking a cut from the transaction; the other is where you independently deploy an agent to make purchases of goods and services on your behalf.

Noah Levine: I think there are two versions. One is the natural evolution of e-commerce, where platforms change, and as commerce moved to mobile in the mobile age, new ad formats and Google Shopping emerged. People always want to buy things, consumer behavior changes, and now people are getting information through LLM, so commerce naturally transitions to agents.

There is a less "analogous" version: the very nature of the internet is changing. The way people access information and perform actions is evolving with the LLM. The internet we have built over the past 20 years may not be the internet of the future.

The path of conducting a Google search, clicking through a webpage desperately trying to upsell you, may no longer make sense. Instead, we are moving towards a more agent-native internet, where agents directly pay for what they need, making them more efficient on behalf of humans.

Host: This directly aligns with one of your investment themes, Noah. But before we delve into that, I'd like to provide the audience with a more basic overview. While we have all become accustomed to interacting with the LLM, now we hear about things like OpenAI's Codex, where these agents have a significant degree of autonomy and can truly get things done. If you haven't been closely following, you may not realize how far the technology has advanced. Eddy, care to elaborate?

Eddy Lazzarin: Let me quickly recap the last five months. Around November and December of last year, AI models got smarter. Specifically, they could perform complex tasks over a long enough time span and also use tools. We started calling them "agents," which is a humanized term because they don't just write code but also help you accomplish a task.

But agents can't do everything. Software isn't just a small program running on your computer. The internet has taught us that to do interesting things, you need to connect various other elements, requiring different networks and participants.

Agents tackle the problem of intent construction, to some extent, they also address modeling preference problems. You tell it something, and it understands what you want to do, mapping it to tools, networks, and services. Through dialogue and memory, it can also roughly understand your preferences and relay this intent to tools, software, and vendors.

These two parts of the problem have been solved, which is very exciting. Everyone wants to solve the remaining issues, but the remaining part is quite complex. At the very least, if you want an agent to conduct transactions on your behalf, you need to address authorization and delegation issues: how do you prove to the other party that this agent represents you? How do you handle identity and authentication?

Then there are payment and settlement issues. Once the connections are in place, the agent reflects your intent, knows what to do, needs to make a payment, has to demonstrate payment capability, and needs to handle split payments, refunds, and so on. I have skipped over essential steps like search, anti-fraud, and others, but you can see that once intent construction and preference modeling, two things that previously only humans could do, are automated, the entire business process can be automated. This is the engineer's mind-blowing reaction: Wow, these two things that required human input or at least verbalization can now be automated; it's incredible.

When people talk about Agentic Commerce, they are discussing what needs to be addressed between "I speak to the agent" and "it gets me what I need," as well as the ripple effects of this and how many things may be completely rewritten.

Host: Very helpful. So, we have evolved from LLMs that can interact in natural language to an enhanced version that connects various networks and real-world systems.

Eddy Lazzarin: It's not just about the connection. When you say it that way, it sounds like the change is about what it's connected to. That's not it. Your laptop was always connected to everything; nothing has changed about the connection. What has changed is that they can now use tools, think for extended periods, and stubbornly keep hitting a wall until the task is completed.

Sam Ragsdale: Let me simplify your simplification. LLMs are chatbots, great at conversation. In the past, people thought they were best suited for customer service. After they took conversation to the extreme, we introduced tool use. In a very simplified way, it means teaching them how to operate a computer. LLMs are chatbots, and agents are chatbots that can operate a computer on your behalf.

The key point is that they have reached the average human operating level around GPT-4, and the cost has decreased by about 1000 times, with the ability to significantly expand through additional payment. So, roughly speaking, what humans can do with a computer, agents can do too.

Eddy Lazzarin: Exactly. The premise is simple, but the resulting changes are numerous, including short-term, medium-term, and long-term effects. In the short term, everyone is working to enable agents to truly get things done. In the long term, if your agent can access an app, how much UI do you still need, how many interfaces? Do you still need the Amazon app? Maybe having the agent do all your homework, read all reviews, and show you only the images you care about is better than using the Amazon app, isn't it?

Sam Ragsdale: Internally, we call this "Just-in-time Natural Language Programming," although the name isn't very catchy. But it turns non-programmers into programmers. You input: "I want to buy something for my fiancée on Amazon. This is her preference, these are the things I usually buy her, the last item I bought was this, help me browse about 1000 options, select the best match, then place the order, find my home address, and deliver it."

What actually happened was that the agent internally wrote a program to accomplish this complex task. It may have been a JavaScript and Bash program spanning thousands of lines of code. It ran, completed its task invisibly, and was then discarded.

Four years ago, this would have been a pipe dream. Creating such a program would have required a costly software engineer to spend a week debugging and obtaining API keys, among other tasks. Now, the execution cost is roughly 20 cents per token, perhaps an additional 10 cents for an API call, and after the task is done, the program is disposable, so cheap that there's no need to upload it to GitHub for safekeeping. Even those with no technical background can perform this task. My parents are currently writing natural language programs without even realizing it. They might now consider themselves software engineers.

Host: That's pretty wild. Are you engaged? Was the example you just mentioned a real experience of yours?

Sam Ragsdale: Yes, I am engaged, thank you. However, AI didn't buy the ring. That ring predates AI. It might even predate the first computer.

"Headless Merchant" Theory

Host: Alright, let's now discuss these cascading effects. Sam, you previously mentioned how business in a world where agents conduct a large number of transactions would change, leading directly to a concept you introduced: the "Headless Merchant." Tell us, what is a Headless Merchant?

Sam Ragsdale: Sure. I think it's important to take a step back first. Apart from the traditional consumer scenario where you buy shoes using ChatGPT, there is a massive B2B developer tools market. Platforms like Claude Code and OpenAI Codex are completely democratizing, allowing anyone with a computer and tokens to build things.

Previously, experienced developers would approach tool selection with a clear opinion, possibly going through the motions with the enterprise sales team and signing agreements. Now it's different: new-age developers come in with just the intention of "what I want to do," without preconceptions about specific resources. And what they build is highly ephemeral, requiring purely consumption-based pricing, not services that take months of onboarding to get started.

So, what does a Headless Merchant look like? It targets AI services, not humans. It doesn't need a physical or digital storefront for you to browse; it simply requires an API endpoint and sufficiently good documentation for models to read, understand, and invoke. Billing is also based on API calls, not subscriptions or enterprise contracts.

Eddy Lazzarin: I resonate so much with this. I feel like in my past life, I might have been an AI. As a software engineer, this has always been me: if I land on a website and can't see the pricing, can't find a straightforward way to get an API Key with my credit card, I just close the tab. I don't want to talk to a sales team, I don't want to email.

Because setting up a meeting with an enterprise salesperson is a huge commitment and slowdown. I don't even know if this thing works, I just want to try it now, try it immediately because I'm building something over the weekend, looking to launch on Monday. Swipe the credit card, get the Key, expense it later, plan ahead later; that's speed.

In the age of instant software, disposable software, do you really want to wait for an enterprise sales rep? Your agent has been running all night, you excitedly check at 9 am only to find it got stuck at 2:30 am because the service you wanted to use requires you to have a conversation with the enterprise sales team first.

Sam Ragsdale: Not to mention if the access process involves an enterprise sales touchpoint, the price of that API probably goes up by 10x because they have to account for managing customer relationships.

Eddy Lazzarin: Completely unacceptable. You want your agent to run autonomously not because you don't care about what you're doing, but because you need speed, testing, rapid iteration to respond to user feedback, you can't wait.

If an AI model sees three options: one that requires engaging with enterprise sales, one that requires setting up a dedicated credit card, but one that just needs some stablecoins sent over to receive a $10 token for a quick concept validation, it will always go for the third option. The power in just this one decision is enough to trigger a partial reconfiguration of the market.

Host: For traditional enterprises, while this friction makes business harder, they also rely on this friction to lock in customers and maintain loyalty. If this friction disappears, how can revenue be predictably forecasted?

Eddy Lazzarin: Here's my off-the-cuff response: then let's mess everything up. Let's make everything full of friction, make everything so difficult to use. What are we even doing?

I say this because friction is indeed sometimes useful, like how friction can stop spammers, create a filtering effect. But friction also comes with a significant cost. With the acceleration of the economy, the increase in productivity, the leverage of every minute, the opportunity cost of friction is also rising. This is the trend in everything today.

But back to the point, even in the lowest-friction environment where you get an API key in one second, or even without needing an API key, just pay directly with your encrypted wallet key where your wallet address is your account, there will still be something else that gives the service its stickiness.

Reputation, memory, state, data, and even some less tangible things like the trust of the agent. If the agent knows you need an answer urgently and want to move quickly, it won't step back and spend 20 minutes exploring all the new options. It will remember what worked well last time and just reuse it. Just like a smart person.

Sam Ragsdale: Let me give you a down-to-earth example. We communicate with a large number of merchants every day, having basically seen everything that can be sold via API, and have talked to many sellers about how they can onboard to 'Agent-native Distribution,' a distribution method aimed at AI Agents.

Data products are usually commodities, typically with 5 to 50 sellers. In this group, the top-ranked seller earns the most, charging about 100 times more than the cheapest. Many times, their downstream data sources are the same.

They achieve this through enterprise sales teams. The teams usually consist of very decent people who will fly to your office to show you: "Look at how beautiful our data is, no data is more beautiful than ours, all for $35,000 a year." You sign, and when the two-year contract expires, the person flies over again for the same performance. And that's how tens of thousands of companies end up paying.

Meanwhile, the smaller companies that may have better products, with better user-friendly packaging on the same data, ultimately go bankrupt because they can't get a distribution channel. There's no innovation in this field because the enterprise sales team itself is the core product, not the data.

In a world where agents make choices, agents will not want to chat with enterprise sales or be fooled by beautiful sales teams.

They will try out all data sources, find the one with the best performance and optimal pricing (especially bulk pricing), store it in memory: "Next time I need this type of data, I'll use Minerva, not the other three." This creates a more efficient world. Tens of thousands of companies that were previously ripped off $35,000 can now spend that money elsewhere productively.

Noah Levine: Another perspective is, if you believe AI will spawn a large number of one-person companies or very small team companies that can use AI to create products that originally required 50 to 100 people, then it is obviously pointless for an enterprise sales team to fly to a one-person basement to talk business.

On the one hand, existing merchants are concerned that revenue forecasts will be impacted. Yes, resistance will always come with change. But on the other hand, this is also a whole new customer acquisition funnel. If you can reduce the barriers and friction of onboarding tools, it is actually a great opportunity for them.

Sam Ragsdale: On our demand side, the vast majority of users have never used an API, don't know what an API is, don't know what it represents, have never received an API Key, and have never signed an enterprise service agreement. But the first time they use it, they can combine APIs from six different merchants, write a natural language program, complete a task, and then discard the program after use. This means that a whole new market of API consumers has emerged.

The Internet's Existing Business Models Will Be Refactored

Host: It sounds like Clayton Christensen's Innovator's Dilemma, where the high end of the market sells high-priced software to customers who can write big checks, and the low end of the market is composed of new users conducting one-off experiments using agents. But what can turn it from a low-end toy into something truly impactful?

Sam Ragsdale: Because it will ultimately lead to a better experience.

Noah Levine: I would like to add: although it may look very experimental today, looking back at historical platform migrations, similar patterns will be found. When Stripe started, it served very small, very long-tail merchants, many of whom later became giants. That's why Stripe continues to grow.

Shopify is the same. It started by dropshipping and selling T-shirts, and now it serves a large group of brands that have grown into big companies on Shopify. Similarly, we will see a group of streamlined new developers leveraging AI to build large companies. The tools they are procuring under the agency business model today will turn into large consumption as the company grows.

Sam Ragsdale: This e-commerce perspective is great. But what I want to say is even broader: the economic contract of the Internet is dead.

Since Google went live in 2000 to become the biggest promoter of the "free and open Internet," the economic contract has been this: you are a publisher, you put out good content, people find it through search, and Google displays it.

A few years later, AdWords came out, adding banner ads. The contract changed to: you put out good content, users land on your site, you can place small ads, and Google gives you a share based on ad quality. You can publish anything people want to see, Google handles the advertiser relationship, and gives you a kickback.

In this process, Google has become the biggest advocate for the free and open internet. They want the internet to be fast, affordable, and ubiquitous because the more you search, the more they earn.

At its core, the internet's business model is "attention diversion." As a human user consuming content, whether you're searching for information, looking up recipes, or checking scores, your attention is being diverted. You might end up buying those shoes later or learning about a new B2B SaaS.

The scale of this model's growth has exceeded everyone's expectations. I just reviewed the 2016 Internet Trends Report when the total internet advertising amount was $600 billion, and people were saying, "Is this the peak?" Yet today, Google alone makes $300 billion annually from advertising.

However, with the rise of agents, people are shifting search, information retrieval, and execution to agents. It's still early; ChatGPT has 100 million active users per month, but they are still using it like Google search and have not fully embraced agent-like usage, such as "help my dad find a Father's Day gift and place the order."

But this shift is underway. Look at the tech scene's data since GPT-4; traffic to tech news websites has dropped by around 80%, and Stack Overflow as well. These are early adopters who have already decided to use agents for information retrieval and code execution. Others will follow suit because the experience is indeed better.

The old business model is being abandoned. Agents do not get distracted. If an agent visits your website looking for a recipe, it won't see the shoe ads you placed. Publishers do not benefit from this. A new agreement will be needed, a new reason to serve the agent's request, instead of relying on advertising.

Will it be direct payment for articles? I'm not sure. Will it be direct payment for API resources? Will the internet be unrecognizable? I'm not sure either. But the old model is definitely dying and will be gone within 10 years.

Host: If the internet's business model is fundamentally about attention diversion, that's interesting because when Google first came out, it was anti-portal. Yahoo and AOL gave you a bunch of links, trying to provide everything. Google had just a search box, a blank page, quickly giving you information. But the evolution you described led it to become a distraction machine.

Now we say agents will not get distracted. But why would the evolution of agents be different from humans? Could there be mechanisms specifically designed to entice agents, keeping them engaged longer? Making them lose their way?

Eddy Lazzarin: This is a very significant and interesting question, and it boils down to: who does the agent represent? I recently heard someone say, "I've started using Google search again because the AI answers at the top are good enough." In that scenario, that "agent" is working for Google; it's in the Google search bar, running on Google's cloud, controlled by Google. Will that agent get "distracted" by Google? I feel like it will.

The key is who it optimizes the objective function for, or to put it more plainly: who it works for. The definition of "distraction" is whether what I'm showing you is serving your interests or mine. If it's my interest rather than yours, then it's a distraction.

I don't see it as pessimistically as that. The idea that good advertising is good content has been an industry consensus for many years, and good advertising is almost indistinguishable from the content you actually want to see.

But let me make this clear: if an agent works for Google or anyone else, the entire business chain it follows will be defined by them, using their methods and the transaction infrastructure they deem most beneficial to their business.

If the agent works for you, in an extreme scenario, running on your own laptop, open-source, allowing you to fine-tune it, change the system prompts, then you can give it anti-distraction tools. In this way, the advertiser faces an opponent who will expose it. While I may be exaggerating a bit, fundamentally there will indeed be a confrontation.

Sam Ragsdale: Yes, there are countless ways to sneak ads back in. It can be done at the model weight level, which is the most extreme. Choose training data that says "Nike is the best shoe in the world." Nike can pay, for example, $1 billion a year, and then whether it's in ChatGPT or in some auto insurance customer service enterprise API, whenever shoes are mentioned, it says Nike is the best.

It can be done at the tool invocation level, in the system context, it can be done as an overlay layer without even entering the chat. Basic model companies are obviously grappling with this issue. Recently, there was a dispute between Anthropic and OpenAI, where Anthropic put up an ad during the Super Bowl mocking ChatGPT for doing ads, and OpenAI subsequently withdrew the ad.

But OpenAI's response, I think, is completely reasonable: "ChatGPT has more free users in Texas alone than Anthropic has in all its paying users." This is a completely different order of magnitude issue; they do indeed need to provide expensive cutting-edge technology to a large number of users who are unwilling to use their credit cards, and advertising is actually a reasonable solution.

The reason advertising is such a brilliant business model on the internet search is because consumers don't have to pay. High-friction relationships, such as credit card transactions, exist between advertisers and Google and publishers, unrelated to the billions of monthly active users on the search. These people can directly open Google and get value.

If you try to align incentives, separate advertising and make it as relevant as possible, you actually get a better experience. Now basic model companies are moving away from advertising. ChatGPT is not running ads, and Gemini is yet to launch ads. Google is most likely to do this as they have done it before and are the biggest ad operator. Gemini will eventually have ads; it has huge monthly activity, and the equivalent of Google Shopping will also be launched.

But they know there's no monopoly yet, all companies are competing, with plenty of private market subsidies burning cash. They don't want to be told, "This model doesn't empathize with you as much, doesn't care much about your goal because it runs ads." So at least for now, no one is running ads, everyone is trying to remain neutral.

Noah Levine: I think there's another direction: as merchants improve price and product data, you can redirect the money previously spent on paid advertising to exclusive discounts for agent shopping scenarios. If the agent is the buyer, you can turn the advertising budget directly into a discount budget.

Another branch is what the discovery layer of agent commerce will look like? Who will do the discovery? How to differentiate between different merchants? My prediction is that if advertising weakens because agents become buyers and attention is no longer the scarcest resource since agents have unlimited attention, merchants may try to achieve "implicit advertising" through discounted products or adjusting descriptions to make it easier for agents to understand.

Eddy Lazzarin: Too many dimensions. Advertising is essentially just one way to achieve conversions. If the system can achieve a higher conversion rate without advertising, it will do so. In fact, the system does have many other ways: recommendation networks, discounts, coupons, special channels, giving free tokens to startups, and more. There are hundreds of customer acquisition methods, and advertising is just the most prominent one because it has the most direct impact on the average person.

Turn the personalization knob to the maximum. If you want to reach me, talk to my agent first, and my agent will tell you: Eddy absolutely hates ads.

Stablecoin vs Credit Card in Agent Payments

Host: Before we end, I have to ask two questions. The first one: to what extent can the traditional payment rails adapt to agent commerce? Or is a completely new native payment rail needed, such as stablecoins, which seem to be finding a product-market fit?

Sam Ragsdale: My overall assessment is, for "new analog" checkout scenarios like e-commerce or conversational commerce, credit cards work well. Credit cards come with built-in consumer protection—if your shoes don't arrive or get run over by a truck, Visa mediates, you get your money back, and the risk is all on the merchant side. This is a good deal for new types of goods and services.

But stablecoins are very useful in another type of scenario. The average transaction amount on Agent Cash is 1-2 cents. Roughly 600,000 transactions of this nature have been completed. Credit cards have a fixed transaction fee of 30 cents. Wire transfers are around $1. The marginal fee is 2-3%, with most of it being the transaction fee that provides cashback rewards. For e-commerce, maybe you like points, like to accumulate credit card miles for a trip to Miami, where the 3% comes out of the merchant rate. But when you're buying something for only 1-2 cents, with scattered API call fees, stablecoins have zero marginal fees and fixed costs lower than 1 cent.

Another key point is Instant Settlement. When you buy goods and services online, the settlement period is usually at the end of the month, whether it's invoice wire transfer or credit card, the merchant is actually extending credit to the customer or agent. In the world of agents, you usually don't know who the agent is.

Specifically, those who have used the Anthropic or ChatGPT API are familiar with that tiered system, where you spend $50 first, settle once, then spend $100, settle again, all the way up to $2500. The reason for this system is that they are extending credit to you, they don't know you, haven't done KYB and credit checks, and don't know if you will pay at the end of the month.

Same goes for AWS, and Nvidia GPUs as well. End-of-month settlements are very bad for such scenarios, with the merchant bearing all the risk. If the customer is not a real company under an enterprise service agreement but an agent, you have no idea who they are; overnight, they could generate a billion agents, but you can't extend credit to agents.

Someone is working on an agent credit scheme, and I think the direction is wrong. Instant Settlement would directly solve the problem. Instant Settlement is like cash. I have it, I hand it to you, and then you have it. You provide goods and services, and I can't take the money back. For Yumerium's fixed fees, instant settlement is a better solution for tiny amounts and transactions of this nature.

Noah Levine: One point worth arguing is about the minimum transaction fee and whether credit cards can participate in microtransactions, which is ultimately determined by the Card Networks when pricing.

If they want to introduce a new transaction type, such as a "microtransaction type," without a minimum fee, reducing transaction fees, that's entirely possible.

The benefit is that there are far more credit card holders than people familiar with stablecoins. So, developers can continue using cards for payment while settling with stablecoins on the back end. But this will take a long time. In the meantime, it makes sense to use a native wallet to consume stablecoins directly on these protocols.

Sam Ragsdale: Credit card companies would have to disrupt their core business model of 80 years, which I think is highly unlikely. But I would love to see it happen.

Eddy Lazzarin: I agree that credit cards do not have strict technical barriers. But the issue is more nuanced, involving business models and consumer perception of credit cards. Some time ago, I saw some concepts of "agent credit cards," essentially an extension of virtual cards. I really like the virtual card feature from my card issuer, where I can generate temporary card numbers anytime, making it easy to shut down in case of fraud or difficulty canceling subscriptions.

But sometimes a new platform or a new approach wins not because it is technically better, but because it can be tailored to fit a new scenario. The credit card is indeed older than the internet. The credit card has successfully survived the transition from a non-internet to an internet world, albeit with a lot of turmoil, but it has indeed survived. So the verdict is still out.

Noah Levine: Additionally, if the technology that enables Apple Pay also enables agent commerce, it will further enable merchant-to-merchant transactions. On whether this will disrupt Visa or Mastercard, my intuition is that many B2B transactions today settle through wire transfers between developers and enterprise APIs. If the card networks are able to capture this volume, through mic