Dialoguea16z Cripto: What will happen when AI purchases you

The end of ads
source: a16z crypto
Original language: Deep tide TechFlow

Edit Guidance

The podcast gathered a16z Crypto CTO Eddy Lazzarin, an investment partner Noah Levine, and Sam Ragsdale from a16z who started a business as Agent Cash, who engaged in a high-intensity discussion on the technical status of AI agents, payment infrastructure, and the survival of the credit card system。

The core judgement is that the immediate settlement of the stable currency, the nature of the zero-margin cost, the 1-2 cents micro-transaction in the natural matching agent economy, and the credit card transaction cost system (2-3 per cent marginal fee + 30 cent fixed fee) are vulnerable in the world。

“The economic contract for advertising is dead and will be completely destroyed in 10 years.”。

Expensive Notes

AI AGENT NATURE

LLM is a chat robot, Agent is a chat robot that can operate your computer. People can do things with computers, agents can do."

"AS OF LAST NOVEMBER OR SO, THE AI MODEL BECAME SMARTER. THEY CAN PERFORM COMPLEX TASKS OVER A SUFFICIENT PERIOD OF TIME AND USE TOOLS. WE STARTED CALLING THEM "AGENTS," BECAUSE THEY DIDN'T JUST WRITE CODES, BUT THEY HELPED YOU WITH THE WHOLE JOB

"We call this "Programming Real Language." The user describes the demand in natural languages, and the agent writes a JavaScript program that may be in thousands of lines on the backstage. It only takes 20 cents token to generate, 10 cents API call, and then throws it out when it's over. Four years ago, it took an expensive software engineer a week to complete."

No front-end business and commercial reconstruction

• "What's Headless Mercant?" It's for AI service, not for people. There is no front end to the website, only an API endpoint and a good enough document to allow models to read, understand and call

• “The leading sheep in the data industry charge 100 times the lowest price, using the same downstream data source. Their core product is actually a sales team, not the data itself. In a world where agents make decisions, agents are not fooled by a pretty sales team. It will try all the data sources, find the best, the best price, and remember it."

"You got the agent all night. At 9 a.m., it was stuck from 2:30 a.m. because the next step was to call the corporate sales team

The End of Advertising Mode

• “The economic contract on the Internet since 2000 is based on distraction. The agent won't be distracted. If it visits your website looking for recipes, it won't see the shoe ad next to it. The old model will die in 10 years

· "The total Internet advertising in 2016 was $60 billion, when everyone felt it was at the top. Google makes $30 billion a year in light advertising today. But when GPT-4 came out, the traffic on the technology news site dropped by about 80 percent, as did Stack Overflow. These are early adopters who have decided to use agents for information acquisition and code execution. The people in the back will follow because the experience is really better."

stable currency vs credit cards

· "Agent Cash has an average turnover of 1-2 cents. The credit card fixed fee is 30 cents. The transaction rate is completely absurd in this scenario. In 2026, loyalty should go to the merchants, not to the card you paid for.”

• “Card cards do appear earlier than the Internet and have successfully survived the transition from non-Internet to Internet. It did survive, though it was stunned enough. So the conclusion is still open

• “If someone with a credit card company is listening and you have a money transfer license, you can make real-time stabilization coins for your clients, so that they can pay in them. I strongly recommend that you consider this.”

The future that consumers experience

– "If the agent is shopping for you and you're putting it into credit card optimization skills, now you can see exactly each card's ROI." When you have zero loyalty to credit cards, all psychological manipulations are lost."

"One day you'll realize you never really like shopping."

Structures for the opening of proxy business houses

Moderator:Hello, I'm with a 16z Crypto's CTO Eddy Lazzarin, investment partner Noah Levine, and former a 16z Crypto co-worker, Sam Ragsdale of Merit Systems, who is working on an Agent Cash project, and we'll talk later。

Before doing so, I would like to lay the background. There's too much going on in the AI field right now, and you can't keep up until you've been staring 24 hours. So, what's the world like now? Sam, you're building on the first line. You wanna talk first

Sam Ragsdale: I like to start with a classification, a framework borrowed from Erik Repel, co-founder of Coinbase x 402。

This classification divides proxy commerce into two categories. The first is Conversational Commerce, which is the closing of accounts in ChatGPT. And you said to ChatGPT, "I'm a man living in West Village in New York who's going to Equinox to get some shoes to fit in my social circle." It's gonna recommend a pair of Nike, and then you buy it。

The second is to entrust the money to the agent, who will pay for the task。

Dialogue business will certainly happen. ChatGPT, Gemini, Claude, and all the forward models that come out of the back, will be closed. It's good for consumers, helping them find something better; it's good for businesses, it's a higher conversion rate; it's good for platforms, it's 5 to 10%. A new generation of Google Shoping。

In another world, agency capacity is currently limited. A lot of people make agents do difficult things, like "Do me sales outreach," and they say, "I won't, I can't get that information." If the agent had a small balance and could spend a few cents on services that it could not use, it would have become stronger。

SO NOW THERE ARE TWO WORLDS IN PARALLEL: ONE TO RECOMMEND GOODS THROUGH THE TRADITIONAL LLM INTERFACE, TO FINISH YOU OFF THE LAST STEP OF THE DEAL, AND THE OTHER TO DEPLOY AGENTS INDEPENDENTLY TO BUY YOU GOODS AND SERVICES。

Noah LevineI think there are two versions. One is the natural evolution of the electrician, the transformation of the platform, the migration of mobile-era commerce to the mobile end, the emergence of new forms of advertising and Google Shoping. People always want to buy, consumers behave differently, and now the way people get information becomes LLM, and business naturally moves into agents。

THE INTERNET ITSELF IS CHANGING IN ITS FORM. THE WAY PEOPLE GET INFORMATION AND EXECUTE OPERATIONS IS CHANGING WITH LLM. THE INTERNET WE BUILT OVER THE LAST 20 YEARS MAY NOT BE THE FUTURE INTERNET。

This route may no longer make sense by searching for Google and clicking on a web page that works so hard to sell upwards. Instead of a more proxy Internet, the agent pays directly for what it needs to make it more efficient for humanity。

ModeratorThis is a direct response to one of your investment themes, Noah. But before going forward, I want to make a more basic science pop for the audience. You're used to interacting with LLM, but now you're hearing about OpenAI's Codex, and these agents are already quite autonomous and can really do things. If you don't pay close attention, maybe you don't know how far the technology has gone. Eddy, you wanna talk

Eddie Lazzarin: LET ME GO OVER THE LAST FIVE MONTHS QUICKLY. ABOUT LAST NOVEMBER AND DECEMBER, THE AI MODEL GOT SMARTER. SPECIFICALLY, THEY ARE ABLE TO PERFORM COMPLEX TASKS OVER A SUFFICIENTLY LONG PERIOD OF TIME, WHILE ALSO USING TOOLS. WE STARTED CALLING THEM "AGENTS," WHICH IS A HUMANISTIC TERM, BECAUSE THEY DON'T JUST WRITE CODES, THEY DO A JOB FOR YOU。

But agents don't do anything. Software isn't just a little program on your computer. The Internet tells us that you have to connect a lot of other things to do something interesting and that you need networks and participants。

The agent addressed the issue of intent and, to some extent, of modelling preferences. You tell it one thing, it understands what you're trying to do, and it maps it on tools, networks and services. Through dialogue and memory, it also provides an overview of your preferences and conveys this intention to the tools, software and suppliers。

It is very exciting that these two elements have been resolved. All wanted to solve the remaining problems, but the remainder was complex. At least, if you want the agent to make the deal for you, you need to resolve the issue of authorization and commission: how do you prove that the agent represents you? What about identity and authentication

Then comes the issue of payment and settlement, once connected, the agent reflects your intentions, knows what to do, it needs to pay, it needs to demonstrate capacity to pay, it needs to deal with split payments, refunds, etc. I'm skipping the important links of search, anti-fraud, but you can see that once the intent is to build and model the preferences of two things that were previously only human capable of doing are automated, the whole business process is automated. That's how the engineer's brain reacts: God, these two things that humans have to enter or at least say, can now be done automatically. It's incredible。

When people talk about "agent commerce," they talk about what needs to be addressed between "me and agent" and "it gets what I need" and the chain reaction to this, because many things are completely rewritten。

Moderator: VERY HELPFUL. IN OTHER WORDS, WE EVOLVED FROM LLM, WHICH INTERACTS IN NATURAL LANGUAGES, TO AN ENHANCED VERSION THAT CONNECTS NETWORKS AND REALITY SYSTEMS。

Eddie Lazzarin: It's not exactly a question of connectivity. You say that like it's something that changes. No, it's not. Your laptop is connected to everything. Nothing changes in connection. Instead, they can now use tools, think for a long time, and be stubborn enough to hit the wall repeatedly until the mission is completed。

Sam Ragsdale: I'LL GIVE YOU A SIMPLIFIED VERSION. LLM IS A CHAT ROBOT, VERY GOOD AT TALKING, AND PEOPLE USED TO THINK THEY WERE THE BEST FOR THE SERVICE. WHEN THEY HAVE MADE THE DIALOGUE EXTREMELY CRITICAL, WE HAVE DEVELOPED TOOLS TO USE. IT'S VERY SIMPLE TO SAY THAT THEY LEARNED TO OPERATE COMPUTERS. LLM IS A CHAT ROBOT. THE AGENT IS A CHAT ROBOT THAT CAN RUN YOUR COMPUTER。

THE KEY IS THAT THEY REACH AVERAGE LEVELS OF HUMAN OPERATION AROUND GPT-4, WHICH IS ABOUT 1,000 TIMES LOWER, AND THAT THEY CAN SIGNIFICANTLY EXPAND THEIR CAPABILITIES WITH EXTRA MONEY. SO, ROUGHLY, WHAT PEOPLE CAN DO WITH COMPUTERS AND AGENTS CAN DO。

Eddie LazzarinThat's it. The premise is simple, but the resulting changes are numerous, short-, medium- and long-term. In the short term, everyone is opening channels to make agents really capable. Long term, if your agent can get App, how many UIs and how many interfaces do you need? Do you need Amazon App? Isn't it better than having the agent do all your homework, read all your comments and show you the pictures you care about

Sam Ragsdale: We call this "just-in-time natural language programming" within us, even though it's not very catchy. But it turned non-programmers into programmers. You entered: "I want to buy something for my fiancée on Amazon, which is her preference, and this is what I usually buy for her, and this was the last time I bought it, and I went through about 1,000 options, picked out the best match, and then came down, found my home address and sent it."

What actually happened was that the agent had written an internal program to do this complex thing. Could be a thousand lines of JavaScript and Bash programs. It was implemented, but the user could not see it and then it was thrown out。

Four years ago, this was the night. Writing such a program requires expensive software engineers spending a week debugging, acquiring API keys, etc. Now the execution cost is about 20 cents token, possibly plus 10 cents for the API call, and throw away the program after buying something, so cheap that it doesn't need to upload GitHub. People who don't know anything about technology can do it. My parents are writing natural language programs now, they don't even know it. They may now call themselves software engineers。

ModeratorIt's crazy. Are you engaged? That was your real experience

Sam RagsdaleI'M ENGAGED, THANK YOU. BUT THE RING DIDN'T LET AI BUY IT. THAT RING APPEARED EARLIER THAN AI. PROBABLY BEFORE THE FIRST COMPUTER。

"No prior business."

Moderator: Okay, let's talk about these chain reactions. Sam, you mentioned earlier how business can change in a world where agents do a lot of transactions, directly linked to a concept that you have proposed: "Headless Merchant." Tell us what it's about

Sam Ragsdale- Okay. I think it's necessary to take a step back. In addition to the traditional consumption scenario of buying shoes with ChatGPT, there is a huge B2B developer tool market. Like Claude Code, OpenAI Codex, these platforms are completely democratized, and anyone with computers and token can build things。

In the past, senior developers used to select tools with clear views, possibly following the business sales team and signing them. Now it's different: when the new developers come in, there's only the intention of "what I want to do," and there's no stigma about what resources are actually used. And what they build is highly temporary and requires services that are fully utilised and do not require months of access to the process。

SO WHAT'S THE HEAD OF THE BUSINESS WITHOUT A FRONT? IT'S FOR AI SERVICE, NOT FOR PEOPLE. YOU DON'T NEED A PHYSICAL OR DIGITAL STORE TO BROWSE, YOU NEED AN API ENDPOINT AND A GOOD ENOUGH DOCUMENT TO MAKE MODELS READ, UNDERSTAND AND CALL. THE CHARGES ARE ALSO BASED ON API CALLS, NOT SUBSCRIPTIONS OR ENTERPRISE CONTRACTS。

Eddie Lazzarin: I am so resonating. I think I might have been an AI in my last life. As a software engineer, I've always been like this: if I can't get a price on a website and I can't get a credit card to get the API Key entrance, I close the page. I don't want to talk to the sales team. I don't want to e-mail。

Because the timing of business sales is a huge commitment and a slowdown. I don't even know if this thing works, so I'd like to try it now and now, because I'm doing something this weekend, and I want it released on Monday. It would be quick to take a credit card and get a refund later and see the plan later。

In the era of instant software, temporary software, do you really want the agent to wait? Your agent ran all night, and you were excited to wake up at 9 a.m., and it was stuck from 2:30 a.m., because the service you wanted to use made you talk to the corporate sales team。

Sam RagsdaleNOT TO MENTION THAT IF THE ACCESS PROCESS INCLUDED A BUSINESS SALES COMPONENT, THE API WOULD BE ABOUT 10 TIMES MORE EXPENSIVE BECAUSE THEY HAD TO ARRANGE FOR SOMEONE TO MANAGE THE CUSTOMER RELATIONSHIP。

Eddie Lazzarin: Not at all. You have to act on your own, not because you don't care about what you do, but because you can't wait because you want speed, you want testing, you have to respond quickly to user feedback。

If the AI model sees three options: one needs to link business sales, one needs to set up a dedicated credit card, but one gets a ten dollar token proof of concept if you send some stable coins, it always chooses a third. This power alone is sufficient to trigger the restructuring of some markets。

Moderator: For traditional businesses, although these frictions make business difficult, they also rely on them to target customers and maintain loyalty. If these frictions disappear, how can income be reliably projected

Eddie Lazzarin: First give you my mouth page answer: then we'll make everything rotten. It's hard to rub everything. What are we doing

I say that because friction does sometimes work, for example, by blocking spam senders, and friction can have a filter effect. But friction also costs a lot. As the economy accelerates, productivity rises, and leverage increases every minute, the opportunity cost of friction increases. This is the trend of everything。

Back to the point, even in the lowest friction environment, you get API Key in a second, you don't even need an API Key, you pay directly with an encrypted wallet key, the wallet is your account, and there's something else to make the service sticky。

Honor, memory, state, data, even something less tangible, like proxy trust. The agent doesn't take 20 minutes to explore all the new options if he knows you need answers and wants to move fast. It will remember that the last time it worked very well and then went straight back. Just like smart people。

Sam Ragsdale: I'll give you an example of the atmosphere. We talk to a large number of businesses every day, basically looking at everything that can be sold through API, and talking to a lot of vendors about how they can access the "Agent-nature Distribution" (Agent-native Distribution), which is the original distribution for AI Agent。

Data products are usually bulk commodities, typically between 5 and 50 sellers. In this group, the top earns the most and the cost is about 100 times the cheapest. And many times, their downstream data source is the same。

They do that by the corporate sales team. "Look how beautiful our data are, there's no data better than us, $35,000 a year." You sign the contract, the man who's two years old will fly over and do the same show again. And then tens of thousands of companies pay。

Small companies that may be better off and have made better user-friendly seals on the same data because they do not have distribution channels and eventually go bankrupt. There is no innovation in this area, as the corporate marketing team itself is a core product, not data。

In the world where agents make choices, agents don't want to talk to corporate sales, they don't want to be fooled by a pretty sales team。

It will try all the data sources, find the one that works best, priced best (especially bulk pricing), and put it in memory: "The next time you need such data, use Minerva, not the other three." This creates a more efficient world. Tens of thousands of companies that had been killed for $35,000 could spend it in other productive areas。

Noah LevineTHE OTHER ANGLE IS THAT IF YOU BELIEVE THAT AI WILL GENERATE A LARGE NUMBER OF COMPANIES, OR VERY SMALL TEAMS, THAT CAN USE AI TO PRODUCE PRODUCTS THAT WOULD HAVE REQUIRED 50 TO 100 PEOPLE, IT IS OBVIOUSLY POINTLESS FOR THE CORPORATE SALES TEAM TO FLY TO SOMEONE'S BASEMENT TO TALK TO HIM ABOUT BUSINESS。

On the one hand, existing businesses fear that income projections will be affected, and yes, the changes will be contradictory. But on the other hand, it is also an entirely new customer access funnel, and if you can reduce bottlenecks and friction in access tools, it is a great opportunity for them。

Sam Ragsdale: On our demand side, most users have never used API, they don't know what API is, they don't know what it means, they've never taken API Key, they've never signed an enterprise service agreement. But for the first time, they can combine six APIs from six different businesses, write a natural language program, finish the job, and then abandon it. This means that API consumers have a brand-new market。

The current business model of the Internet will be reconstituted

Moderator: Sounds like Clayton Christensen's innovators, with high-end markets selling sky-priced software to old players who can sign big checks, low-end markets being new users who do one-off experiments with agents. But what makes it from low-end toys to really powerful things

Sam RagsdaleIt will eventually become a better experience。

Noah LevineI would like to add that, although it seems experimental today, looking back at the migration of historical platforms will reveal similar patterns. Stripe started with very small, very long-term businesses, and many later grew into giants, which is why Stripe continues to grow。

The same goes for Shopivy, who first shipped and sold T-shirts, and now serves a large number of Shopies that are branded as big companies from zero. Similarly, we will see a very streamlined group of new developers building large companies through AI, which today buy tools under proxy business models, which will turn into large consumption as companies grow。

Sam RagsdaleThe electrician is very good. But what I want to say is even bigger: the economic contract on the Internet is dead。

Since Google became the biggest "free and open Internet" promoter in 2000, the economic contract is like this: you're the publisher, you're the publisher, you're the good stuff, people search it, and Google shows it。

A few years later AdWords came out and added the banner ad. The contract becomes: you put out good content, the user drops down your website, you put a little ad, Google divides you by the quality of the browsing. You can publish anything people want to see, Google handles advertising and gives you kickbacks。

In the process, Google became the biggest promoter of free access to the Internet, which they want fast, cheap and everywhere, because the more you search, the more they make。

After all, the Internet business model is "difficult." When a human user consumes content, whether looking for information, a recipe or a score, you get distracted, maybe later you buy those shoes, or learn about a new B2B SaaS。

The scale of this model has grown beyond all expectations. I just went through the 2016 Internet Trend Report, when the Internet commercials totalled $60 billion, and people said, "Final," but Google made $30 billion a year in light today。

However, after the presence of the agent, people are moving search, information acquisition and execution to the agent. It's early, ChatGPT has 100 million months to live, but they're still searching like Google, and there's no real proxy for it, like, "Get my dad a father's day present and order."。

But it's on its way. Look at the data from the technology community: GPT-4 has seen about 80 percent of the traffic on the technology news site, as has Stack Overflow. These are early adopters who have decided to use agents for information acquisition and code execution. The people in the back will follow, because the experience is really better。

Old business models are being abandoned. The agent won't be distracted. If it's on your website looking for recipes, it won't see your shoe ads. The publisher did not receive any benefit from it. A new contract and a new reason for your service agent's request, rather than advertising, will be required。

WILL IT PAY FOR THE ARTICLE DIRECTLY? I'M NOT SURE. WILL IT PAY FOR API RESOURCES DIRECTLY? DID THE INTERNET MEET AT ALL? I'M NOT SURE. BUT THE OLD MODEL MUST BE DEAD, IN 10 YEARS。

Moderator: It would be interesting if the Internet business model was ultimately a distraction, because Google was the first one to come out. Yahoo and AOL give you a bunch of links, everything you want. Google just a search box, blank pages, quick information for you. But the direction you're describing is precisely where it becomes a distraction machine。

Now, we say the agent is not distracted, but why would the agent be different from humanity? Is there a mechanism that is specifically designed to lure agents and keep them lost for longer

Eddy Lazzarin:It's a big and interesting question, and at the heart of it is: who is the acting representative? I've recently heard people say, "I'm searching again with Google because the top AI answer is good enough." In that scene, the "agent" worked for Google, which was in the Google Search Bar, running on Google's clouds, and Google controlled it. Will that agent be distracted by Google? I feel like I will。

The point is it optimizes whose target function, or it works for whom. The definition of distraction is whether what I show you serves your interests or mine? If it's my interest, not yours, it's distraction。

I don't understand that pessimism. The industry consensus that good advertising is good content has existed for many years, and good advertising is almost indistinguishable from what you want to see。

But let me be clear: if the agent works for Google or anyone else, the whole business chain that he goes through will be defined by them, using the methods that they set out and the trading infrastructure that they think is best for their own business。

If the agent works for you, in extreme cases, running on your own notebook, open source, you can fine-tune it, change the system hint, then you can give it a counter-divisive tool. In this way, the advertiser is faced with an opponent who will break it. Although I am exaggerating a little, there is, in essence, confrontation。

Sam Ragsdale: Yes, there are countless ways to put advertising back. It can be done at the model weighting level, which is the most radical. When you choose the training data, you choose the 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 an API, just talk about shoes and say Nike's best。

You can do it at the tool level, in the context of the system, and you can do it in stacks without even talking. Basic Models are apparently addressing this issue. Recently, there was a dispute between Anthropic and OpenAI, and Anthropic put ads on the Super Bowl and mocked ChatGPT, and OpenAI then dropped the ad。

But OpenAI's response seems perfectly reasonable to me: "ChatGPT has more free users in Texas alone than Anthropic has all paid users." This is a problem of very different magnitudes, and they do offer expensive front-line technology to a large number of users who do not want to take credit cards, and advertising is a reasonable solution。

Advertising is such a talented business model on Internet searches because consumers do not pay. The relationship of high friction, such as taking credit cards, exists between advertisers and Google and publishers, and has nothing to do with the billions of live users searched. Those people who just open Google get value。

If you try to match incentives, separate advertising and be as relevant as possible, you actually get better experience. Now basic model companies are away from advertising. ChatGPT isn't running an ad, Gemini hasn't done an ad. Google's most likely to do this. They've done it before. They're the biggest advertisers. Gemini's gonna have an ad sooner or later, it's gonna be huge, and Google Shoping's equivalent。

They knew, however, that there was no monopoly, that all companies were competing and that large private market subsidies were burning money. They don't want to be said, "This model doesn't have much common sense for you, it doesn't care about your goals because it runs advertising." So, at least for the moment, no one is running advertising, trying to be neutral。

Noah LevineI think there's another way to go: as businesses do better and more transparent with price and product data, you can change the amount that was spent on fee advertising to an exclusive discount on proxy shopping. If the agent is the buyer, you can turn the advertising budget into a discount。

The other branch is the discovery layer of proxy commerce. Who's gonna do the discovery? How do we separate different businesses? My prediction is that if advertising weakens because the agent becomes a buyer, because the agent has unlimited attention, and attention is no longer the most scarce resource, businesses may try to "hidden advertising" by discounting the product, or by adjusting the description to make it easier for the agent to understand。

Eddie Lazzarin: too many dimensions. advertising is essentially just a way of getting conversion. if the system could achieve higher conversion rates without advertising, it would do so. in fact, there are many other ways of doing so: recommending networks, discounts, coupons, special channels, free token for start-ups, etc. there are hundreds of ways to get visitors, and advertising is only the most visible because ordinary people feel the most directly。

If you're going to touch me and talk to my agent, my agent will tell you that Eddie hates advertising。

role of vs credit cards in proxy payments

Moderator: I must ask two questions before the end. First: To what extent does the traditional payment track fit into proxy commerce? Or is there a need for a new primary payment track, such as a stable currency, which seems to be finding convergence points in product markets

Sam RagsdaleMy general judgment is that credit cards are very useful for a "new edition" of the electronics or conversational business. Credit card built-in consumer protection, shoes missed or hit by a truck, Visa helps you decide, you get the money back, and the risk is on the business. This is a good trade in new goods and services。

But the stabilization currency is very useful in another scenario. The average turnover on Agent Cash is 1-2 cents. About 600,000 such transactions have been completed. Credit card fixed fee 30 cents. The wire transfers are close to $1. The marginal rate was 2-3 per cent, the majority of which was for transaction costs, and credits were returned. For an electrician, maybe you like points, you like to save credit card miles for a vacation to Miami, 3% from a commercial rate. But when you buy something with only one-two cents, a fragmented API call fee, the stabilizer is a zero-margin fee and a fixed fee is less than one cent。

There is also a key point: immediate settlement. If you purchase goods and services on the Internet, the settlement cycle is at the end of the month, regardless of whether the invoice is wired or a credit card, the business is actually providing credit to the customer or agent. In the proxy world, you usually don't know who the agent is。

Specifically, people who have used Anthropic or ChatGPT API know about the graded system, paying $50 for it, $100 for it, up to $2500. The reason for this system is that they're giving you credit, they don't know you, they don't have KYB and they don't know if you'll pay at the end of the month。

AWS, like Nvidia GPU. The end-of-month settlement was very bad for this type of scenario, with businesses taking full risks. If the customer is not a real company with an enterprise service agreement, but an agent, you have no idea who it is, and you can generate a billion agents overnight, but you can't give the agent letters。

Someone's doing the proxy credit program, and I feel in the wrong direction. Immediate settlement would directly solve the problem. Immediate settlement is like cash. I have, pass it to you, you have it. You provide goods and services, I can't take back the money. The fixed cost of the Asia-America classification, immediate settlement, is a better option for very small amounts and transactions of this nature。

Noah LevineIt is worth refuting that the final pricing of minimum transaction fees and the ability of credit cards to participate in microtrades is decided by Card Networks。

If they want to introduce new types of transactions, such as "micro-trading" — no minimum costs, lower transaction rates — that is entirely possible。

The advantage is that there are far more consumers with credit cards than people who are familiar with stable currencies. So you can keep the developer paying with a card and the back end with a stable currency. But it takes a long time. Until then, it makes sense to consume these agreements directly in primary purses in stable currency。

Sam RagsdaleCredit card companies are going to subvert their 80-year-old core business model, which I think is extremely unlikely. But happy to see it happen。

Eddie LazzarinI agree that there are no strict technical barriers to credit cards. But the problem is more subtle, involving business models and consumer perceptions of credit cards. A few days ago I saw the concept of a proxy credit card, which is essentially an extension of a virtual card. I like my card issuer ' s virtual card functionality, which generates temporary card numbers at any time, and which is closed directly in case of fraud or subscription cancellation difficulties。

Sometimes, however, new platforms or approaches have won, not because of technical necessity, but because they can be tailored to new scenarios. Credit cards are indeed older than the Internet. Credit cards have successfully survived the transition from non-Internet to Internet, and although they have been exhausted, they have survived. So the conclusion is still pending。

Noah Levine: In addition, if Apple Pay becomes a possible technology, agent commerce will also be activated. As to whether this would destabilize Visa or Mastercard, my instinct is that today a lot of B2B transactions are settled by wire transfer between developers and business API. If C.O. can catch this amount, by micro

Original Link