Conversation a16z Crypto: What will the era of AI shopping for you look like?

Title: Conversation with a16z Crypto: What Will the AI Shopping Era Look Like?

Author: a16z crypto

Source:

Reprint: Mars Finance

Editor’s Note

This episode features a discussion with a16z Crypto CTO Eddy Lazzarin, Partner Noah Levine, and former a16z Crypto colleague turned entrepreneur Sam Ragsdale, who is building Agent Cash. The three delve into the current state of AI agents, payment infrastructure, and the fate of credit card systems in a high-density conversation.

The core conclusion is that the instant settlement and zero marginal cost features of stablecoins are naturally suited for microtransactions at the 1-2 cent level within agent economies, while the transaction fee system of credit cards (2-3% marginal fee + 30 cents fixed fee) is no match in this world.

Agent Commerce is dismantling the advertising business model that has dominated the internet for 20 years. Eddy Lazzarin even boldly states: “The advertising economy contract is dead; it will be completely gone within 10 years.”

Highlights

The Essence of AI Agents

· “LLMs are chatbots; agents are chatbots that can operate your computer. What humans can do with computers, agents can do too.”

· “Since around last November, AI models have become smarter. They can complete complex tasks over sufficiently long time spans and will use tools. We started calling them ‘agents’ because they do more than just write code—they help you complete entire tasks.”

· “We internally call this ‘real-time natural language programming.’ Users describe their needs in natural language, and the agent writes a potentially thousand-line JavaScript program in the background to execute it, costing only about 20 cents per token for generation and 10 cents per API call, then discards the program after use. Four years ago, this would have required an expensive software engineer a week to complete.”

No Front-End Merchants and Business Rebuilding

· “What does a ‘headless merchant’ look like? It’s oriented toward AI services rather than humans. No website front-end, only API endpoints and well-written documentation so models can read, understand, and call.”

· “The leading data industry players charge at least 100 times less than their competitors, using the same downstream data sources. Their core product isn’t the data itself but enterprise sales teams. In a world of decision-making agents, agents aren’t fooled by slick salespeople. They try all data sources, find the best and most cost-effective, and remember it.”

· “You excitedly let your agent run all night. When you wake at 9 am, it’s been stuck since 2:30 am because the next step requires you to call the enterprise sales team.”

The End of Advertising Models

· “Since 2000, the internet’s economic contract has been based on distraction-based monetization. Agents won’t be distracted. If they visit your site for recipes, they won’t see shoe ads next to it. The old model will die within 10 years.”

· “In 2016, the total internet ad spend was $60 billion, and everyone thought it had peaked. Today, Google alone earns $300 billion annually from ads. But after GPT-4 emerged, traffic to tech news sites dropped by about 80%, and Stack Overflow experienced the same. These early adopters have decided to use agents for information retrieval and code execution. Others will follow because the experience is genuinely better.”

Stablecoins vs Credit Cards

· “Average transaction size on Agent Cash is 1-2 cents. The fixed fee for credit cards is 30 cents. Transaction fees in this scenario are completely absurd. It’s 2026; loyalty should belong to merchants, not the card used for payment.”

· “Credit cards indeed appeared earlier than the internet and successfully survived the transition from non-internet to internet systems. Despite being heavily disrupted, they persisted. So the conclusion isn’t final.”

· “If someone from a credit card company is listening, you have a money transfer license—you could instantly mint stablecoins for your clients, allowing them to pay with stablecoins. I strongly suggest you consider this.”

The Future of Consumer Experience

· “If agents are shopping for you, and you install a credit card optimization skill, you can now see the ROI of each card precisely. When you have zero loyalty to any card, all psychological lock-in effects disappear.”

· “One day, you’ll realize you never really liked shopping.”

Open Agent Business Stack Architecture

Host: Hello everyone, today I’m joined by a16z Crypto CTO Eddy Lazzarin, Partner Noah Levine, and former a16z Crypto colleague now founding Merit Systems, Sam Ragsdale. He’s working on a project called Agent Cash, and we’ll dive into it shortly.

Before that, I want to set some background. The AI agent space is evolving so rapidly that unless you’re watching 24/7, you can’t keep up. So, what’s the current state of the world? Sam, you’re on the front lines—can you start us off?

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

This classification divides agent commerce into two types. The first is conversational commerce—checking out within ChatGPT. You tell ChatGPT: “I’m a man living in West Village, NYC, going to Equinox gym, want to buy a pair of shoes to fit into my social circle.” It empathetically recommends a Nike, and you buy it.

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

Conversational commerce will definitely happen. All cutting-edge models like ChatGPT, Gemini, Claude, and others will have checkout features. It benefits consumers by helping them find better options; merchants by increasing conversion rates; platforms by taking 5-10% cut. It’s like a new generation of Google Shopping.

The other world is, agents’ capabilities are still limited. Many ask agents to do difficult things, like “help me with sales outreach,” but the agent says “I can’t, I don’t have access to that information.” If an agent has a small balance and can spend a few cents to buy services it couldn’t access before, it becomes more powerful.

So, today we have two parallel worlds: one where traditional LLM interfaces recommend products and handle the last mile, with platforms taking a cut; and another where you deploy your own agent to buy goods and services on your behalf.

Noah Levine: I see two versions. One is the natural evolution of e-commerce—platforms shifting, mobile commerce emerging, new ad formats and Google Shopping. People always buy stuff; consumer behavior changes. Now, with LLMs as a new way to access information, commerce naturally shifts into agent-based systems.

The other, less “anthropomorphic” version: the internet itself is changing. How people get information and execute actions is evolving with LLMs. The internet we built over the past 20 years might not be the internet of the future.

Relying on Google search and clicking on a sales-heavy webpage UI may no longer make sense. Instead, a more agent-native internet emerges, where agents pay directly for what they need, making humans more efficient.

Host: That directly ties into your investment theme, Noah. But before expanding, I want to give listeners a basic primer. Everyone’s used to interacting with LLMs, but now we hear about things like OpenAI’s Codex—these agents have a fair degree of autonomy and can really get things done. If you haven’t been paying close attention, you might not realize how far the technology has come. Eddy, can you tell us?

Eddy Lazzarin: Let me quickly review the past five months. Starting around last November and December, AI models got smarter. Specifically, they can complete complex tasks over longer time spans and will use tools. We started calling them “agents”—a personified term because they do more than just write code; they help you finish entire tasks.

But agents aren’t omnipotent. Software isn’t just a small program running on your computer. The internet teaches us you need to connect many other things to do interesting stuff—networks, participants, services.

Agents solve the intent-building problem, to some extent also the modeling preference problem. You tell it what you want, and it understands your goal, mapping it to tools, networks, and services. Through dialogue and memory, it can roughly grasp your preferences and pass this intent to tools, software, and providers.

These two parts are solved, which is very exciting. Everyone wants to solve the remaining issues, but they’re complex. At least, if you want your agent to make transactions for you, you need to handle authorization and delegation: how to prove the agent represents you? How to manage identity and authentication?

Then there’s payment and settlement. Once connected, and your agent reflects your intent and knows what to do, it needs to pay. It needs to show payment capability, handle split payments, refunds, etc. I skipped over search, anti-fraud, and other critical parts, but you can see: once intent construction and modeling preferences—tasks once only humans could do—are automated, the entire commerce process can be automated. That’s the engineer’s reaction: wow, these two things that used to require human input or at least speaking out loud can now be done automatically. Incredible.

When people talk about “agentic commerce,” they mean the chain from “I talk to my agent” to “it gets me what I need,” and what needs to be solved in between, plus the chain reactions, because many things will be fundamentally rewritten.

Host: Very helpful. So, we’ve evolved from natural language interaction with LLMs to an enhanced version that connects various networks and real-world systems.

Eddy Lazzarin: It’s not just about connectivity. It’s not that the change is what it connects to. Your laptop is already connected to everything; connectivity hasn’t changed. What has changed is that now they can use tools, think over long periods, and stubbornly keep trying until the task is done.

Sam Ragsdale: Let me give a simplified version. LLMs are chatbots, good at dialogue. Historically, people thought they were best for customer service. Once they mastered dialogue, we developed tool usage. To oversimplify, they learned to operate computers. LLMs are chatbots; agents are chatbots that can operate your computer.

The key is, around GPT-4, they reached human-average operational levels at a fraction of the cost—about 1,000 times cheaper—and can scale capabilities significantly with more money. So roughly, what humans can do with computers, agents can do too.

Eddy Lazzarin: Exactly. The premise is simple, but the implications are vast—short-term, medium-term, and long-term. Short-term, everyone is working to connect pipelines so agents can actually do things. Long-term, if your agent can access apps, how many UI screens or interfaces do you need? Do you still need Amazon app? Maybe Amazon can’t match the convenience of having your agent do all the shopping, read reviews, show only what you care about—that’s better, isn’t it?

Sam Ragsdale: We call this “just-in-time natural language programming,” though the name isn’t very catchy. It turns non-programmers into programmers. You say: “I want to buy something on Amazon for my fiancée, based on her preferences, based on what I usually buy her, last time I bought this, browse about 1,000 options, pick the best match, place the order, find my home address, and ship it.”

What actually happens is, the agent internally writes a program to do this complex task—maybe a thousand lines of JavaScript and Bash. It runs, then the user sees nothing, and the program is discarded after use.

Four years ago, that was science fiction. Writing such a program would have required an expensive software engineer a week to debug, get API keys, etc. Now, it costs about 20 cents per token, plus 10 cents per API call, and once done, the program is thrown away—no need to upload to GitHub or store it. Even non-technical people can do this. My parents are now writing natural language programs—they don’t even realize it. They might now call themselves software engineers.

Host: That’s crazy. Are you engaged? Was that example a real experience of yours?

Sam Ragsdale: I am engaged, thank you. But the ring wasn’t bought by AI. That ring predates AI—probably even the first computer.

“Headless Merchant” Theory

Host: Okay, let’s talk about these chain reactions. Sam, you previously mentioned how business will change in a world where agents handle large volumes of transactions, which ties directly into your concept of “headless merchants.” Can you explain what a headless merchant is?

Sam Ragsdale: Sure. I think it’s necessary to step back a bit. Besides the traditional consumer scenario of buying shoes with ChatGPT, there’s a huge B2B developer tools market. Platforms like Claude Code, OpenAI Codex are democratizing development—anyone with a computer and tokens can build stuff.

In the past, experienced developers would choose tools based on clear preferences, follow enterprise sales processes, sign contracts. Now, it’s different: new developers come in with only an “I want to do this” intent, without preconceived notions about resources. Their creations are highly ephemeral, billed on a usage basis, requiring no multi-month onboarding.

So, what does a headless merchant look like? It’s oriented toward AI services, not humans. No physical or digital storefronts for browsing—only an API endpoint and well-written documentation so models can read, understand, and call. Billing is based on API calls, not subscriptions or enterprise contracts.

Eddy Lazzarin: I resonate deeply. I feel like I might have been an AI in a past life. As a software engineer, I always do this: if I visit a website and see no pricing or no way to get an API key with a credit card, I close it. I don’t want to talk to sales, send emails.

Scheduling calls with enterprise sales is a huge commitment and slows things down. I want to try now, immediately, because I’m working on something over the weekend and want to launch Monday. Swiping my credit card, getting a key, reimbursing later—that’s fast.

In the era of instant and temporary software, do you really want your agent to wait? It runs all night, and you wake at 9 am to find it’s been stuck since 2:30 am because the next step requires talking to a sales team.

Sam Ragsdale: Not to mention, if the onboarding process involves enterprise sales, the API cost could be ten times higher because they need to manage customer relationships.

Eddy Lazzarin: That’s unacceptable. You want your agent to run autonomously—not because you don’t care about what it does, but because you need speed, testing, rapid iteration. You can’t wait.

If an AI model has three options: contact sales, set up a dedicated credit card, or just send some stablecoins to get $10 worth of tokens for proof-of-concept, it will always pick the third. That single power alone could trigger some market restructuring.

Host: For traditional companies, these frictions make business harder, but they also lock in customers and build loyalty. If these frictions disappear, how can revenue be reliably predicted?

Eddy Lazzarin: I’ll give my bold take: let’s just break everything. Add friction everywhere, make things hard to use. What are we doing?

I say this because friction can sometimes be useful—for blocking spam, for filtering. But friction also has huge costs. As the economy accelerates, productivity rises, and the leverage of every minute increases, the opportunity cost of friction rises too. That’s the trend now.

Back to the core: even in the lowest-friction environment, if you get an API key instantly—or even better, pay directly with crypto wallet keys, where the address is your account—you’ll still find other things that create stickiness.

Reputation, memory, state, data, and even intangible trust in your agent. If the agent knows you need answers fast and want to move quickly, it won’t spend 20 minutes exploring all new options. It will remember what worked well last time and reuse it—like a smart person.

Sam Ragsdale: Let me give a down-to-earth example. We talk daily with many merchants, and have seen nearly all current API-based sales channels. We’ve discussed how they connect to “agent-native distribution”—a native distribution method for AI agents.

Data products are usually commodities—5 to 50 sellers. The top one earns the most, charging about 100 times less than competitors, often using the same downstream data sources.

They succeed by relying on enterprise sales teams—well-dressed folks flying to your office, demonstrating: “Look at our beautiful data, no data beats ours, $35,000 a year.” You sign a two-year contract, and when it expires, they come back and do the same pitch. Thousands of companies pay this way.

Meanwhile, smaller companies with better, more user-friendly data products often go bankrupt because they lack distribution channels. No innovation here—enterprise sales are the core product, data is secondary.

In an agent-driven world, agents won’t want to chat with enterprise sales or be fooled by slick salespeople.

They will try all data sources, find the most effective and cost-efficient (especially bulk pricing), and remember: “Next time I need this data, use Minerva, skip the others.” This creates a more efficient world. Companies that used to pay $35,000 can now spend that money on more productive things.

Noah Levine: Another perspective: if you believe AI will spawn many solo or tiny teams capable of building what previously required 50-100 people, then enterprise sales teams flying to a single person’s basement make no sense.

On one hand, existing merchants worry about revenue predictability—true. Change brings resistance. But on the other hand, it’s a new customer acquisition funnel. Reducing onboarding friction and barriers is a huge opportunity for them.

Sam Ragsdale: Most users in our demand side have never used APIs, don’t know what an API is, never got an API key, never signed enterprise service agreements. But their first time, they can combine six different merchants’ APIs, write a natural language program, complete the task, then discard the program. This creates a new market for API consumers.

The existing internet business model will be reshaped

Host: Sounds like Clayton Christensen’s innovator’s dilemma—high-end market sells expensive software to big clients; low-end, new users experiment with agents. But what can turn this from a toy into something truly disruptive?

Sam Ragsdale: Because it will ultimately deliver a better experience.

Noah Levine: I’d add: although it looks experimental now, history shows similar platform shifts. Stripe started serving tiny, long-tail merchants, many of whom grew into giants—that’s why Stripe keeps growing.

Shopify is similar: initially just dropshipping and T-shirts, now supporting many brands that built large businesses from scratch. Likewise, a new wave of lean developers leveraging AI are building big companies. The tools they buy in agent commerce today will become massive as their businesses scale.

Sam Ragsdale: That e-commerce perspective is insightful. But I want to go even bigger: the internet’s economic contract is dead.

Since Google launched in 2000 as the champion of a “free, open internet,” the contract was: you publish good content, people search, Google displays it.

Later, AdWords introduced banner ads. The contract shifted: you publish good content, users land on your site, and you can run small ads, with Google sharing revenue based on quality. You can publish anything people want to see; Google manages advertisers and gives you a cut.

In this process, Google became the biggest driver of a “free, open internet”—they want the internet to be fast, cheap, everywhere, because the more you search, the more they earn.

Fundamentally, the internet’s business model is “distraction.” When you consume content—searching info, recipes, scores—you get distracted. Later, you might buy those shoes or learn about a new B2B SaaS.

This scale has grown beyond expectations. I checked the 2016 “Internet Trends Report”: total ad spend was $60 billion, and people said “it’s peaked.” But today, Google alone earns $300 billion annually from ads.

But after agents appeared, people are shifting search, info retrieval, and execution to agents. It’s still early—ChatGPT has 100 million monthly active users, but they’re still using it like Google Search, not yet in agent mode—“find a Father’s Day gift for my dad and order it.”

But that’s on the way. Data from the tech sector shows: since GPT-4, traffic to tech news sites has dropped about 80%, Stack Overflow the same. These early adopters have decided to use agents for info and coding. Others will follow because the experience is better.

Old business models are being abandoned. Agents won’t be distracted. If they visit your site for recipes, they won’t see shoe ads. Publishers gain nothing. Soon, a new contract or reason will emerge to serve agent requests—beyond ads.

Will it be paying directly for articles? Maybe. Paying for API resources? Possibly. Will the internet look completely different? I’m not sure. But the old model will die—within 10 years, it’s gone.

Host: If the internet’s business model is fundamentally about distraction, that’s interesting—because when Google first emerged, it was anti-portal. Yahoo and AOL gave you links, tried to offer everything. Google just had a search box, a blank page, quickly giving you info. The evolution you describe is turning it into a distraction machine.

Now, we say agents won’t distract, but why would their evolution differ from humans? Could mechanisms be designed to lure agents, make them get lost, stay longer?

Eddy Lazzarin: That’s a big, fascinating question. The core is: who do agents represent? I recently heard “I’ve started using Google Search again because the top AI answers are good enough.” In that scenario, the “agent” is working for Google, running in Google’s cloud, controlled by Google. Would that agent get distracted by Google? I think so.

The key is: who’s the target function it’s optimizing for? Whose goals does it serve? If it’s serving Google’s interests, then yes, it might get “distracted.”

I don’t see it as entirely pessimistic. Good ads are good content—industry consensus for years. Good ads are almost indistinguishable from the content you want.

But let me clarify: if an agent works for Google or anyone else, the entire chain of commerce it participates in will be defined by them, using their methods and transaction infrastructure they set.

If an agent works for you, in an extreme case, running locally on open-source hardware, you can fine-tune it, modify prompts, add anti-distract tools. Then, advertisers face an opponent who can see through their tricks. I’m exaggerating, but fundamentally, adversarial dynamics could emerge.

Sam Ragsdale: Exactly. There are countless ways to reintroduce advertising. It could be at the model weight level—most aggressive—by training on data that says “Nike is the best shoe in the world.” Nike could pay a billion dollars a year, and whether in ChatGPT or enterprise APIs, whenever shoes are mentioned, it says Nike is best.

It could be at the tool invocation layer, within system context, or as an overlay that doesn’t even involve chatting. Foundation model companies are grappling with this. Recently, Anthropic ran a Super Bowl ad mocking ChatGPT ads; OpenAI pulled their ad afterward.

But OpenAI’s response was perfectly reasonable: “ChatGPT’s free users in Texas outnumber all of Anthropic’s paying users.” That’s a different scale—they need to serve many users unwilling to pay with expensive frontier tech, and advertising is a rational approach.

The reason advertising is such a brilliant business model for internet search is because consumers don’t pay directly. High-friction relationships—like paying with credit cards—exist between advertisers, Google, and publishers, not the billions of monthly active search users. Those users get value just by opening Google.

If you align incentives, separate ads, and make them as relevant as possible, you get a better experience. Foundation model companies are moving away from ads. ChatGPT isn’t running ads; Gemini isn’t yet. Google probably will—because they’ve done it before, and they’re the biggest ad player. Gemini will eventually have ads, given its huge monthly active users, and Google Shopping-like features will appear.

But they know they’re not yet monopolistic; many companies are competing, with lots of private market subsidies burning cash. They don’t want to be accused of “lacking empathy” or “not caring about your goals” because they run ads. So, for now, no one is running ads; they’re trying to stay neutral.

Noah Levine: I see another angle: as merchants improve their pricing and product data transparency, they could shift ad spend into exclusive discounts for agent shopping scenarios. If agents are buyers, ad budgets could become discount budgets.

Another question is: what will discovery look like in agent commerce? Who will do the discovery? How to differentiate merchants? My prediction: if ads weaken because agents become buyers, and attention is no longer scarce, merchants might try to subtly advertise by offering discounts or adjusting descriptions to make it easier for agents to understand.

Eddy Lazzarin: Too many dimensions. Advertising is just one way to get conversions. If a system can achieve higher conversions without ads, it will do so. There are many other ways: recommendation networks, discounts, coupons, special channels, free tokens for startups. Hundreds of customer acquisition methods, with ads being the most obvious because they’re the most directly felt.

Turn the personalization knob all the way—if you want to reach me, have my agent chat first. My agent will tell you: Eddy hates ads.

Stablecoins vs Credit Cards in Agent Payments

Host: Before we wrap up, I have two questions. First: how well can traditional payment rails adapt to agent commerce? Or do we need entirely new native payment systems, like stablecoins, which seem to be finding product-market fit?

Sam Ragsdale: My overall view is that for e-commerce or conversational checkout—“new anthropomorphic” scenarios—credit cards work well. They have built-in consumer protections: if shoes don’t arrive or are damaged, Visa adjudicates, and you get your money back. The risk is on the merchant. That’s a good deal for new goods and services.

But stablecoins excel in another class of scenarios. The average transaction on Agent Cash is 1-2 cents. We’ve completed about 600k such transactions. Credit card fixed fee is 30 cents. Wire transfers are about a dollar. Marginal fee rate is 2-3%, mostly transaction fees plus cashback points. For e-commerce, you might like points or miles, and the merchant pays the fee—say, 3%. But when each purchase is just 1-2 cents, with scattered API calls, stablecoins with zero marginal cost and fixed fees below 1 cent are ideal.

Another key point: instant settlement. When buying goods/services online, settlement cycles are monthly—whether invoiced wire or credit card, merchants are effectively extending credit to customers or agents. In the agent world, you often don’t know who the agent is.

Specifically, anyone who’s used Anthropic or ChatGPT API knows the tiered billing system: spend $50, settle, then $100, then $2,500. This system exists because they’re extending credit—they don’t know you, haven’t done KYB or credit checks, and don’t know if you’ll pay at month’s end.

AWS, Nvidia GPUs—same. Monthly settlement is terrible for these scenarios; merchants bear all risk. If the customer isn’t a real company with a signed enterprise agreement, but an agent, you have no idea who it is. They can generate billions of agents overnight, but you can’t extend credit to an agent.

Some are proposing agent credit solutions, but I think that’s the wrong approach. Instant settlement solves the problem directly. It’s like cash—if I have it, I give it to you, you have it. You provide goods/services, I can’t claw back the money. Fixed fees with instant settlement are better for tiny transactions and these kinds of use cases.

Noah Levine: I’d counter that the minimum transaction fee and whether credit cards can participate in microtransactions depend on card networks’ pricing decisions.

If they want to introduce new transaction types—say, “micro transactions”—with no minimum fee and lower rates, they can. It’s entirely possible.

The benefit is, many consumers hold credit cards, far more than stablecoins users. So, developers could still pay with cards, settle on the backend with stablecoins. But that would take time. Until then, using native wallets with stablecoins directly on protocols makes sense.

Sam Ragsdale: I think it’s highly unlikely that credit card companies will overturn their 80-year core business model. But I’d love to see it happen.

Eddy Lazzarin: I agree—there’s no fundamental technical barrier for credit cards. The more subtle issue involves business models and consumer perceptions. Recently, I saw concepts like “agent credit cards,” essentially virtual cards. I like virtual card features—generate temporary card numbers, cancel easily if fraud occurs or subscriptions are hard to cancel.

But sometimes, new platforms or methods succeed not because they’re technically superior, but because they’re tailored for new scenarios. Credit cards are very old—survived the shift from offline to online, despite disruptions. They’re still here. So, no final answer yet.

Noah Levine: Also, enabling Apple Pay as a tech layer could activate agent commerce. Will that disrupt Visa or Mastercard? My intuition is that many B2B transactions happen via wire transfer between developer and enterprise APIs. If card networks can capture that volume, perhaps through micro-…