The DoorDash problem arrives for Amazon

Let’s talk about AI and what I’ve been calling the “DoorDash problem.” This is about to define the next battle in AI, and it might completely transform not only how you order a sandwich — but also how the entire internet economy works in general.

So what, exactly, is the DoorDash problem? Briefly, it’s what happens when an AI interface gets between a service provider, like DoorDash, and you, who might send an AI to go order a sandwich from the internet instead of using apps and websites yourself.

That would mean things like user reviews, ads, loyalty programs, upsells, and partnerships would all go away — AI agents don’t care about those things, after all, and DoorDash would just become a commodity provider of sandwiches and lose out on all additional kinds of money you can make when real people open your app or visit your website.

The DoorDash problem is not specific to DoorDash — it’s just the example I like to use because I think sandwich delivery is a funny proxy for the structure of the global economy. But “who owns the customer?” is a big problem for all of the service companies that came up in what you might call the App Store era: Uber, Lyft, Airbnb, Taskrabbit, Zocdoc, you name it.

I’ve been asking the CEOs of these companies about the DoorDash problem on Decoder for months now, because I’ve been predicting that eventually one of them is going to decide it doesn’t want to give up its customers to AI and try to block agents entirely.

Recently, my prediction came true — but it wasn’t a small player that decided to push back against AI; it was one of the biggest players of all. Earlier this month, Amazon sued Perplexity to try and prevent its AI powered Comet browser from shopping on Amazon.com, a move Perplexity has characterized as “bullying.”

So, it’s here: the first major front in the war over who gets to browse the web and who controls the economic experiences of the future has opened up. Now, it’s time for us to really dive deep into the DoorDash problem.

The sandwich delivery complex

Once upon a time, if you wanted to order a sandwich from your favorite local sandwich shop, you’d actually head over there and order at the counter, or make a phone call directly to the store. It was face-to-face interaction with another human being, and that’s basically what the entire economy looked like, outside of some mail order catalogs and QVC.

The famous dot-com bubble of the late 1990s was fueled by the belief that all of these sorts of interactions would happen on the internet — that instead of going in person or ordering over the phone, you’d visit websites to order sandwiches, or buy pet supplies, or have groceries delivered. The bet was that a huge part of the economy would move to the internet, and the companies that got there first would get huge and make everyone rich.

Looking back, this was broadly the right bet. The problem is that “the internet” in the late 90s looked like desktop browsing on huge tower PCs with CRT displays, accessed over 56K dial-up modems. It was the right idea with the wrong form factor, all way too early. People didn’t actually do all their shopping on those machines, and the bubble popped.

But all of these ideas came back around in the smartphone era, because the smartphone was and remains the perfect form factor for ubiquitous access to the internet. Apple launched the iPhone in 2007 and the App Store in 2008, the same year as Android and what would become the Play Store. The app economy started to actually boom.

Companies like Uber, Airbnb and, yes, DoorDash exploded in popularity, and we really have seen huge chunks of the economy move to the internet because of it. Yes, it took enormous sums of venture capital money, and you might say it was wasteful, but that money was spent, and these services are all huge businesses today. They are so huge, in fact, that spending a bunch of VC money to start another Uber now is a really bad idea — there’s a reason you don’t see that happening. You just can’t do this again on smartphones.

But a lot of the same VCs who funded the smartphone app economy are looking for the next big thing, and they’re backing AI — specifically, agentic AI. We’ve all seen the demos and heard the promises. The bet — the bubble — is that we’ll move the economy yet again, from smartphone apps to AI agents that just do all the work for you. And this isn’t some idea from a few wide-eyed startup founders, although there are definitely a few of those in the mix. Apple, Google, Amazon, Microsoft, OpenAI – they’ve all announced plans for assistants or AI-powered web browsers that would work in this way.

Now, here’s where the problem comes in. That promise of the agentic economy is built on an extremely brittle relationship between the DoorDashes of the world and the AI companies who deploy the agents. It’s not at all clear how things will play out to everyone’s benefit. It’s also not clear how anyone can stop it.

Think about it this way: it’s easy to build an agent that can go get you a sandwich if you simply assume that DoorDash exists. That there’s a company building and maintaining a big database of restaurants and menus, another database of available delivery drivers, as well as the ability to take your order, send it to a restaurant, process the transaction, get a delivery person all lined up in time, and handle customer support if something goes wrong.

But if people stop using the apps and websites and start sending agents instead, that business really starts to break down. Because DoorDash and all the other service providers make their money by having a direct relationship with customers they can monetize in lots of different ways. It’s basic stuff like promotions, deals and discounts, ads for other stuff, their own subscriptions like DashPass and Uber One, and whatever other ideas they might have to make money in the future.

But AI doesn’t care about any of that stuff — if you ask for a car to the airport, an AI might just open Uber and Lyft and always pick the cheapest ride. These big App Store era services might just become commodity databases of information competing on price alone, which might not actually be sustainable, even if it might be the future. In fact, this past May at the Google I/O developer conference, Google DeepMind CEO Demis Hassabis said that he thinks we might not need to render web pages at all in an agent-first world:

Demis Hassabis: In the nearer term, the web is going to change quite a lot if you think about an agent-first web. Like does it necessarily need to see renders and things like we do?

I asked Google CEO Sundar Pichai about that comment on Decoder, and he tried to clarify it a little:

Nilay Patel: I’ve been thinking about this in the context of services like Uber, DoorDash, and Airbnb. Why would they want to participate in that and be abstracted away for agents to use the services they’ve spent a lot of time and money building?

Sundar Pichai: Two things, though. First, there’s a lot to unpack, a fascinating topic. The web is a series of databases, etc. We build a UI on top of it for all of us to conceive.

This is exactly what I wanted — “the web is a series of databases.”

It is. But I think I listened to the Demis and Sergey conversation yesterday. I enjoyed it. I think he’s saying for an agent-first web, for a web that is interacting with agents, you would think about how to make that process more efficient.

I’m just saying, turning all these services into databases for agents might kill those services. And then what will the agents do? That’s the DoorDash problem.

So far, this hasn’t really been a problem. All these AI bots haven’t been disrupting the flow of commerce between consumers and the companies, so service providers have been content to let the AI companies experiment. In some cases, they’ve even announced integrations with AI chatbots, because why not? But as more and more people turn to AI first, and as more and more AI companies start making noise about agents, it’s clear that this is going to be a real fight in the years to come. The thing I didn’t expect is that it’s starting with Amazon.

The great AI browser fight is here

If you haven’t used Perplexity, the company arrived on the AI scene in late 2022, right after the launch of ChatGPT. It makes an AI search engine powered by a combination of its own models and existing AI models from OpenAI and others.

The original pitch was as a kind of competitor to Google. As of earlier this year, Perplexity now also makes its own browser called Comet, which is built on Google’s Chromium engine but is really a vehicle for the company’s AI agent technology.

Perplexity promotes Comet as “the browser that works for you,” and much like Microsoft Edge with built in Copilot, Chrome with its deep Google Gemini integrations, and now OpenAI’s Atlas browser, it’s an attempt to rethink how we use the web, or rather how we direct AI to use the web on our behalf.

Because people do so much shopping on the web, Perplexity built a shopping agent that can automate browsing and buying on Amazon, which, notably, is in violation of Amazon’s terms of service. According to Amazon’s legal complaint, Perplexity ignored several requests to disable this functionality, and instead began masking its AI agents as human Chrome users and even updated Comet to circumvent Amazon’s technical efforts to stop it.

Amazon’s lawsuit claims Perplexity is violating the Computer Fraud and Abuse Act, which is a controversial law but still basically the big federal anti-hacking legislation. This is a big deal. Here’s Amazon’s statement:

We think it’s fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions whether or not to participate. This helps ensure a positive customer experience and it is how others operate, including food delivery apps and the restaurants they take orders for, delivery service apps and the stores they shop from, and online travel agencies and the airlines they book tickets with for customers.

Agentic third-party applications such as Perplexity’s Comet have the same obligations, and we’ve repeatedly requested that Perplexity remove Amazon from the Comet experience, particularly in light of the significantly degraded shopping and customer service experience it provides.

That’s Amazon making direct comparisons to food delivery apps and online travel services in the context of AI agents. Baby, that’s the DoorDash problem.

Now, I’ve asked a lot of CEOs in the past year about this exact problem, and the wild thing is not a single executive I’ve talked to seems nearly as concerned as Amazon. They’re certainly not filing lawsuits. In fact, some of them seem downright relaxed.

For example, here’s what Lyft CEO David Risher had to say when I asked him about the DoorDash problem:

David Risher: The most extreme version of what you’re saying is I go to ChatGPT and I say, “Please come pick me up.” And some rando comes to pick me up, and there’s no guarantee, there’s no service. That would be bad. I don’t think a lot of people would be super excited about just some rando coming, picking me up in an unbranded service, and whatever it is.

So if it’s not going to be an unbranded rando picking me up, then it probably has to be one of the guys who are doing existing rideshare, and that’s us. And then we’ve got all sorts of ways where I think we can compete.

If they try to disintermediate, you say, “Well, no, I actually have a preference here.” And we’re going to do a whole bunch of different things to make sure that you have a very, very strong preference for asking for us by name, not just saying, “I want to get to a place.”

I’d say this is the most common response to my questions about the DoorDash problem — that people have a lot of existing strong brand relationships, and that you’re still more likely to want your AI to get a car from Uber or Lyft instead of some random company. Maybe these service providers just charge higher fees to customers that come from AI, and it all works itself out. Besides, you still have to build the service — connect supply and demand, manage drivers, fielding customer service requests — and that’s really hard, so it’s unlikely anyone else does that in a purely AI-native way.

Actually, the best version of that argument I heard came from Zocdoc CEO Oliver Kharraz, who more or less said “go ahead and try”:

Oliver Kharraz: Here are the questions you should ask yourself. Question number one: Are these agents simply going to completely displace you? Anyone who’s running a business that interacts with the real world knows that that’s not going to be the case, because of that learning curve, because of all the edge cases, and all these things. Even if the AIs were to start learning about them, we’re so much further ahead that we can always deliver a better experience. So this is the coast of England problem. Our cartographers have been at this for 20 years; there’s no way that anyone would catch up to us anytime soon. So they’re not going to put us out of business.

Nilay Patel: I look at some of these companies, and they say, “Well, screw it. We’re just going to go click around on your website. We’re just going to open a browser, and we’re going to click the buttons for the user, and we’ll do that in the background.” And you might never know. You might never know that this happened.

Perplexity is going to do this with its browser. Knowing Perplexity, that is probably how its agent will work. That destroys your leverage. You have to detect their agent and say, “You can’t do automated browsing.” And there’s no framework. There’s no negotiation framework for that.

While they do that, they’re not making any money, and I make money as I used to. So that’s actually cool. Give me free traffic.

So I don’t think there’s really a threat there. And if they are going to negotiate, if they do want to have some of that money, I think these companies that are the Ubers, the Airbnbs, the DoorDashes of this world, are in a unique position to dictate their terms in a way that they could never do with Google.

I got the same kind of response from Taskrabbit CEO Ania Smith, who more or less dared Apple to start a Taskrabbit of their own:

Ania Smith: Then, there will be some economic way to figure out a system where it works for both parties. Siri, in this case, wants to be able to provide these types of services, and they can’t really do it without Taskrabbit because only Taskrabbit actually has a network of thousands of Taskers. We cultivated that network. We know who they are and we understand their skills. They’ve all been background checked.

Siri, Apple, or whoever is not going to do that. They’re just not going to be able to do that. We will always have the strength and supply, and we will continue to go where the customer is to make sure as many of them have access to our platform as possible.

Oliver and Ania’s perspectives here are important, because I think they represent a fairly defiant attitude toward the agentic web. ZocDoc has spent more than almost 20 years building a platform to help find you a doctor, and doing the hard work of integrating with the chaotic U.S. healthcare system. Taskrabbit is really good at getting someone out to hang up a TV. Those moats are so broad and deep that these CEOs don’t really feel like they have any competition, let alone from AI.

At the same time, when I asked Uber CEO Dara Khosrowshahi about the DoorDash problem, he was pretty open that it is something he thinks about:

Dara Khosrowshahi: It’s a really good question. We talk about it all the time, and time will tell as to what the right decision is. I believe in running open platforms. I don’t believe in companies that try to fight the course that technology is taking. They always get left behind. I think we should go where consumers want us to go. I do think that’s all nicey-nicey, but at the same time, one of the advantages that we have is our unique inventory.

So I think, as it relates to AI and these agents, we want to work with these players. We come from a place of strength because of the unique inventory and the fragmentation of these markets that we’re organizing. We’re going to build our own agents as well. As long as we’re thinking about the consumer, our earners, and our merchants, I think we’ll be okay.

And Dara was also one of the few CEOs to give me a straight-up answer when I asked how much he might consider charging AI companies if they try to sit between him and his customers.

Nilay Patel: So, I come to you and I say, “I want to be able to get cars from Uber.” What is the percentage toll? What is the extra margin you would have to charge me, in dollars and cents, to make it worth it for me to take the customer away from you in that way? Because when the customer opens your app, you get to cross-sell them into Uber One, or ask “would you like some food when you arrive?” There’s all of these other incremental opportunities you have to forgo if I take that customer.

Dara Khosrowshahi: So I have a weird philosophy on this. Initially, I charge you zero. I think that companies sometimes–

Alarm bells just went off at every level.

No, listen! People spend so much time trying to figure out what the economics might be when the first thing is to try it out. Is it going to be a good experience or not? Is your scheduling actually going to work, or is it going to be off and the driver has to wait for 10 minutes, which is terrible? Let’s just figure it out. Then, once you optimize the experience, we can measure. Are you an incremental consumer for Uber or are you totally cannibalistic?

If it’s cannibalistic, then I’m going to charge a lot of money. You can’t have any money because you’re getting the benefit and my content. You’re not bringing me any business at all. If it is incremental, then I would pay some take rate. Is it a 5, 10, or 20 percent take rate? It depends on the incrementality.

But I think so much innovation has slowed down because companies try to figure out the economics first. Figure out the experience first and then the economics. Listen, if I do a bad deal for a year, who cares? I’m going to renegotiate with you. I’m building stuff for the next 10 years. Success or failure isn’t going to be determined by my take rate being 5 or 20 percent in year one. It can set precedent, and precedents are dangerous. That’s why I would say to charge zero. Let’s try it out. Let’s see what the experience is. Let’s try to measure out what the value add is, and then the economics essentially will take care of themselves.

If you’re making the mental comparison to the App Store era like I’ve been, that might sound familiar: build the business first, get all the customers, and then figure out what to charge in order to make a profit. That’s always been the Uber way, and it’s not really a surprise to hear it again in this context.

So at this point, you might be wondering: what does DoorDash think about the DoorDash problem? We’ve been trying to get DoorDash CEO Tony Xu on the show to go deeper on all of this, but in the meantime Head of Communications Ali Musa did send us a statement:

We have launched several integrations with partners who offer DoorDash incremental demand channels, such as ChatGPT, Google, and Yelp. No single channel monopolizes customer attention. What makes DoorDash valuable is not where you place an order, but the end-to-end experience, including reviews, personalization, real-world fulfillment, and support. A channel would have to replicate all of our end-to-end functionality to fully serve a customer. Each time we’ve integrated with partner properties, we’ve continued to grow faster and at larger scale.

Tony, if you’re listening, I’d love to have you on the show to unpack that a little more. I did get a more fleshed out version of that idea from Airbnb CEO Brian Chesky, one of my favorite Decoder guests, who told me that he just doesn’t think a single company can build all of the interfaces for everything:

Brian Chesky: There’s this AI maximalist view that there’s going to be like one or two AI models and one or two applications that rule them all and you use this one app and this one model for everything in the world. If you take that to its logical conclusion, you also start to go to this place where almost one company rules everything, and I think there’s numerous problems with the AI maximalist view that it’s one company to rule them all.

I don’t think companies just want to be data layers, and so these platforms or these new interfaces are only as good as the companies that participate, and the companies will only participate if they can have a relationship with their own customer.

Again, I’ve been asking about this problem for months, and hearing all of these very pragmatic answers. Sundar, when I asked about him about it, made a very down-to-earth comparison to brands and retailers and credit cards, all of whom take cuts from every transaction:

Sundar Pichai: It’s tough to predict, but I do think that over time, if you’re removing friction and improving user experience, it’s tough to bet against those in the long run. And so I think if you are lowering friction for it and then people are enjoying using it, somebody’s going to want to participate in it and grow their business. And would brands want to be in retailers? Why don’t they sell directly today? Why won’t they do that?

Because retailers provide value in the middle. And why do merchants take credit cards? I’m just saying. So there are many parts, and you find equilibrium because merchants take credit cards so they see more business as part of taking credit cards than not, which justifies the increased cost of taking credit cards.

Similarly, Microsoft CTO Kevin Scott told me here on Decoder that all their work on agents won’t come to anything if they don’t make the economics work:

Kevin Scott: So if you have a business and you want your business to be able to transact with users via their agent, that has to make good business sense in order for you to be willing for that to happen at all. You can’t just hack your way around that and expect it to be a durable thing. Even if you can temporarily figure out some kind of technical magic to get around the brittleness of the actual technology, you also have to get rid of the brittleness in the business model.

So here you have all of these smaller players saying they want to see how things play out, and even saying they think they have a lot of leverage. You have Google and Microsoft saying it needs to all make business sense, with Google’s CEO making a direct comparison to retail.

So why is Amazon, a literal retail giant, going on the offensive against AI agents first, and hardest?

We were just discussing how all the CEOs I’ve talked to on Decoder have framed the DoorDash problem for me — it’s mostly been very calm, and even confident, even though I’ve been predicting that eventually someone would push back.

But I was completely surprised that it was Amazon that pushed back first, and against Perplexity of all companies. Perplexity isn’t exactly a household name, and Comet isn’t grabbing all the market share from Chrome anytime soon.

At first I thought Amazon was just picking on a smaller player to make a point and set some precedent. But the more I thought about it, the more it occurred to me that Amazon might have the most to lose in the world of AI agents — it has the biggest ad business, the most lucrative subscription service in Prime, and it is a force-of-habit default for millions of shoppers worldwide. Letting an AI agent disrupt all of that would be dangerously costly. And on top of all that, Amazon might be far less differentiated and easier to replace than something like Uber.

Hear me out. First of all, ads are a huge business for Amazon. Any time you shop with Amazon, you see sponsored products, sponsored brands, and regular old display ads. Lots of companies spend millions in marketing budgets buying preferred placements across Amazon, and it’s become a meaningful part of its business and the entire advertising ecosystem.

In fact, Amazon reported $17.7 billion in advertising revenue in its most recent quarter, a 24 percent increase year over year that puts it on track to pull in more than $60 billion in ad revenue in 2025 alone. That makes Amazon’s ad business the third-biggest among tech giants, behind only Google and Meta, and it’s also now diving head first into AI advertising, with tools for brands to generate images, product descriptions, and even video ad spots. Amazon’s ad business is so robust that it’s now selling all of the technology that powers it to other retailers. For Amazon, this is a huge, profitable business, one that’s growing to rival AWS.

But if an AI agent does your shopping for you, you don’t see the ads, and you certainly won’t care if some brand paid to put their product first on the list. So browsers like Comet with agentic shopping features are an existential threat to Amazon’s huge and growing advertising business.

AI browsers are also a threat to Prime itself — subscribing to Prime is a major reason people shop on Amazon by default. But if you can just tell your browser to buy something at the lowest price from any retailer, maybe you shop on Amazon less, and maybe that Prime subscription feels less worth it over time. All of the other companies we talked to talked about how you trust their brands to do things in the physical world like sending you a car, but Amazon’s already turned itself into commodity provider of weirdo drop-shipped products from alphabet soup companies. It wouldn’t take a lot to disrupt that if people are just sending agents out to do the shopping for them.

This is a problem Amazon is already trying to solve on its own platform — it has its own AI shopping assistant, called Rufus, that can navigate the huge number of products and vendors on the site, and it’s trialing an AI bot called “Buy For Me” that can go out and make purchases for you on third-party websites.

Maybe you don’t trust some fly-by-night company to book a doctor or hang a TV or send you a car in time to catch your flight. But you might be perfectly happy to just take the cheapest price on a USB-C cable from anyone offering it, which would make a lot of Amazon’s retail competitors very happy.

Amazon simply can’t risk losing that relationship with its customers or its status as the default shopping destination. If Amazon does partner with agentic AI companies, it’s only going to do so on favorable terms it negotiates directly. We know this because Amazon CEO Andy Jassy said so on the company’s most recent earnings call. He said that the agentic web reminded him of the early days of search engines as “sources of discovery for commerce,” adding, “you had to kind of figure out the right way to work together.”

It’s weird to think of Amazon as having one of the least defensible moats in tech, but AI scrambles everything, and it feels like Amazon has a lot at stake if the courts ultimately side with Perplexity.

“Software is becoming labor”

That all brings us to Perplexity, and why it thinks it can win in a legal battle with Amazon without one of those deals. Perplexity’s point of view isn’t exactly subtle — we can just look at the company’s own blog post, titled “Bullying Is Not Innovation.” As you might imagine, it makes some pretty bold claims about how the future of the internet should work.

First off, Perplexity has a history here that’s important to mention. The company first began making a name for itself a few years ago, when it launched its AI-powered search engine in December 2022. Instead of partnering with news organizations, Perplexity began scraping websites for the information it would compile into AI summaries for its users.

In some cases, newsrooms were finding paywalled content surfacing in Perplexity search results, which eventually resulted in the company getting sued by Dow Jones, publisher of The Wall Street Journal and The New York Post. And the lawsuits haven’t let up. In September, Perplexity was sued by Encyclopedia Britannica and Merriam-Webster, and a month later, it was sued by Reddit.

Since then, like OpenAI, Perplexity has started going out and striking deals with news organizations to license their content. [Disclosure: The Verge’s parent company, Vox Media, has struck one of those deals with OpenAI.] But overall, Perplexity is very much an “ask forgiveness, not permission” player in the AI ecosystem, and the company certainly loves getting in trouble for attention.

All that said, openly violating Amazon’s term of service is a pretty risky stunt. But the company isn’t backing down. In its blog post, the company casts itself as an innovator fighting for the user and striving to create a better internet, and Amazon as a bully trying to squash progress with legal threats. It called Amazon’s lawsuit a “threat to all internet users” and an attempt to “make life worse for people.”

The company writes, “User agents are exactly that: agents of the user. They’re distinct from crawlers, scrapers, or bots. A user agent is your AI assistant — it has exactly the same permissions you have, works only at your specific request, and acts solely on your behalf.”

The argument here is that an AI agent is the same as you — if you can access a website, your agent should be able to as well. If you’ve been at all thinking about how AI will interact with labor, well, here it is — Perplexity says explicitly that “software is becoming labor” to justify its agents accessing Amazon’s site in violation of Amazon’s terms of service.

To be honest, I have no idea how to feel about that. I couldn’t even tell you how to begin to feel about that. In fact, I invite you to tell me how to feel about that: our email address is [email protected], and we really do read all the emails.

In any event, Perplexity doesn’t want permission for its agents to access websites, and it doesn’t think it needs it. Its blog post ends pretty defiantly: “Amazon also forgets how it got so big. Users love it. They want good products, at a low price, delivered fast. Agentic shopping is the natural evolution of this promise, and people already demand it. Perplexity demands the right to offer it.”

Look, I don’t know how this fight is going to end up. I don’t know how the industry solves the DoorDash problem. I don’t even know if all these AI agents will work well enough to make this problem any bigger than Perplexity starting some shit.

What I do know is that everyone is thinking about it, and if all this AI investment is going to be worth it, the companies building agents need to start thinking about how all the people that actually do the real labor these services provide get paid real money that makes those jobs worthwhile. Because the thing that really brings down DoorDash is not having anyone to deliver sandwiches at all.

_{Questions or comments about this episode? Hit us up at [email protected]. We really do read every email!}