（预览）后AI时代的互联网现实、未来创作者如何成功、创业公司邮件、F1版权与另类摇滚

Hello, and welcome to a free preview of Sharp Tech. Hello, and welcome back to another episode of Sharp Tech. I'm Andrew Sharp, and on the other line, Ben Thompson. Ben, how are you doing?

Doing okay, Andrew. Doing okay. I'm back here in Taiwan, addressing some things, taking care of some things, as it were. It's summer. It's very hot. I'm reminded why I'm usually not here in the summer. But other than that, I'm all good.

Mm.

There you go. Well, that's good. I'm in a good mood. It's summer here as well. I'm still in New England. I'm on Nantucket. I've been in the sun all day, so I might get a little punchy on this one.

But with a week between shows, I'm always very excited to see you on the other line here. So we'll dive into it.

Right. In contrast to twice a week, when you're just getting pretty sick of me. It's a—

Exactly.

Common refrain.

It's a real grind—

Yep.

Seeing you every—

Yep.

72 hours.

I get it.

Oh, my God. I don't know how I do it all year long. But a good—

That's what the money's for, as D'Andre Fletcher would say.

Midsummer rundown. Indeed, that is what the money's for.

We're going to be bouncing all over the place. However, we will begin with some responses to your article on Monday. The title was “Content and Community,” and you were riffing on the implications of what we discussed at the end of last week's show, which was Cloudflare's plan to block AI crawlers on Cloudflare-protected websites on the one hand.

On the other hand, Cloudflare will give AI companies and Cloudflare-protected websites the opportunity to use a pay-per-crawl system, which would collect a fee from AI companies every time they crawl a website, and then pay most of that money to the website owner.

So, regarding that idea and a new framework for the post-AI internet, Finn wrote in and said, “Salutations. Cloudflare's push to have AI providers pay for content crawls makes no sense. None. A second Negroni makes me feel generous, so I'll share with you why that is.

“First, we can factor out the use case of training, as that is a single request to pull a piece of content once, and that's that. Maybe they put in another request in 6 months for Llama 3.3-500B_26-04. Crawls during inference—”

Redis cache.

Look, I was completely underwater with that entire sentence. “A Redis cache and crawl pages at some cost-efficient minimum. Ben posts on a schedule. I know when he'll post. Congrats, Ben. You made 4 cents last week. Given the size of the holes in your analysis here, maybe that's fair.

“Cloudflare's marketing speak and some motivated reasoning had you confuse a local optimum with a default.”

Oh, boy. Finn is bringing the heat. What do you have to say for yourself? What do you think?

Yeah, Finn needs a few more Negronis, because he needs to relax. It would be my initial response here.

Zooming out, a challenge about writing about this particular topic—and remember, this goes back to me writing about the agentic web, and why ads were great for the human web, and we're going to need something different as we go to AI—is that it's all very theoretical, pie in the sky.

There's a huge “How do we get from here to there?” sort of aspect to it, and it's really light on details, which I sort of admitted at the time. But it also makes me hesitant to even write about this stuff, because you get Finn with a Negroni coming along saying, “Bad, bad, bad, bad,” picking apart all the details—X, Y, Z.

And can I also just add, for the sake of the listeners who are hearing the email and not reading the email, I do need to note for the record that the entire email was written in Sam Altman's house style: all lowercase letters.

Oh, man. Terrible.

So we have to deduct about 50 points from the grade right off the top. But again, I do appreciate the energy from Finn there.

No, I mean, in Finn's defense, I agree. This overall structure of the Cloudflare proposal is not going to scale, is not going to get somewhere. So when I talk about it, it's more about this: directionally, there's something here.

Mm-hmm.

And by the way, to get there, you're going to need someone exerting power to make it happen.

Yeah.

That's my broader overall take on the Cloudflare piece.

Now, the reality is—and this is the problem with the internet in general—you can go back to regular websites versus Google. Google needs the web, but not any one website.

Right.

Any one website, they all need Google. So you have this real mismatch in power and need, and this applies to the copyright cases. The problem with all these plaintiffs bringing the case about the AI models—and both judges had to admit this, including Judge Chhabria, who wanted to rule for the plaintiffs—is that they need a lot of content; they don't necessarily need your content.

Mm-hmm.

So what's the actual case that you have here? And that's just a reality writ large.

Overcoming that by saying, “Oh, I'm going to…” You know, the interest of any one publisher—you can't rely on this system. And I think a problem with Cloudflare's proposal is the fact that it's sort of geared to existing systems.

Yeah.

Now, broadly speaking, are we going to need content generation for AI? I think so. I don't think that, if they're just going to pay for content generators all along, that's going to scale sufficiently.

Mm-hmm.

Maybe it will. Maybe the argument is that you actually don't need that many facts and that's going to be sufficient. And by the way, that might be true. I might be wrong. They might have gotten enough to be sufficiently smart, and now they just need one wire service to keep them updated on current events, and they can sort of rely on that.

Synthetic data, or, yeah.

Yeah. Well, one wire service to keep them updated on current events, and they can sort of rely on that. But if you do think that they need significantly more content, I think the answer is some sort of market for content, and that spurs people to come in and fill that need.

Mm-hmm.

However, this was the part I was getting at in the article, and I think we talked about it on this podcast. That's why I brought up the Facebook analogy.

Yeah.

What makes these the true advertising businesses—which, there are really 3, I would say—is that they're Google, Facebook, and Amazon. Actually, Amazon's maybe the best example here. Amazon has built a multibillion-dollar advertising business, and all the advertisers are sellers on Amazon.

Mm-hmm.

And they literally advertise on Amazon.com. It's completely self-contained. These entities create their own markets. They create their own customers. This is why Meta is impervious to a CPG boycott that they tried in 2020, because all the Meta advertisers that matter are entities that only exist because of Meta.

Right.

That's why when COVID first struck, Meta's revenue barely got hit at all, whereas everyone just assumed, “Oh, a recession's coming. All advertisers are going to get pulled.”

If you're an entity that exists because of Facebook, and all your business is based on performance marketing, where you pay a price to acquire a customer, you literally need—

You're not boycotting Facebook.

—to advertise or die. You know, you're going to die, right?

And the point is that they pulled these new business models into existence. These aren't the Unilevers or the legacy brands that have dominated advertising in the past. It's an entirely new advertising market.

That's right.

Now, Google—

It's kind of unique.

Google's kind of unique because the whole market was there waiting for Google, which was this open web: disparate, unable to find anything, too much stuff. Google comes in and just captures it, and that's part of why Google's such a dysfunctional company.

Because their whole life has been easy. They played on easy mode. They created a brilliant product, and they've just been harvesting ever since, which, by the way, Google's business model is kind of harvesting, right?

I've talked about this on here. I compare it to Uber and Airbnb, right? Airbnb somehow got all this amazing press for years and years, and everyone dumped on Uber. It was always confusing to me, because I feel the externalities of Uber are very positive: drunk drivers off the road, more commerce, people getting around. You're just expanding transportation.

Mm-hmm.

Whereas Airbnb, it's like, does anyone actually want an Airbnb next door to them?

Right.

The externalities are, I think, much more negative than an Uber. And I think there's a similar thing when it comes to advertising. Google somehow doesn't get all the opprobrium that Facebook does, despite the fact that a huge amount of Google's revenue is scavenging off links that would've been clicked anyway.

Mm-hmm.

When you search for something and click the sponsored link, and the real link is right below it, what service did Google provide?

Absolutely.

That's also a fair criticism of Amazon and its massive ad business these days.

All search engine advertising is a little suspect.

Yeah.

There's value, but by and large, it's so easy because the user is telling you what they want that it's kind of just a tax in many respects. Whereas the reason I've always defended Facebook's advertising is that it's actually introducing you to new products that you didn't know existed, which I think is a much more positive manifestation of advertising and why it's a good thing.

So, we're sort of sidetracked here from the point.

You did, as is your wont. But before you take it full circle and go back to pay-per-crawl, I just want to say that reading your article on Monday was gratifying, because on the last show, when we were discussing this, I was pretty bearish on the idea that pay-per-crawl was a model that could somehow revive the economics of the open internet and ad-supported websites. That was some of the language being used by Matthew Prince in the blog post announcing this. It was like, “We're the saviors. This is a life preserver for people who are being left behind by the AI internet.”

The reason I was bearish is that I think it's worth being clear about some of the headwinds those folks are facing in the legacy internet. Google is sending less traffic than ever to external websites.

Yep.

The distribution mechanism is dying. The ad model has been dying for 15 years at this point, and it's in worse shape than ever. That's also compounded by diminished audiences for all these websites.

Compounding things again is the idea that, in the future, LLMs will be the portal to the internet for most people, and most people won't visit websites at all, which is already happening. So I don't know how futuristic that take actually is.

And then the main thing, on top of all that, with respect to media companies, is that the cost structure just doesn't make sense. The cost structure of good, ambitious media companies that do the sort of work people want to support and preserve in the years to come hasn't really made sense for a long time. So it would be awesome if the AI licensing market were big enough to solve all those problems and make it 1997 again for some of these people, but I'm not going to hold my breath on that becoming any sort of reality anytime soon.

Yeah, I don't think it's going to happen at all.

Yeah.

The reality is, if you make this market for AI-generated content, the people who are actually going to win this market—and this is sort of the full-circle point—

Exactly.

—are going to be new entities that are created specifically because this market was created.

And that's what you wrote on Monday, where I was like, “Aha, now I see why Ben was taking a wait-and-see approach with what might be possible for this sort of framework over the next 10 or 15 years in terms of—

Right. I think the analogy—

—yielding success.”

The analogy I would use is ghost kitchens for content.

Mm-hmm.

A ghost kitchen is this idea that DoorDash and Uber Eats have sort of created this new market for food delivery. Now, most of the suppliers in that market are existing restaurants, which makes sense. They say, “Oh, a new revenue stream. We can add this on,” and so on. Let's ignore how this has destroyed the service level within the actual restaurant and all those sorts of things that you and I, as old fogies, complain about.

Yeah.

But there's also a new opportunity, which is to create virtual restaurants. Literally, you're completely structured around food delivery. It's a store that doesn't exist in the real world; it only exists in the apps. Then your cost structure is optimized for it, along with your workflows and all those sorts of things.

In many respects, the virtual restaurants actually have much better food because the food is structured for delivery. It assumes it's going to be 30 minutes until it's actually consumed. All those bits and pieces matter.

I could see ghost kitchens for content. If there's an incentive structure created to generate content and there's some sort of market for it, I think that's what would actually win. It would be entities that exist only because this market structure was created, just like Meta advertisers only exist because Meta advertising exists.

Yeah.

I think there's a—

Google, again, is kind of the exception. They captured the web. The web was just sitting there waiting to be captured. But the more normal state of affairs is that these entities create their own suppliers.

Apple created its own app suppliers by virtue of the App Store existing.

Yeah.

Did existing app developers go to the iPhone? Yes, but that's a very small portion. A huge majority came in to fill the market.

There's an eternal chicken-and-egg question about what comes first. I've thought about this a lot and written about it a lot in the context of app stores. When I was at Microsoft working on the Windows 8 App Store, this was something we thought a lot about. What comes first? Do you get developers first, or do you get the market?

It's absolutely the market that comes first. Apple dominates the App Store because it created the phone that people wanted. People were there, and then Apple created the mechanism for developers to come in and serve that market.

That is absolutely the direction in which it happens, and I think that is a principle that extends broadly.

Okay. One question before we shift to the future of content makers. First of all, we should note that this vision for the future is a lot less rosy than the Content Independence Day language from Cloudflare.

Yeah, to be clear, I’m with Finn. Finn was being a little—what's the word when you're talking a lot of smack?

Dickish? Um—

Yeah.

That would be one word.

No, no, no. I mean a nice word.

But amusingly dickish. I appreciate it. It's fine.

I don't see the Cloudflare mechanism really going anywhere. To the extent that it does, it's almost more like brand marketing. The big guys could sign up for this deal, like The New York Times or whatever it might be. But those are also the folks who need the least help.

Yeah.

Is this going to save your—

Exactly.

What we're envisioning is people who are basically coming to this market to write exclusively—not necessarily exclusively, but primarily—for AI crawlers, which isn't—

Oh, I think exclusively. Why go through the trouble of having a website and all those sorts of things? Literally, just create content for—

And these aren't going to be amazing jobs or great jobs. There's a reason why the Scale AIs of the world are outsourcing labor to the Philippines or wherever it might be.

Yeah.

Let's be clear about what the reality here is. The barbell effect comes for everyone.

Mm-hmm.

There's the prestige side, where they have direct consumers that go to them directly, and on the other side are the mines, the salt mines of content.

Right.

You're getting paid, but it's not amazing work or a great job. It's a commodity. It's a total commodity.

Well—

That's the thing: the LLMs don't even want content. They want tokens. It's total commodification.

So lowercase Finn here was voicing skepticism that was shared by a few other listeners.

Yeah.

Which I think is totally reasonable, just to be clear.

But in terms of how big this market could be, how many times would AI models actually crawl if they had to pay? Couldn't they just index a page—

Oh, to be clear, that's what they do.

And come back?

Index. They build indexes.

Yeah.

Yeah. All that is correct. I don’t see this as ever being any sort of meaningful revenue stream. I’m not going to turn on Cloudflare. I don’t use Cloudflare, but I would not turn it on anyway.

Mm-hmm.

I think that’s a fantasy.

Yeah.

Yeah.

I think that’s a fantasy.

Okay. To keep it moving, Paul says, speaking of publishers: “Hey guys, I really enjoyed your conversation about Cloudflare last episode, and I wanted to run an idea by you. Do you think the current online publishing transition from Google Search to an AI free-for-all, and now potentially to pay per crawl, is comparable to the music industry’s transition in the 2000s? They went from CDs to pirated music to streaming, and today’s music industry is much less lucrative than it was in the days of CDs. So do online publishers just need to accept this new reality and start figuring out their equivalent of an Eras Tour in order to make money for themselves?”

Ben, what do you think?

I think it is sort of an interesting analogy. To be clear, the Eras Tour is a singular sort of thing. That’s for Taylor Swift.

Yeah.

But I think it’s an important distinction. You can’t look at this as a whole. There’s going to be this massive bifurcation. The New York Times is like Taylor Swift in this analogy: a super-successful publishing model. I wrote years ago that we can’t always use The New York Times as our stand-in for publishing, because they’re exceptional.

Mm-hmm.

They’ve actually figured out the internet in a way that, if anything, has harmed all the other publishers. They scarf up all the best talent. Everyone who has a subscription just subscribes to The New York Times instead of subscribing to someone else.

Yeah.

Everyone’s competing to be people’s second subscription, and that’s not a very good place to be.

And also, we should be clear and explicit: they solved their problems with subscriptions and not advertising, and it took other publications longer to get there and longer to see the light there.

Yes. And a fully integrated approach, which is that our editorial mission is going to be defined by being worth a subscription.

Yeah.

This goes back to very explicit memos. I wrote about this; I think I did at the time. They did this sort of exposé on the working conditions at Amazon—not just the warehouse workers, but those poor office workers. They work so hard, blah, blah, blah.

Mm-hmm.

It was very controversial. People in tech were upset about it. I defended The New York Times. I said, “Number one, it does kind of suck to work at Amazon.”

I was going to say, from what I’ve heard, that’s not wrong.

I mean, it sucks so bad to work at Amazon that it sort of single-handedly saved Microsoft, because Microsoft was up there in Seattle not doing very well. But no one wanted to move. If they were in Silicon Valley, Microsoft would’ve lost—

Less attrition than there might have been.

Right.

That’s great, yeah.

Microsoft would’ve lost all their talent. If you’re up in Seattle, it’s like, “Do you want to work for Amazon?” “Okay, I guess I’ll soldier on with Office or Windows or whatever it might be.”

Yeah.

Which is why all the tech companies opened offices in Seattle. There was actually a huge opportunity there, but they didn’t really do that until around the 2010–2011 era, around when Microsoft started to get their crap together.

Mm.

There was this huge opportunity to take all of Microsoft’s talent in the 2000s that never really materialized because it sucked so bad to work at Amazon, and no one in Seattle would do that.

Horror stories coming from Amazon.

Right.

Yeah.

It sort of saved Microsoft. But my other point was, look, you’re the tech industry. You’re the most important industry in the world. You’re going to get critiques like this.

Yeah.

Internally, The New York Times was like, “This is amazing. People really like this. This is the stuff people will pay for. We need to do less day-to-day stuff and more in-depth things like this that prompt people to pay, to feel good about paying.”

So, again, setting aside your view of The New York Times and your personal opinion about their politics, from a business perspective, it’s exactly what I’ve been calling for.

Yeah.

You have to have a whole, holistic strategy that combines everything. Your business model and your editorial need to work hand in hand. This whole idea of the firewall between business and editorial was a vestige of being a geographic monopoly.

Mm-hmm.

It turned out you got that for free. That wasn’t the key to you being successful. Going forward, once that was gone, once you had to compete, your whole business needed to be aligned in terms of what you’re trying to accomplish. The New York Times did that. They did that sooner than everyone else, and they’ve won.

And they were early to bundling as well.

Right. Again, setting aside the partisan political component, which a lot of our audience is probably annoyed at The New York Times, from a business-analytical perspective, it’s very impressive—what they’ve done over the last 10 to 15 years.

Anyhow, they’re the exception. Everyone else has had a bad time. It’s been getting worse, and now AI is just accelerating what was already occurring. AI isn’t killing them; AI is accelerating the death that was occurring. Maybe it’s a mercy killing, to a certain extent.

Yeah.

And so this is all about—

To put a final point on that, even The Washington Post has been struggling in the past couple of years. If The Post is struggling, you can imagine what midsized institutions are dealing with in the modern internet era.

Oh, yeah. You go to their webpages and they’re horrific. There are ads everywhere. I’ve talked about how I had to do a 180 on the whole ad-blocker thing, when I was very principled about it. I was like, “Look, if you’re going there, you have to pay the price.” But you literally can’t surf the pages. These videos just take over the screen. It’s unbelievable.

Mm-hmm.

How is this going to be sustainable? It’s not. They’re just scratching out every penny.

And it hasn’t been.

Yeah, no, exactly.

Spoiler alert, yeah.

The reason I like this email and the reason I like this analogy is that the Eras Tour is something of a red herring. I believe it made over $1 billion a couple of years ago when Taylor Swift was touring the country and then the world. But the idea of making your money in music through live shows has proven fairly successful for a larger cohort than just Taylor Swift, and it gets at some of what you were writing about with the second half of Monday’s article, in terms of an actual optimistic future for people who are publishing on the internet. Does that—

It’s a good point.

Does that resonate with you?

Well, it’s funny. It does. I was going to make another point, where I think the music industry and Spotify are interesting. I think one of the big questions is, how do you price content for this whole pay-per-crawl thing or whatever? That’s part of the whole problem. How do you actually create some sort of market here? I could see, in the long run, if there can be some sort of market maker that works like Spotify, in that there’s a pool.

Mm-hmm.

How does pricing work on Spotify? Everyone’s money goes to Spotify, and then it’s split up based on who gets plays, right? It’s actually fairly straightforward. That’s why it’s not like you get paid a penny per play or whatever it might be.

Yeah.

It’s what your share of all plays is, and then that’s your share of the pool of money that goes to music. I can understand the consternation artists have about that. But at the end of the day, when you have a completely open market and anyone can put music on Spotify, you need some sort of constraining mechanism to actually apply a price to the stuff that’s in there.

Mm-hmm.

Could there be a market maker on the internet that has a pool of money, and everyone goes into it, and then there’s some aspect of—you know, there’s going to be a huge analytics and measurement problem here that needs to be figured out?

Yeah.

But then, based on that, it’s split up so that it’s not some sort of uncapped amount, or whatever it might be. By the way, this speaks to it being a commodity. If you want to have unlimited pricing power, you have to get out of the pool, and you have to have some sort of model that works independently. This is like, look, there’s an amount of money in the world for iron ore.

Yeah.

So you could go and dig for iron ore or oil, right?

Well, I mean, that’s the thing: publishers have even less leverage than individual musicians because at least music is distinct. One song is distinct from another. The reason this current information is valuable is because these models need facts. People aren’t seeking out prose from The New York Times or anybody else. And so you’re not—

Yeah.

—going to have very much negotiating leverage with anybody that’s crawling your site for AI because ultimately, the information is available in a bunch of different places.

That’s right. We’re talking about token-generation farms.

Yeah.

Again, it’s not even content. It’s just the statistical association of tokens that they’re looking for.

And it’s also who won the Wizards game? There are going to be a lot of different places you can go to find out who won the Wizards games. Spoiler alert: not the Wizards, at least this decade. But that’s part of the deal.

Thank you, Andrew GPT. But yes, Sharp GPT.

Mm-hmm.

But, yeah, this is how I think this market is going to shake out. And just to go back to Finn’s point, everything on the table now is not going to work. All this discussion is definitely looking at what might work.

One thing that I have to be cognizant of, and everyone in this space needs to be cognizant of, is not getting bound up in idealism. You have to look at the cold, hard realities. And that’s why I don’t think the existing publishers are going to survive this. It’s going to be something new that comes up. And by the way, maybe it’s not. Maybe we actually don’t need that much content.

Mm-hmm.

The LLMs are going to generate all the content. There are just going to be a few people plugging facts in. Again, I think there will be a market for this, but let’s be clear about what that market is. That market is mining tokens.

All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at email@sharptech.fm.