Windsurf 疯局：硅谷生态系统侵蚀之际大科技获胜，Cloudflare 要修复互联网经济

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

Feeling really washed, Andrew. I got on here—

Oh, boy.

I’m like, “We gotta do a podcast.” My voice is a little gravelly. Not sure how it is. You jump on with kind of a weak, gravelly hello, if I’m gonna be honest with you.

Really?

Yeah. We were both in Vegas at Summer League last weekend. It’s been 5 days. We might be totally washed.

I mean, look, I had a cold on the way into Vegas.

Uh-huh.

And I was kicking it on the way in.

Uh-huh.

And now, I have to say, the cold is still lingering, and 72 hours in Vegas with not very much sleep is not a good recipe for restoring yourself to full health.

Eh, sounds like excuses. You’re just getting old.

Well, look, I think the gravel adds character. So as we age here and we both become increasingly washed, we’re only gonna sound cooler. We’re all just trying to get to Tommy Heinsohn’s level, where you could hear the packs of Marlboro Reds that he smoked throughout his lifetime when he was announcing Celtics games. So that’s my personal goal in podcasting. You’re free to join me if you’re down.

I’m not sure how to respond to that. I think we should keep going on the podcast.

There’s never a bad time—

To invoke Tommy Heinsohn.

…to start smoking Marlboro Reds. Yes.

1. The Windsurf Deal Collapses

Well, washed or not, it’s great to be back. It’s great to see you, and we’re gonna begin this episode with Windsurf. We’re coming in batting cleanup after a week of news and rumors, occasional outrage, surrounding a company called Windsurf that makes an AI-assisted coding tool. For anyone who’s unfamiliar with this story, I’m gonna try to offer a brief recap of the context here before we get into a handful of emails.

Back at the beginning of May, OpenAI reached an agreement to buy Windsurf for $3 billion. You and I mentioned it in passing on the show. It wasn’t exactly a bombshell of a deal. However, between May and last week, OpenAI was struggling to close the Windsurf deal, in part because of tension between OpenAI and Microsoft, because Windsurf didn’t want Microsoft to have access to its IP, and OpenAI and Microsoft couldn’t reach agreement on that point. This is according to reports, at least. And then on Friday of last week—

Which, by the way, I think was day 61 after the deal was announced, because they had a 60-day sort of exclusivity window.

Ah, there you go. Well, on Friday of last week, news broke that Google struck a deal worth about $2.4 billion to hire Windsurf Chief Executive Officer Varun Mohan and co-founder Douglas Chen, along with a small group of Windsurf staffers, to work at its DeepMind AI unit. Windsurf, in that deal, reportedly retained $100 million that could be paid out to those not hired by Google if the company shuts down in the near future.

I think, just as a clarification on that point, that was the assumption. The stated reason was, “No, Windsurf is still a fine company. They have customers. They’re gonna continue serving them.”

So $100 million to continue operations in that scenario. Is that right?

I don’t think anyone said what it was for, but the public statements were, “Oh, Windsurf’s gonna keep being an independent company.” They just got completely lobotomized, but everything’s great.

The remainco.

Yeah.

Absolutely. Yeah. Well, all of this set off a weekend of speculation and outrage throughout the tech industry, at least on tech Twitter, about what would happen to the remaining employees at Windsurf after it had been hollowed out. This was particularly concerning for the employees whose stock had not yet vested and who weren’t sharing in the $2.4 billion that Google was paying for the licensing and the leadership team.

And then since then, after 96 hours of outrage, news broke that Cognition, another startup working on AI software development, has agreed to acquire Windsurf, including full ownership of the Windsurf IP, as opposed to Google’s license. And so those Windsurf employees will now have a job at Cognition, and it’s possible that the $100 million left at Windsurf will be split among the existing employee base. That has not been confirmed.

Well, they have said that they’re going to—so, when you get stock options in a company, it’s usually on a 4-year vesting schedule. That’s varied and changes over time, but say you’re promised $100,000 in stock over 4 years. You get $25,000 worth of stock a year.

Anyone who was there and hadn’t been there for a year had nothing vested. If you’d been there for 1 year, a quarter of it had vested. And so I think the news from Cognition is that they’re going to accelerate vesting, so you’ll get everything that’s promised to you. And, according to some reports and people digging up documents, that $100 million broadly matches the outstanding promises to employees.

Mm-hmm.

So there is an angle where this is actually an even better deal for the employees, because they’re getting their 4-year vest accelerated immediately.

Yeah.

If I were a Windsurf employee that got paid out, I might just want to bail. Take the money and run. But yes, by and large, in a narrow sense, there’s a happy ending here. I think that was kind of inevitable given all the outrage.

Right.

That, if you’re above the line, as it were, stuff gets solved. Had no one noticed or cared—

And if not, then—

Everyone would have been normal.

…tough luck.

Yep.

Yeah. Well, and you mentioned that for PR reasons alone, it behooved everyone involved to try to take care of the remaining employees at Windsurf. But that’s the narrow sense. Let’s broaden the aperture a little bit.

2. Startup Risk Gets Repriced

Tony says, “Ben, can you comment on the second-order implications of Google swooping in and hiring the core team at Windsurf? This feels like it is downstream of the broken tech regulatory regime, but I also sense that there’s a ton of negative implications for rank-and-file startup employees and the ecosystem as a whole. If early employees can no longer expect to share in the upside of these exits, then folks will be more hesitant to join startups. I just joined a Cursor competitor myself, and now I’m a bit more nervous about my prospects. I’m a designer, and there’s no way I would be part of the acquisition core if a Windsurf-type situation happens.”

So what do you think? Where should we begin?

I completely agree. This is a bit of a slow-motion train wreck that I would usually be very excited to claim being-right points for. Here, I have to mournfully accept my grant of being-right points.

I told this story, I think, on Dithering earlier in the week, but throughout the 2010s, I was pointing to the Facebook acquisition of Instagram.

You’re just using that to highlight how people were thinking about regulation wrong. And that—

The reason that was a problem is you had an aggregator in Facebook buying a burgeoning aggregator in Instagram. Even though it was small, you could see where it was going. And there’s—

Mm-hmm.

There are 2 aspects of this. 1, it was knowable if you knew what was happening. It was clear to me when they bought it that it was a problem. I analogized it to Microsoft buying Netscape in 1993. But there was no metric that said it was bad.

And by the way, in popular culture, Facebook was mocked endlessly for spending $1 billion on this tiny little photo-filter app or whatever. And—

Yeah.

Sometimes people are just smart and figure it out, and it’s not—

Mm.

I think there’s a general movement amongst your people—the sort of antitrust-solves-the-world people—that has too much self-ascribed agency. We could have stopped this; we failed, da-da-da-da-da.

And sometimes it’s just the case that Mark Zuckerberg was ahead of the curve in this case and made a very smart acquisition. Twitter was trying to buy them as well. There certainly is a bit where people in tech who were really examining these companies and understood how they grow knew that Instagram was going to be a big thing. But—

Yeah.

It’s important to keep in mind that 98% of the world thought this acquisition was insane. And just because it worked out in spectacular fashion doesn’t mean that, ipso facto, all acquisitions are bad and must be stopped.

Unfortunately, I realized in 2020 that was the way it was being interpreted. I was at this antitrust conference at Stanford that was put on by the DOJ. I was a panelist on some sort of thing. And it was jarring to me at that conference how basically the accepted wisdom about Facebook buying Instagram was that big companies should not be allowed to acquire.

Right.

That was basically the takeaway. And to me, I went back and I wrote First, Do No Harm, basically saying, “If this is the takeaway, we’re on the verge of destroying the startup ecosystem,” because what makes Silicon Valley remarkable is the extent to which new companies and new ideas are almost completely de-risked.

How so?

Well, everyone knows about the financial angle. It's remarkably easy to raise money in Silicon Valley compared to anywhere else in the world because you have an entire financial category in venture capital, and then you go down to the seed investors. Everyone knows the way to get rich in Silicon Valley: make a little bit of money, then become a seed investor. All you need is to get into one little company that goes big and you've made it, right?

Everyone is so eager to give away money. It's an amazing cultural situation where, in most countries and most places, you have to beg to get money, and you're going to get all these terms around it. People are like, “No, $50,000 check, no strings attached. Here you go. Another lottery ticket for me.” It's amazing. It's priceless how important that is.

And that's how you get some of the biggest and most important companies that sound stupid and start as toys or dumb ideas. The fact that basically anyone can get money is amazing. Everyone's aware of that part.

The part people weren't thinking about at that conference, and what we're seeing manifest now, is that if you want to build a company, you don't just need money. You need people. That money goes to pay for the people. That's the cost, right?

Mm-hmm.

It's not really paying for compute anymore because, with cloud services, you pay as you go. When you start out, Amazon will give you $100,000 or $200,000 in credits. Azure will do the same thing. Your compute and all that stuff is basically free.

What all the money is going for is employees to actually build the product and, eventually, over time, customer acquisition and sales and all that sort of thing.

Mm-hmm.

Employees who could be working at big tech companies themselves but are willing to take a risk at a startup because of the security that this ecosystem has afforded them.

That's the thing.

Yeah.

They're not taking a risk, right?

Mm-hmm.

Usually, this is another thing that I think people who are not in the Valley—some of our listeners can probably attest to this much more than people in the Valley can appreciate—in lots of areas, you don't take the stable job. You do the crazy startup that sounds dumb, and it fails, and you're screwed. You're out. You're done.

Right.

Absolutely, because you've got a family to support. I've got a friend right now who's about my age and debating whether to go join a startup. It's a startup that's not in Silicon Valley, and it's a cool opportunity. There is some financial upside.

But he's got a son who's 2 years old, and if it doesn't work out 3 or 4 years down the line, that really complicates things. He's going to be in his mid-40s, and it just makes it a lot more complicated to get a job at that point. These are the considerations that normal employees elsewhere in the country have had to weigh, and apparently that isn't as common in Silicon Valley.

The sort of implicit deal in Silicon Valley—and, again, this is just a broad characterization; I'm not saying this always happens—is that there's a lot of startups that fail and just go away, and nothing ever happens. But, by and large, there are new jobs to be had very easily.

More importantly, if you get a little bit of traction or you did something nifty, but it doesn't work out, you can do a startup that's just focused on a cool idea, and you'll get funded even if it's not clear what the market could be.

Mm-hmm.

Take Uber. There was that famous set of financial bloggers saying that Uber had no chance, and they were characterizing it by the size of the taxi market. They weren't actually considering that Uber might vastly increase the market for personal transportation, which is actually exactly what happened.

Obviously, you want to go to a VC and say, “There's this huge market opportunity. We're going to fill it.” By definition, someone is probably already doing that, right? Some of the best startups—it's not clear what the market is. They create their own market.

Actually, that's the key characteristic of aggregators: they created their own market. Facebook advertisers were created because Facebook existed. That's why they're sort of antifragile, and their customer base isn't going anywhere. The same thing with Google, and the same thing with a lot of these companies.

You have this situation where you can go in and just focus on a cool technology and say, “I think there'll be a market. We'll see what happens.” If it doesn't materialize but you still have a cool technology, guess where there are markets that could use that technology? The big tech companies.

They have billions of users. There's a real advantage in just buying a startup with a cool idea. That idea immediately gets propagated to billions of people. That's actually a huge consumer upside from big tech buying small companies: it immediately becomes available.

Whereas if a company has to go through and acquire customers, it might be 20 years before that technology is broadly available, right? There's a real consumer benefit to big tech acquiring companies, in that it disperses the technology.

Now, the counterpoint is that a lot of ideas go there to die, which is true. That is what happens to big tech companies. I'm not denying that. I'm mostly focused on the upside here, while acknowledging that there are downsides.

But the other thing is that a lot of companies have this sort of deal: “Okay, we'll acquire you.” What happens in a lot of these acquihires is that no one makes any money.

Mm-hmm.

Often, the founders are actually the ones who really get wiped out. What happens is that the investors might get their money back at 1×. They put in $20 million, and they get $20 million back. They usually have preference in the capital stack, so they get their money before anyone else does.

If you see a company that raised $50 million acquired for $50 million, usually the founders made nothing because they're lower in the preference stack. All that money goes back to the investors, and then the founders get nothing.

What they get—and, critically, what all their employees get—is basically a 4-year job at Google or a 4-year job at Facebook. They might all go work for the big company, and maybe they're going to work on their product.

Usually, they're going in as a team, and the company says, “We're going to support this,” and then they get dispersed and sent to other divisions and things along those lines. But from an employee perspective, if you're a founder, if you're the founder of Windsurf, you're fine from this deal. You're getting paid big money.

But if you're someone who did design, for example, and any future Google coding thing is going to go to Google's product, they don't need more designers. They have designers. Ideally, even if it doesn't work out, you didn't make any money, but you have a good job. That's a tremendously risky situation.

You have a good job. And you're not left high and dry after working 80- or 90-hour weeks trying to get the Windsurf product off the ground and making it attractive enough so that Google wants to hire the leadership team.

If you're just left holding the bag after all that, that's really frustrating. Granted, in this specific case, that's not what is going to happen to those employees, but it's a valid concern.

Yeah. Well, let's not anchor on this specific case. Let's zoom out on the role that big tech companies acquiring small companies played in the ecosystem. The vast majority of those companies have failed or are not going anywhere and are not remotely an antitrust concern.

Mm-hmm.

These are failed companies. This was an essential part of the Silicon Valley ecosystem. It's not just that you can get risk capital that will invest in you. You can get employees who will take a chance because what's the worst that can happen?

Previously, the worst that could happen was that I get a job at Google. Now, the worst that can happen is that I don't have a job, and, by the way, the hiring environment has gotten tight. It's a lot easier to get into Google if you're bought by Google than if you have to apply for a job with a gazillion other people and deal with all these sorts of things that are going on.

Yeah.

That conference was such a wake-up call for me. I'm like, these people have no idea what they're about to do. It's explicitly what I wrote. I said the number one problem with this focus on acquisitions being bad, without any real understanding of what made the Instagram acquisition unique, is that it's going to really screw up the employee incentives to work at startups.

Yeah.

That's basically exactly what has happened.

3. Big Tech Skips Full Acquisitions

Well, and also, with this specific Windsurf example, I think the reason the outrage hit a fever pitch over the last week or so is because the context for all this now feels more urgent. You had Meta buying Alexandr Wang and members of Scale AI. You had Microsoft, a year ago, hiring some of the team at Inflection AI.

Well, basically all the team from Inflection. Yeah, Microsoft started this, and then Google did the same with Character AI. So, no, this is—

Yes.

You’re right. This is not the first one.

That’s the workaround.

Yep. The problem I see is that the entire concept that we buy the whole company was a cultural norm. It was a societal norm, in that if you’re a startup company, you need an HR person, right? You need a person who does payroll. There are all these company-specific functions that Google does not need.

They don’t need all the support staff. They don’t need the—

Right.

The employees who are lowest on the totem pole—the people all companies need to function but who don’t differentiate the company in any way—if you think about it from a big tech perspective, those are the people who get laid off, right?

Mm-hmm.

Those layoffs are painful and get bad press and all that sort of thing. If you’re a big tech company, it’s like, “Can I get just the engineers and just the people with the ideas?” That’s way better, and I don’t need to get a regulatory investigation.

Right. There’s not a 6-month waiting period. Even if the merger or acquisition isn’t blocked, just from a time standpoint, it’s more efficient to do it this way.

Absolutely. Everything about this is better for big tech, and the worry I have is that we broke it, and it’s done. It’s not coming back.

There’s no reason for a big tech company to do a full-scale acquisition unless, ironically enough, it actually does raise anticompetitive concerns because it’s a really successful company that’s worth paying top dollar for. But what I’m concerned about is the companies—you know, Windsurf has a business, I think, with $80 million in annual recurring revenue or something along those lines. It feels like—

Yeah.

They were getting beaten pretty comprehensively by Cursor in the independent environment. Again, I don’t know. But generally speaking, I think people underappreciate the extent to which startups were de-risked in the Valley, not just from a money perspective but from an employment perspective.

In the long run, for the ecosystem as a whole, there are so many things that, for everyone in the short term, are bad. For a big tech company, buying a whole company has a lot of waste associated with it, but because that was the Silicon Valley expectation of how you did it, they did it.

Yeah.

And no one complained about it.

Exactly. It incentivizes people to found startups and people to innovate, and most of those startups end up failing. Some of them are subsumed by big tech. But it has fueled innovation in the Valley for 40 years now, and so if that model is—

I know, but I want to be super critical here. The thing I’m not worried about is the people spurred to innovate and the startup founders. I mean, there’s some concern there. The concern I have is with the rank-and-file employees, who are not the ones with the ideas.

The problem is, if you’re the person who wants to innovate, you need those sorts of people.

Yeah, you need to hire people.

You need the non-innovative people.

Right.

Right, exactly.

In jeopardy here, yeah.

So, yeah, I think it’s a real bummer. I’m sad about it. I also think that it’s over. If I’m a big tech company, I see no reason to do a full-scale company acquisition ever again. I can pay a lower price and get all the best employees, not have to worry about layoffs. To your point, it can happen right away, and I think it’s done. I think regulators killed the Silicon Valley model, and I think it’s dead.

Oh, boy. Well, we can move to Joseph, who addressed his email specifically to me. He said, “Dear Andrew, and not Ben, because you own this one.” Oh, great.

“In a shocking twist worthy of M. Night Shyamalan, there have been unforeseen consequences for the law-stretching antitrust zeal Lina Khan pioneered during the Biden administration. Instead of a predictable acquisition environment for investors, management, and employees, acquisitions are now made via convoluted backroom acqui-hires and technology licensing. See Scale AI, Character.AI, Inflection AI.

“Now, OpenAI’s Windsurf acquisition has fallen through. Some founders are getting hired at Google, some investors are getting paid out, and a whole lot of regular employees are getting left with worthless equity in a gutted carcass of a startup. Huzzah, we beat big tech. Is this really the point of antitrust? It feels like government fought the last war just in time to hopelessly botch the next one.”

This email is amusing to me now because it really captures the feisty tone that was pervasive on tech Twitter last weekend. So thank you so much, Joseph. He followed up and said, “Actually, I wrote that after a couple of margaritas Saturday night. Apologies if I offended anybody.”

We encourage this. We encourage margarita-fueled emails.

Exactly. Maybe everyone should have 2 or 3 margaritas before they write in.

Yeah, we’re going to have a checkbox: “How many drinks have you consumed?”

The more drinks, the more likely this is read.

Oh, boy. Look, you and the emailers have appointed me as the spokesman for antitrust enforcers, which is a position I may have embraced in the first year of hosting this podcast. My own thinking has evolved a lot since those halcyon days, those first 6 to 9 months.

What I would say to Joseph is that, to your point earlier, I’ve learned about Silicon Valley and the landscape that incentivizes not only startup founders and startup employees but also venture investors. I now fully agree with Ben’s argument that protecting these sorts of exits is really important if you care about protecting the model that has historically led to a ton of innovation in this industry.

The only thing I would add is that it’s a harder problem than it seems, and calibrating the right kind of ecosystem isn’t easy. I mean, there are still plenty of deals that are problematic and do foreclose future competition.

The Facebook acquisition of Instagram is still a problem. The fact that it was—

Right. Well, to take an example from a couple of years ago, Adobe buying Figma was held up by regulators in the EU and the UK, and the DOJ was apparently investigating it as well. Ultimately, they had to abandon the deal. Now Figma is going to IPO. That’s a positive outcome for the future of design software, and it’ll be a positive outcome for Figma employees in the end. Although I can’t imagine being a Figma employee while that acquisition was being held up.

Yeah, for sure.

But it’s worked out, and it wouldn’t have gone that direction without regulatory scrutiny. So I don’t know that there is a one-size-fits-all solution for how this should go, but I agree with Joseph’s sentiment that there should have been more discretion than there was under Lina Khan and the FTC during the Biden administration. But I don’t think we can just abandon regulatory scrutiny entirely.

No, I think I agree by and large. I would just make a couple of points in defense of your position. Own it, Andrew.

One, you could make the case that the Facebook acquisition of Instagram was so injurious to competition that that alone outweighs all the other issues. I don’t agree, but I think that is an argument that could be made. If we just stop one—

And to be fair, that’s exactly what I would have argued in the first 12 months of hosting this podcast. I said, “Look, the bad ones are so bad that something close to a per se ban isn’t the worst thing in the world because, at the end of the day, it’s going to promote competition over the long term.” But at that point I wasn’t as conscious of how much harm you could do to the innovation side of things and the incentives on that side.

Right. So, yeah, I think we’re on the same page there. The other thing is Lina Khan gets most of the blame, and appropriately so. This happened while Trump was president, so there’s definitely a bipartisan aspect to this. Some of the more populist wing of the MAGA movement—J.D. Vance has been very supportive of Lina Khan and that sort of angle.

Again, I think Lina Khan dramatically accelerated it. By the way, the way acqui-hires used to work was that they took a month. No one thought anything of it; it was like, “Of course we’re not going to investigate this,” so they closed quickly.

Yeah.

There wasn’t a time penalty for doing an acquisition. But to your point, the idea of saying, “No, we’re going to look at every single acquisition,” means that even if you approved every single acquisition, you’re now imposing a huge cost and making this kind of thing unviable. In a lot of cases, these startups aren’t going to last for 6 months, right?

Mm-hmm.

The idea that we're going to investigate every single one is itself bad, but it was a bipartisan thing. I think there's an overall lack of understanding, to your point, about what has made Silicon Valley unique. And I think the other bit that you sort of got at is—this is what makes this really hard—and it definitely is.

If I were to characterize how my position has changed, my job professionally is to look at all these sorts of things. I've been thinking about tech and the business side of tech for three decades now, or however long it's been. Actually, more than that.

Yeah.

I think that's wild to think about. So, to someone like me, it was immediately obvious the Facebook acquisition of Instagram was bad, and I agree with you, the Adobe–Figma one is a good example. That one kind of stunk from the beginning. It wasn't clear how this wasn't anticompetitive. Adobe did kill its competing product almost immediately, but even then, the route for Figma to build competitive products was super clear, and it's a good example of, yeah, regulatory intervention was appropriate. We're at a better place afterward.

4. Antitrust Becomes A Binary Choice

I think the challenge is, for someone like me and for you as you're immersed in tech, it becomes clear that any one-size-fits-all solution is bad, and you should have judgment and pick and choose the right outcomes. But you and I are not the regulators, and I think what happens with a lot of politics is that the choice is actually a binary one. It's like people who complain about the political candidates: "Well, I'm not going to vote because they're both terrible," blah, blah, blah. The way it works is you have a binary choice, and you pick one.

Yeah.

And I think that actually applies to a lot of politics. So the question actually isn't, can we make very sophisticated decisions about distinguishing between Facebook buying Instagram and Adobe buying Figma and all these little acquisitions? Or is it, do we want to blanket-allow small-scale acquisitions or blanket-ban small-scale acquisitions? That's the actual choice in front of us.

And so this is where I've shifted significantly on this point. I used to be the guy saying, no, and trying to draw very narrow guidelines saying, "Oh, do this here, do this. Don't do this there."

Do this but not that.

"Extinguish this."

Yep.

Right. And appreciating the reality of politics, which is that you're choosing the least-bad option, I would rather blanket-allow than blanket-ban.

Yeah. Well, and I think that is a reasonable argument when you look at the logistics of how any of the enforcement would work going forward. The other thing that I do wonder about, in the wake of some of the more notable acquihires over the last 12 months, whether it's Inflection or now Windsurf—

But they're not acquihires because there's no acquisition happening. It's—

Well, that's true.

You need a new word for it.

Just hiring.

Yeah, exactly.

This is just good old-fashioned hiring. But I mean, I wonder how much of this is unique to this exact moment in AI—

Yep.

—where suddenly there are these talent wars.

We've been talking about the talent wars, right? Yeah, so it aligns very much.

And a handful of brilliant researchers and founders are just literally more valuable to these companies than the fledgling startups that they founded. And so I don't know.

5. Foundation Models Reorder AI Startups

The Character.AI one is actually a really interesting example in this regard. Character.AI was founded by the guy who invented the transformer, which is at the core of LLMs. And the reality is, I think Character.AI started out training its own models to be these entities that you talk to, before Elon Musk decided to just go all the way with the waifus.

Oh, yeah.

It became actually pretty clear that you're not going to keep up as a startup with the foundation-model companies. If you're building a company, it should be built around the models that other people are investing in. At that point, it didn't make sense for the founder of Character.AI to be at Character.AI. He should have been back at Google or back at OpenAI or one of the foundation-model companies.

And in that case, it made sense: We're not going to do deep research anymore. We're not going to do deep learning. We're not going to train our own models. We're going to build a product on top of foundation models. That means the people in our company—it would have been a waste of resources and talent for Noam Shazeer to be stuck at Character.AI. It's actually better for everyone that he's back at Google because he's doing what he's good at.

Yeah.

And so there is some subtlety to your point about this point in time. It wasn't clear for the first few years of AI whether it was going to be widely dispersed, with lots of people doing models, or whether there would be a few foundation models. It seems pretty clear there's going to be a few foundation models, and people will build on top of them.

And given that, some reorganization kind of makes a little bit of sense. So maybe we'll come through this and it'll all be fine. Maybe all these companies were actually malformed to begin with because they didn't understand the structure of the industry. That's possible.

I am worried it's just such a better deal for big tech to hire these people.

Well, and not only that, I mean, look, the Silicon Valley model has been an aberration relative to the way the rest of the world does business or has done business.

That's why it's called "First, do no harm." You have this amazing thing. If you don't understand it, be careful what you're messing with.

Yeah. It's not directly analogous to the end of high-trust societies the way we've talked about in the past. I mean, there was a social convention that appears to be on the way out.

I think it's very analogous.

There was a social convention—

Yep.

—that appears to be on the way out.

Noam Shazeer. I knew it was Noam. I thought it was Shazeer, but I didn't want to get it wrong. I had it right in my head. Noam Shazeer, founder of Character.AI, inventor of the transformer, now back at Google.

There you go. Well, hello, Noam, if you're out there listening.

Arvind says, "I read Ben's Tuesday follow-up on the Cognition acquisition of Windsurf, or at least what's left of Windsurf after the Google deal. I don't think Ben explicitly addressed whether this follow-up reduces the stinkiness of these kinds of deals. If this becomes the new norm with investors, where founders and early employees get the big payout and remaining employees get sucked up by another startup, can the tech startup ecosystem still be healthy?

"After all, there can still be big economic incentives for entrepreneurial types to start companies and join early-stage startups. The main difference is that not everyone is guaranteed that big tech job after the acquisition. Is that so bad?"

What do you think? Because what I'm reading into that email is a new framework that might be possible, where it accelerates consolidation among startups while some of the founders are taken care of by big tech. Is that a dream within reach?

Well, I think you just quoted my Stratechery update, so I'm glad you got that from the email.

Okay, good.

So, really circular.

Influencing the influencers here.

That's right. My follow-up: I accept that the old model's dead. I think it's going to take a while for people to come to grips with it.

Yeah.

My view, just looking forward, is that it's very hard to see. This entire thing was governed by social convention. That social convention was broken by regulators, who gave big tech an out to no longer abide by this convention, right?

Mm-hmm.

And it is funny, and I think there are some people who are upset about me blaming regulators. They're like, "Big tech wants to do this." But that's missing the sort of guardrail that did exist—

Mm-hmm.

—which was social mores.

Yeah, social mores.

Right. And so I think the old model's dead. By my next update, I'm like, I mourned it on Monday, and by Tuesday I've moved on. So, given the fact that it's dead, what might it look like going forward?

This did end up in an interesting position, which is that certain talent goes to Google. Cognition makes a product called Devin. I think Devin is probably a little aggressive as an agent that does everything. I think there's real value in having an IDE like Windsurf does, where you're getting data, and it's more of a gradual move to an agent world where you start helping developers, you get more data, you do more and more for the developers, and then you end up in a Devin-like world.

Mm-hmm.

So I think there's a bit where these companies make sense together. Windsurf does have existing customers, and the people who were building Devin are in the same broad adjacent space as the people who were building Windsurf. And by the way, there are a lot of situations where some startups are pretty complementary and would be better off together.

Right. Yeah.

And they never end up together because either one dies or X, Y, Z, whatever might happen. And so you could definitely spin a case that we're getting pushed to a more compelling model that makes startups more viable as a category precisely because they'll be forced together sooner, consolidate sooner, and combine forces sooner. We'll see. I think it's a reasonable thing to be hopeful for. The old model was pretty good.

Yeah.

I'm not sure it's going to be better.

Look, you're just searching for silver linings here as the social conventions erode in Silicon Valley. I think that's what I enjoyed about your take on Tuesday.

It's dead. I don't think—yes, there will still be company acquisitions, et cetera, but I just think the guardrails that prevented this sort of thing are gone.

Yeah.

Even if regulators back off and say, "We're not going to investigate," it doesn't matter. If you're a big tech company, why do you want to acquire a little company? This is actually much better, much cleaner, much easier. You don't have to deal with all the rigmarole of acquiring a company. It's just a new hire, which is very trivial.

But, yeah, maybe there'll be some outcomes. Things change, things develop. This is the critique I have when I trust people: things aren't static.

Yeah.

Stuff changes and responds, and I think this was a good outcome. We'll see what happens with this new combined company. Hopefully, it's a model of interesting ways to build startups going forward. But there's going to be a lot of proof points to show that it's better.

Yeah. Well, and you get being-right points in the interim here, unfortunately. Mournful being-right points because you did foresee a lot of what has shaken out over the last couple of years.

The thing about big tech companies, as someone who is still a member of the antitrust enforcement movement, is that big tech companies tend to be pretty adaptive. And so they've quickly—

I know. Yeah.

—adapted to the new environment here. Very frustrating.

No, we have to be pretty honest that the net outcome here is a benefit for big tech companies—

Yeah.

—and screwing the little guy. There's no real debate otherwise. I guess you could say that in the very long run, it's worse for big tech because there are fewer spurs for innovation and fewer startups they can acquire.

Mm-hmm.

But that's a 10-year, 15-year roadmap thing. If people in antitrust were worried about 10 or 15 years, they'd relax a lot because it turns out that stuff changes very rapidly.

Well, if there were no regulatory enforcement, would Meta have just tried to buy OpenAI and ChatGPT 2 years ago? And would that be sort of the alternate doomsday scenario?

Well, people have agency in these decisions. They tried to buy Snapchat; Snap said no. Big mistake by Snap.

But Instagram could have said no. People do have agency, and I think that can sort of be forgotten here. If you were in OpenAI, would you sell? Absolutely not. No way. Sam Altman is not going to work for Mark Zuckerberg when he has the potential to build a multitrillion-dollar company. That's not going to happen.

6. OpenAI Chooses A Strange Structure

He wants to rule the universe. The only reason I thought of that is because we have this next email from Vikesh. He says:

"Andrew and Ben, something doesn't quite sit right with me about Ben's Chesterton fence take on OpenAI."

And this is in reference to your article on Monday, or your update on Monday.

"What if Sam Altman and Elon Musk did not tear down the Chesterton fence of setting up a normal corporation with a normal cap table governed by normal profit incentives? Would Ilya still have joined that company? Would Dario have bothered to accept the mission? Would the rest of the church of AI true believers at OpenAI have been tempted away from more mundane but lucrative roles?

"And if OpenAI did not get all these true believers to explore this space and ultimately launch ChatGPT, would we still be waiting for the penny to drop at Google about what products could be built from their transformer innovation?

"It's clear that OpenAI's convoluted corporate structure made them lose out on the Windsurf deal, no arguments there. But if Altman and Musk did not mess with what works regarding corporate structures, could humanity have lost out from not appreciating the power of transformers and large language models?

"If we believe that large language models will be a good thing for humanity, should we ultimately be grateful that Altman and Musk pursued a utopian idea like a nonprofit tech company?"

A wonderful zag from Vikesh there. Do you have any thoughts?

No. I think it's a very plausible argument. And I've always ascribed that to why OpenAI is how it is. You need some sort of advantage. We have less compute than Google. We're going to pay less than Google. We have everything less. Why should researchers come work for us? Because—

Let's protect the future of humanity together.

That's right. That's right. It's interesting, in retrospect, looking back on that, and this is, I think, a point that John made on Dithering that was kind of an aha moment for me.

Mm-hmm.

So that was sort of the stated reason. I was super cynical of the deal immediately when OpenAI was formed in 2015. I wrote an update saying, "Spare me the sort of utopian ideals."

There is a world where Y Combinator was threatened by Google because Sam Altman was the head of Y Combinator then. Y Combinator is threatened by Google. Tesla is threatened by Google. It was a very cynical sort of take and—

Yeah.

—skeptical.

I think, in retrospect, the mistake was conflating Altman's motivation and Musk's motivations. I think the "Oh, this is a clever way to hire researchers when we don't have the resources of Google" angle is more Altman style, right?

Mm-hmm.

That's a reason to do it. Did he really believe in the nonprofit mission, or was this a way to get off the ground? He's always been interested in AI, and suddenly he got all the best employees through one simple trick: doing this convoluted structure.

And the whole tie-in between OpenAI and Y Combinator generally was always sort of a tenuous one. Y Combinator is a startup accelerator.

Uh-huh.

Which startups are we talking about, et cetera? I might have been right about Musk's motivation, and you see this with Musk: suddenly SpaceX is investing in xAI.

Yeah.

Real corporate governance win right there, just in general. But there is a bit where, with Tesla in particular, you can clearly see that having leading AI is good for Musk's other companies.

Yeah.

And so I think my cynicism in retrospect was appropriate for the Musk angle, who also was paying for the whole thing, so he had the biggest driver. Whereas I would ascribe Vikesh's angle maybe more to Altman, who, by the way, needed money and Musk was providing it, so he was also incentivized to go along with this angle. So I'm going to absolve Altman a little bit in this deal.

Well, Sam's perspective and strategic incentives still sound pretty cynical relative to the stated mission put forward by OpenAI—

Right. For sure.

—in 2015.

Look, it's a cynicism-off—

We're working with degrees of cynicism. Yeah.

That's right. I'll just say my specific 2015 critique actually held up pretty well for Musk and his companies, and it's being seen in what he's doing with xAI and his companies. That's perfectly in line with my critique of the deal in 2015, which was pretty cynical: this is a sort of cost center that's going to benefit the other companies. That seems to be what's happening with xAI in many respects.

Right.

I think for Altman it was, "Oh, here's a shortcut where I can get the people that I want." I think I mentioned that. I don't know. No one cares about me cataloging what I wrote or did not write.

Well, I can't wait to go back and reread your take in 2015. We have to find that. If we do, we'll put it in the show notes for sure.

Well, someone's not reading Stratechery takes closely. I literally do it all the time. But the—

Oh, wow. Okay. Just busted here—

That's the way it is. I know.

—live on the podcast.

No, there's a good anecdote.

I was on the podcast with Ezra Klein about a decade ago or something, when he first started his podcast and had to scrounge around for people who would just be willing to do a podcast, as opposed to being Mr. New York Times or whatever it might be.

Uh-huh.

One of the things that stuck in my mind from that podcast was that we had just had a general discussion about being bloggers and writing, and one of the important things you learn is that no one remembers anything. He told this anecdote about someone who emailed him, or maybe tweeted him, saying, “Oh, what a great article,” and then, I think the next day, he reposted the link.

Mm-hmm.

The same person responded, saying, “Oh, that’s a really good insight.” “Great article.” They totally forgot not only that they had read it the day before, but that they had already complimented him on it the day before. This is totally the case, and it’s something I try to avoid: more shamelessly repeating takes that I’ve had because no one remembers them, even though I feel bad about it.

No one gives a shit. Yep.

No one really gives a shit about me distinguishing, “Oh, hey, sentence 1 in paragraph 3 was correct—”

In 2015.

“...but sentence 2 in paragraph 4...” But it’s a real challenge because I personally am obviously obsessed with it. I think about it all the time, so sorry when it spills over to the podcast.

Well, listen, thank you so much for sharing your neuroses with me and the audience every week. We have come to love it. I have come to love it. And I love Vikesh’s email because the core of the email comes down to whether you think the best AI researchers in the world and the best AI product people in the world were all so crazy that they only ever would have innovated and pushed forward at a company that was wrapped in the utopian ideology that—

Right.

Sam was selling.

Arguably, OpenAI’s formation actually set a very bad precedent and also broke a social norm. That was my point of the Chesterton defense. You didn’t do a for-profit or a not-for-profit in Silicon Valley before.

Yeah.

But now we have all these wacky companies with wacky mission statements that are not necessarily governed by the profit incentive. I think the more powerful the technology, the more I want it governed by the profit incentive.

Yeah.

That is a limiting factor.

Governed by something rational, absolutely.

That’s right, instead of, “Oh, we promise we’re going to do our best.”

And it’s going to be our idea of what’s safe.

Not so high. That’s right.

Yeah.

Yeah, exactly.

7. Cloudflare Rewrites Web Rules

Exactly. Well, to keep it moving, I’d like to wish a belated happy Content Independence Day to you, Ben, and to everyone out there in the audience. Cloudflare declared July 1 Content Independence Day this month, and I’ll read from The New York Times for a little bit more about what they’re doing.

“Cloudflare,” the Times writes, “a tech company that helps websites secure and manage their internet traffic, said on Tuesday that it had rolled out a new permission-based setting that allows customers to automatically block artificial intelligence companies from collecting their digital data, a move that has implications for publishers and the race to build AI. With Cloudflare’s new setting, websites can block, by default, online bots that scrape their data, requiring the website owner to grant access for a bot to collect the content, the company said. In the past, those whom Cloudflare did not flag as a hacker or malicious actor could get through to a website to gather its information.”

“‘We’re changing the rules of the internet across all of Cloudflare,’ said Matthew Prince, the chief executive of the company. ‘If you’re a robot, now you have to go on the toll road in order to get the content of all these publishers.’”

“Cloudflare is making the change to protect original content on the internet,” Mr. Prince said. “If AI companies freely use data from various websites without permission or payment, people will be discouraged from creating new digital content,” he said.

So, Ben, for any normies in the audience, can you give us a CliffsNotes summary of what Cloudflare is and why they have the ability to institute a change like this? And then, secondarily but more importantly, what do you think about what they’re trying to do to protect websites from being crawled for AI training? Start there, and then we can talk through the pay-per-crawl model separately.

I do have a little bit of a bone to pick with this Content Independence Day thing, which is tied to regular Independence Day—

Mm.

—which always occurs during the Stratechery summer break.

Yep.

Did we talk about this on the podcast previously? I think maybe not. I had sworn I wrote about this a year ago, when they first made it trivial to block AI trackers.

Yeah.

I searched and searched and read through all my July posts from last year. I couldn’t find it. I think it was one of those things where it happened during summer break. I talked about it a lot with people, and then I didn’t write about it.

You wrote a mental article—

Yes.

—or recorded an unpublished podcast on last year’s Cloudflare changes. I’m glad you wrote about it this week, because we had gotten a lot of messages about it over break, and there was not a natural place to address it. So let’s talk through it now.

Yeah. So, Cloudflare—what is Cloudflare? Cloudflare is unique in that they’re a cloud service, but they don’t have a central data center like an AWS or an Azure. They own some data centers all over the world. They’re mostly co-located with ISPs.

They build this sort of network; they’re all on the edge. The edge of the internet is—your phone or your computer is the ultimate edge device. The next bit of interaction with your phone is a human finger, right?

Yeah.

The packets stop at the screen. The closer you are to that, the more on the edge you are, as opposed to being a hyperscaler that’s at the core, the center of the internet. That’s just broadly speaking. Cloudflare operates at the edge. They have all these servers.

If you have Spectrum internet, in Spectrum’s server room or whatever here in Madison, there’s probably a Cloudflare server. That is—

Okay.

So what they can do by being on the edge is absorb all this excess traffic capability because they’re already sort of on the edge. They can absorb a lot of traffic. For example, this is where they got a lot of penetration: distributed denial-of-service attacks, where you have a gazillion requests all directed, say, at stratechery.com. I personally don’t use Cloudflare, but if you got a sort of attack, they can absorb all that with their excess capacity, and normal people can still reach stratechery.com as normal.

It keeps those hackers from being able to just take websites down on a whim. Is that right?

That’s right. Part of this is that the natural counterpart of this is being a content delivery network. We’re not giving you the podcast from 1 server, or even from 1 website if you go to stratechery.com. That’s actually cached locally close to you, so you can download it easily, and we’re not paying a fortune in bandwidth fees and trying to upload it from Virginia or wherever it’s hosted.

Mm-hmm.

A content delivery network is a very standard sort of thing. It’s how you scale a website. Stratechery could have no one on it or it could have 1 million people on it, and it’s fine because the webpage is actually distributed out to the edge and delivered.

This was the natural original business model for Cloudflare, where they got a ton of traction by offering this protection for websites. You could sign up for free, and then they could monetize, and they had the content delivery network. It was a real win-win from an ISP perspective. Why would you let Cloudflare be there? They’re providing some security benefits, but they also have to pay transit fees when stuff is being passed back and forth. If all that stuff is just hosted there with them—

Mm-hmm.

—it actually makes their operations simpler, more secure, cheaper, all those sorts of things. There’s a real win-win situation.

Over time, because Cloudflare has this interesting architecture, they’ve come up with all sorts of new products, whether that’s VPN-type products or cloud-hosting products. In a world where countries start really enforcing that data has to be local, that’s tough for Amazon or something like that, because they’re used to the internet being the whole market, right?

Yeah.

Cloudflare is like, “That’s fine. We have local servers all over the place. We’re highly distributed, and it’s very trivial. It’s just a software problem for us—”

Let us manage it

To solve that.

Yeah.

Yeah, so it’s an interesting business. I’ve written a couple of in-depth articles about them in 2021 that I think still hold true. We can link those in the show notes.

The long and short of it is that, by flipping where they’re located—instead of being in the center, they’re at the edge—they could do a lot of stuff for “free” that was good for their business, a win-win for ISPs and customers, and helped them get a ton of traction. The idea that I, as an individual website owner, can have the best possible protection from a denial-of-service attack for free—sure, I’ll sign up for that. Why not? It was implicit in the way they were structured.

The net result is that a huge number of websites have Cloudflare as a layer above them.

Right. It’s the middle layer between users and the servers, the ISPs.

Right. Exactly. For the purpose of this discussion, that’s exactly right. That’s what matters.

Mm-hmm.

8. Cloudflare Blocks AI Crawlers

A year ago, they said, “It’s not right that these AI crawlers are going out there, taking everyone’s content, and not compensating them. We should make it easy for folks.”

Yes, you can go into your robots.txt file and specify, “OpenAI, don’t crawl me. Anthropic, don’t crawl me,” whatever it might be. But, number one, most people don’t do that, don’t want to do that, or don’t know how to do that.

Or have no idea how to do it, yeah. Exactly.

And number two, robots.txt is an honor system. The crawler has to decide to follow it.

Mm-hmm.

They can just ignore it and still surf your site. Cloudflare said, “We can solve this.” This was a year ago. “We can—” I think they called it literally the big easy button or something like that, where you press it and you don’t have to edit robots.txt or anything like that. You’re just going to block these crawlers.

Yep.

And number two, we can actually enforce the block. Because we sit at this layer where everything passes through, it’s not reliant on the honor system, where they read the robots.txt that says, “Do not crawl.” They’re like, “Okay, we got you. You don’t want us to crawl it, we won’t crawl you.”

There was a bit of a controversy in spring 2024 about Perplexity violating robots.txt, where they were crawling sites anyway even though the sites said not to. People were justifiably upset about that. It ties to our startup thing: that’s a social norm that governs the web, and you’re violating it.

I think the defense of Perplexity was, “Well, Google gets everything. We’re trying to—do you want a competitor for Google or not?”

Right. And Google gets everything because Google has to index every website that appears in Google Search results.

Well, Google respects robots.txt. You can put it on your website and say, “Google, don’t crawl me.”

Right.

Then you’re not in Google.

Then you don’t really have a website.

That’s right.

You certainly don’t have a business.

Whereas Perplexity has a small user base—

A drop in the bucket.

—it doesn’t matter anyway, right?

Yeah.

And so you’re creating the conditions for Google to be eternally dominant because no one can afford to cut Google off, while Google can afford to cut everyone else off. This was my critique of Cloudflare’s big easy button a year ago.

At the end of the day, all you’re doing is helping Google. Now, Google—this is a super-important technical distinction—technically has one bot, but it’s classified as two bots.

Okay.

Bot number one is called Googlebot, and you can use robots.txt to tell Googlebot what to do: surf your site or don’t surf your site. They also have one called Google-Extended. Google-Extended captures information for Gemini specifically and for its training, so you can opt out of your website being included in Gemini’s training going forward. Of course, they’re not going to give you back what they scraped previously.

Yeah.

You can exclude Gemini. The key thing, however, is that according to Google, things like AI Overviews or AI Mode—all the stuff that’s under the search umbrella—are governed by Googlebot policies. They say, “That’s a search product.”

What that means is that you can block Google-Extended, but you’re not actually stopping the collection of your data for AI because Googlebot is still collecting it—

Ah.

—and using it for AI.

Yeah.

My concern is that you’re actually just favoring Google, because no one can afford to cut Google off and you’re making it trivial to cut everyone else off. Then they doubled down this year and said, “Actually, instead of you having to press that big button, we’re turning it on for everyone—”

Right.

—“and you have to voluntarily go in and turn it off.”

And that’s a big shift. The idea that you now have to affirmatively opt in to being crawled—and the numbers I’ve seen are about 20% of the internet’s websites working with Cloudflare—yeah, I don’t know whether this is a good thing because of the concerns about ultimately deepening Google’s moat at the expense of anyone else who might theoretically compete with Google in the future.

I really enjoyed thinking and writing about this article because I had leveled the critique at the political dilettantes—what’s the word? Dilettantes? There’s something along those lines.

Dilettantes?

Dilettantes. “I don’t like that person. I don’t like this person. I’m just not going to vote for anyone. It’s all corrupt. It’s all terrible.” We have a friend that you know exactly what I’m talking about.

Yeah.

He fits perfectly, right?

He’s listening. He knows exactly who he is.

That’s right.

Yes.

You have to make a choice.

Very frustrating to talk about these things.

You have to make a choice. At the end of the day, it’s a binary choice, and this action by Cloudflare really forced me to face this head-on.

I go back a year. Again, I didn’t write about it, and this reaffirms the value of writing for me. I had an initial response a year ago that I didn’t like this because of the favoring-Google angle. This time, my initial response was to be very critical for the same reasons: it’s even worse by default, and look at them exercising their power. They’re abusing their position in the stack. That was my initial reaction.

When you write, you have to dig in and think through the implications of your position. A few months ago, I wrote a piece about the agentic web and the original sin, and—

Yeah.

—people who were mad about advertising on the internet missed the boat, I think. On what I called the human internet, advertising was amazing. It’s easy to see the downside, but there are so many upsides: the explosion of content and all the things we can do, all for free, because it’s a win-win-win. You get all this free content, these services get to make more and more money without raising prices because the advertisers are paying for it, and advertisers get to reach customers they never would have been able to reach otherwise.

Again, as I’ve argued, I think Meta is the best manifestation of this, in part because you’re finding things through Instagram ads that you never would have found otherwise that are actually pretty great, right?

Mm-hmm.

People think about all the spammy ads on websites and focus on the worst parts of it, but with everything there are pluses and minuses. In general, I think the advertising-based web has been a big win.

But with the AI stuff, we talk about AI with Google all the time. If you’re just getting the answer, you’re not clicking the link to an ad.

Yeah.

And over time, if you’re doing deep research, the ads are immaterial. You’re not going to be swayed by an ad, right? If you’re an agent, there’s some aspect of ads that appeals to your humanity, but—

9. The Web Needs New Incentives

There are all these problems going forward about how you actually monetize content and how you incentivize the creation of valuable content that, at the end of the day, the AI needs, right? It needs current events and things that are happening to give good answers. What’s a new system that makes sense?

I put forward that we need to move to some sort of—I talk about stablecoins and being able to monetize content and creating a new marketplace for content, where super-valuable stuff to AI can get compensated more and you’re creating incentives for people to do this.

Instead of having to go to Scale AI and say, “Oh, go out and hire people to do this,” you—

Right.

If you can create a marketplace where individual actors are motivated to create because they can make money from it, that is going to scale way more. It’s going to be broadly beneficial, and I painted this beautiful picture of this future internet with a completely new incentive structure that works for agents and for AI being the primary customers of a website, and all those sorts of things. That’s basically what Cloudflare is trying to force to happen.

Yeah. I mean—

And so I had to sit down and say, “Wait.”

They’re forcing you to put your money where your mouth is on this one.

Exactly. Exactly. I felt like such a hypocrite for my initial response, which was, “It’s really easy as a commentator to point to what you want to happen and then be upset about the means by which it happens.”

Mm-hmm.

The reality is that things don’t change without assertions of power, because you’re upsetting the current system as it is. This is an assertion of power by Cloudflare. It is, from a certain perspective, an abuse of their position.

Yeah.

That’s how things change. If they’re actually driving toward what I think is important, necessary, and potentially very lucrative and beneficial to humanity, I have to be honest with myself and appreciate that if you want to get to something different, you’re going to have to crack a few eggs in the process.

Hmm.

So, ultimately, that’s why it was actually—

Let’s get to that microtransaction omelet on the other side.

Yeah. Exactly. So it was actually—I mean, I’m a sicko, so I enjoy being tortured like this, or self-torturing, as I was writing this.

Yeah.

But it was actually really interesting for me from a personal process perspective. Maybe it’s good I didn’t write about it a year ago, because I would’ve had to retract my argument about how this is bad and say, “Actually, no, this is right.” I think it’s interesting. I think it’s really intriguing.

I think Cloudflare would push back that they’re favoring Google. I think they are. I kind of reject that pushback. But I think their hope is that this will highlight Google’s conflation of search and AI, and it’ll sort of force action—that Google will have to split that up more neatly.

Yeah.

But overall, it is really quite interesting.

Yeah. Well, it’s an interesting time for the internet, because Matthew Prince, in the blog post announcing Content Independence Day, wrote, “The problem is, whether you create content to sell ads, sell subscriptions, or just know that people value what you’ve created, an AI-driven web doesn’t reward content creators the way the old search-driven web did. And that means the deal that Google made to take content in exchange for sending you traffic just doesn’t make sense anymore.”

The second prong of the Cloudflare plan here is to, by default, make it so that every website that runs on Cloudflare is blocking AI crawlers, and yet they’re also announcing this pay-per-crawl offering. Pratik says, “Andrew and Ben, what are your thoughts about Cloudflare’s pay-per-crawl offering? It seems like a great interim option while the copyright fair-use cases get litigated and/or laws get legislated.”

What you had proposed a couple of months ago and what Cloudflare is proposing now is a model that would allow these websites to be crawled, and then you get compensated with, like, 3 cents or whatever the number ends up being.

No. Much less than that, to be clear.

Not very much money.

No, but it’s all volume. The amount of times they hit Stratechery is frankly out of control. They need a little policing, but yeah.

Interesting. Okay.

The fractions of a cent could end up adding up to something meaningful?

That’s what I’m a little fuzzy on. If this is technically feasible, I have no objection to what Cloudflare is trying to do. I’m less convinced of the idea that this would materially change the fortunes and future prospects of websites and creators online that used to be ad-supported.

I think that ship has sailed at this point, and I don’t know that there’s a new model that would salvage things for the future. But I’m hopeful that something like that would emerge, and I think this is a worthy attempt, at least.

Well, the problem for an individual website is, if I were to exclude Stratechery from the crawlers, obviously it’d be dumber, but they’re not going to feel the dumbness, right?

It’d be so much dumber. Shed some tears for the LLMs in that scenario.

I thought that one—I was testing your level of how much you’re spacing out on the podcast by trying to slip that one by you.

Oh, my God.

It took you a couple of seconds, but you did catch it, so the latency is increasing, but you’re still connected.

There you go.

So it’s the collective-action problem.

Yeah.

What Cloudflare is doing is acting as a collective negotiating committee for all publishers and saying, “You might not feel the pain of any one, but if you get all of them…” And the AI companies do need this content. A world of one-to-one negotiations doesn’t scale to the long tail, where some of the most interesting stuff is. You do want to create this. It’s good for the AI companies if there is an incentive structure over time to create compelling new content. They don’t want to pay people to do this, right? That doesn’t scale.

The licensing deals are not sustainable—

That’s right.

—if that’s going to be the model for the next 40 years. Yeah.

Right. So there is a real long-term benefit to everyone to creating this marketplace of content creation and this new sort of thing. It’s hard to get there when you’re dealing with a bunch of disparate, small actors who have no market power.

That’s why it’s interesting: Cloudflare is basically acting as a single entity standing in for 20% of the internet, which actually—

Mm-hmm.

And just to mention the email from Pratik, I think he’s making the same mistake a lot of people are in conflating this with copyright. Or it speaks to why the whole copyright angle is totally wrong.

With search, the search trade-off—which, again, has gotten less and less lucrative for Google over time—but by and large, the trade-off was that Google was going to scrape your site. They were going to violate copyright by putting a snippet from your site in search results, a word-for-word copy, but you were going to click on it, then you were going to go to that site, and they could make money.

With AI Overviews, the actual problem is that it’s not violating copyright. The problem is that it’s rewriting your content and summarizing it. It’s derivative of your content.

This is why I think these copyright cases are losing. Both judges—the one who ruled that this is fair use, Judge Alsup, who was very clear that this is fair use, and Judge Gabriel, who was like, “Oh, this is really bad, but I need specific details,” and wanted to rule against it—

But I can’t just ignore the lack of evidence here. Yes.

Both of them were like, “The output is clearly not violating copyright. It is clearly transformative content.”

Yeah.

This speaks to how everyone on the copyright angle is totally missing the boat. The problem with AI search overviews is that they are transformative, and because they’re transformative, you’re getting a 3-sentence summary, and you have no need to click through to the article.

Mm-hmm.

Whereas when Google was posting 3 sentences that were an exact copy, that actually increased the likelihood that you would click through to the article.

Right. Yeah.

It speaks to why I think the whole copyright-pursuit angle is a total mistake. It’s totally misunderstanding what the threat is, what the issue is, and why this is a problem for publishers. The actual problem is the transformative nature, which is why copyright law is not the right angle to protect publisher interests.

Well, I would say that the similarities between the conversation we had a week ago about copyright law and some of the considerations that inform that debate—we’re talking about the future market of content, whether it’s books or journalism or whatever these LLMs are training on—and this is another case where the concern is rooted in the future economy of the internet.

That’s where, for me, I come back to the fact that the internet was at its best when people thought that the ad-supported ecosystem was going to scale, and VCs were funding websites all over the internet that were doing ambitious work. Those days are just over at this point, because everybody realized there’s a better way to serve ads than what they were doing in 2007 and 2008.

I just don’t know how robust any of this is actually going to be, no matter what model they come up with. But good luck to Cloudflare trying to make it work.

To your point, no, that's not what you want. You want a marketplace.

But we've had something like a marketplace for the last 10 years, and there hasn't been enough money there to make it viable for most people. I mean, if you're not The New York Times—

Feels pretty viable to me.

Well, a subscription model is kind of a separate situation.

No, the whole point, though, is the subscription model. When I started, it was unique; there wasn't a Substack. The point is, you need to figure out the right business model. It's a business model issue. The whole point of this thing is trying to create new business models, creating the conditions for a new business model to emerge.

Mm-hmm.

Ideally, if you want something that is going to scale from small to large, it has to be a marketplace. It can't be a top-down imposition. But the bet here is that via a top-down imposition, you get to a marketplace, so I guess there's a little bit of conflation there.

You can envision this world. You can envision a world where you're compensated, and the more creative, unique, and differentiated you are, actually, the more money you make because you get paid more by the AI crawlers. In one of the Cloudflare posts, they talk about imagining your deep-research report: You give it a budget so it can go and pay more for better resources and things along those lines.

There are lots of issues with that. It gets into why I'm generally anti-microtransactions, where upfront payment decisions for unknown output—

If it's not guaranteed, yeah, exactly.

Yeah, it's a little iffy in the consumer psyche and things along those lines. But by and large, the general outline—and this is the case with lots of things—is that you can see how this could work. The question is, how do you get from here to there? How do you get from the way things are to that future world? You have to break some eggs sometimes to do it.

Sure. I can envision that world. I don't necessarily envision that world if I'm sitting here today, but I do have the capacity to imagine that being possible at some point.

10. Publishers Must Build Destinations

One final question related to this. Eric says, “I found an exchange on X with CEO Matthew Prince that touched on the concerns around deepening Google's moat. Mike Chong asked Matthew Prince, ‘Is it possible to block Google because they are also using AI-generated content in search results now?’”

Prince responded, “Google will need to respect how content is licensed. That may involve splitting up their crawler. I'm optimistic they'll do that for the benefit of the ecosystem. Otherwise, we have a number of other ways to force them to, but hopefully it doesn't come to that.”

And Eric says, “Ben, I'm curious what you make of this statement from Prince. Does Google have any incentive to play nice here? Does Cloudflare have the leverage to ‘force Google’ if Google doesn't play ball?” What do you think?

I don't think Google has any incentive to play nice here. I think this is a huge advantage for them.

Yeah.

I know Cloudflare would push back on this, but it's hard for me to see any way that this isn't. This is where it does kill me to not have written about this a year ago, because I think immediately this was my thought: This benefits Google, before it became an obvious thing.

It's like—see, I'm having neuroses again. You're just going to have to take my word for it.

The theme today.

I love it.

I have the text messages, okay? Does Cloudflare have the leverage to force Google? They could just block Googlebot, but then that's hurting their websites because they're not getting indexed by Google.

Right.

My guess is this is more of a regulatory thing, like regulators being tuned in to which bots are producing search results and are actually violating copyright, strictly speaking, because they're taking exact content, versus which bots are summarizing sites and getting derivative content. That could be a distinction that's drawn.

For the record, part of the reason that this bugs me a tiny bit is there's a little bit of Yelp-ification going on here.

Oh, boy.

Yelp built this company through organic search results, and then has basically been on a 12-year whine fest that they now have to pay to access those customers, or get them to go to Yelp.com instead of just being the top result in an app for free.

Google competed. That's terrible. Google can't compete with us. It's unfair. There is a bit of this from publishers. From the beginning, and I think partly because I'm in this space, I'm just way more critical of everyone else in this space. But you need to build a brand people go to.

Right.

A destination site. If you're dependent on Google, that's a you problem, right?

Well, this is why I look at the subscription business as an independent conversation, because I think people who are in that business and thriving have learned that lesson and internalized it at this point.

I know, but what right do you have? Why is it your God-given right that you get a reader who doesn't know or care who you are, who just clicked a link in Google, and you deserve to make money from that?

And you get to dictate the financial terms and whether you're compensated on the AI side.

Right, and you get to tell Google how to search, or get regulators to come in and protect your model that you can't be bothered to differentiate or do your own customer acquisition. When distribution's free, there is a little bit of entitlement here: I deserve a Google link without any cost. That's kind of what the demand here is.

Well, it's all about perspective.

Google's thing is—

Because I think people who started a website look at Google like it's a utility company, like it's a highway, and Google looks at it like it's a business. It's an entitlement because basically anybody who's started a website in the last 20 years has viewed Google as a fact of life, something that's always there.

Were you the one who sent me the tweet about how millennial writers have the most twisted view of the world because they had the good fortune of coming of age in this golden era of—

That's what I'm referring to: the VC money from 15 years ago.

Yeah.

Yeah, exactly. It hasn't—

Yeah.

It hasn't always been like that, and there's been a rude awakening for the last 7 or 8 years in terms of what the actual economics look like.

Yeah, sorry that business realities exist.

Yeah.

Right, exactly. This is one of my positions that drives people absolutely bonkers. One is me loving Facebook ads, and the other is defending Google's right to—

Google.

—look out for their own interests. Yeah.

Over there in your monopoly hat and the monocle. It's killing me.

That's right. If Google wants to say—

It's driving me up the wall.

“Look, the price of you being in search is that we get to harvest your content for AI Overviews,” why is that not their prerogative? It's very easy. If you don't want Google to use your data for AI Overviews, then block Google's bot. But then my business would die. Well, then it sounds like you need Google.

You should accept their terms.

Yeah, exactly.

Just making everyone upset on this beautiful summer day.

No, but the reason I read that is because we referenced top-down impositions on Cloudflare's part earlier, and I do think that there's going to be some top-down imposition on Google's part if, in fact, Cloudflare tries to force the issue in one way or another. But time will tell.

Either way, I think it's worth a shot from Cloudflare's standpoint, and I wish them luck. We'll see how it all unfolds.

So we've gone too far on Cloudflare and the acquisition ecosystem. We're just going to do a couple of questions at the end here. First, this is from Yi Wen. He says, “Andrew and Ben, longtime listener, love the show. Andrew, belated congrats on the arrival of your second child. My question is basically: How and why did you decide to have a second kid?”

Wow, talk about acquisitions. Yes.

It was. What an acquisition Rose has been. The context for this question is that I welcomed my first child to the world last year, and ever since then I have noticed that most families seem to have 2 kids these days, if not more.

I grew up in the one-child policy era of China and have no idea what I missed out on by not having siblings. Now that I’m a parent, I wonder what my son would miss out on if he doesn’t have a sibling. Do you have any thoughts?

It’s a great question. If I’m being completely honest, Ben, my answer is not particularly thoughtful. My wife wanted a second child, and by the time the topic was broached, it was about a year after our first son, Charles, was born. I loved him so much that I figured, “Well, we can afford this, and I absolutely love Charles, so why not just multiply all the experiences we’re having with Charles and run it back?”

I got a couple of weeks into Rose’s life and I was like, “Man, it kind of sucks to start at the bottom of the totem pole again and do all this over again.” But beyond the initial shock of that experience, it’s been really cool, and I think it will only get cooler. You’re somebody who’s been raising two kids for quite some time. Do you have any takes for Yi Wen here?

Well, my take generally is that kids are amazing, and I really wish we had had more.

Mm-hmm.

I would have had three, four, five. Especially as it’s approaching the end, it’s like, this is so great. I love it. I think from a sibling perspective, I’ve always had siblings, so I can’t speak to the only-child perspective. I do think there’s a bit where my son and daughter are very close.

Yeah.

I know for the rest of their lives they’re going to be very close, and they will always have each other. I think this ties into my general philosophy of parenting, which is that I think most parents do too much and are too invested. They have to be the playmates for their kids and be involved in every aspect of their lives.

I think, in general, it kind of ties to old people trying to be cool and stay—

Mm-hmm.

—young forever, a point I’ve raised on here. That’s the next generation. You’re their parent. You’re not their friend. And by the way, if you want to be friends with them, you do that by being a good parent, and when they’re adults, then you can be friends. In the meantime, I see so many people try to be friends with their kids and reason with them and so on, and it just leads to really bratty kids in general and a lack of discipline.

And by the way, they’re probably going to reject you and not like you when they’re adults, right? It’s—

You’ll have problems—

—it’s doing it completely backwards.

—by the time you get to the point where a friendship actually makes sense, 25 or 30 years old.

That’s right.

Yeah. Okay.

I actually just saw a tweet today, right when we were about to record, so I’m not even sure if I texted it to you. It was something along the lines of someone who observed the difference between people who like having more kids and people who really struggle is the extent to which you care what your kid thinks when you tell them what to do.

Hm.

A big part of raising kids is that they don’t have discipline and they don’t have control. You’re imposing that for them so they can function and have a good life. As they grow up and you’re no longer there, they’ve developed their own discipline and their ability to work hard and control themselves.

I see so many parents who treat their kids like little adults and try to reason with them. They try to make sure they’re happy and all these sorts of things, and it’s such a burden on the parent. It is total nonsense.

The Socratic method—

And it’s not what the kids need.

—with your children is really not what’s up.

They’re not fully developed. They need guidelines. They need guardrails. And does that mean you’re going to be a controlling force in their life forever? No. This is actually the way to make them free adults in the long run, because they will develop over time the self-discipline, self-governance, and executive function to operate in the world by modeling and learning from what you did for them.

I’m so critical of this entire generation of parents that is anti-discipline. It’s so funny. You see kids that are just terrors.

Mm-hmm.

And then you see their parents interact with them and it’s like—

Pretty easy to connect the dots.

And by the way—

Sure.

—these parents are super judgmental about other folks and how they take care of their kids.

Mm-hmm.

And it’s like, guess what? Your kid’s going to be upset at you sometimes, because if they had their way, they’d eat nothing but Oreos and stay up until 2:00 in the morning. The fact that you tell them no and they’re upset doesn’t mean you failed as a parent. That means you’re actually doing your job and providing—

—and preventing them from being brain-dead, fat, and diabetic by the time they’re 17 years old.

Right.

Sure.

Do you want them to make good decisions when they’re older? You don’t do that by convincing them as a 4-year-old to make good decisions. You do that by habituating them to a good life.

Mm-hmm.

And a good life is one with executive function, discipline, and self-control. You instill that through the way you live and allow them to live when you have the power to impose it. Then they will grow up and be independent adults who can function and govern themselves.

I don’t tell my daughter what to do at all. She’s 18 years old. She’s out figuring stuff out. Of course, there are things that I disagree with or wish she wouldn’t do, but you know what? My chance to instill discipline and a way to view life was when she was 0 to 15, or whatever it might be.

Yeah.

And now she’s out there. By the way, I think my daughter and I are going to be very good friends.

Well, and in the interim, she’s had your son for companionship and friendship, and that’s—

Yes, exactly. Thank you.

—the point, to tie it back to the question.

You finished my point. I got on my rant. I got on my soapbox and forgot about the point. No, that’s exactly it.

Her friend in the household is my son. They can go in a corner and complain about Dad and Mom, which—

Uh-huh.

—that’s an important part of friendship, right? And you—

And God knows you earn it. I hope—

That’s right.

—they avail themselves of that corner.

I do wonder if there’s an issue with single kids exacerbating this, because if the single kid feels bad and wants to complain, who does he go complain to? If Dad’s being mean—

Mom.

—then he goes to Mom.

Yeah.

That’s a really dysfunctional sort of situation, where Dad and Mom are split and the kid is playing one off the other. I’m glad you mentioned that. I’m not sure this fully occurred to me, but it probably is easier to have discipline. It’s easier to not feel like you have to fill every moment of your kid’s life when they have other kids there.

Are they going to fight? Are they getting along? Is it hard at the beginning? For sure. But I definitely think, number one, it’s hard to articulate—particularly the older you get and the more successful you are—the extent to which kids are a motivation for you to keep working and keep doing things.

Yeah.

There’s a bit where you want to instill intrinsic motivation in your kids. When you’re a kid, you have extrinsic motivation, which is your parents telling you what to do. Ideally, that habituates you to have good habits, executive function, and all these things, so intrinsically you go out and work hard and do well.

But then you almost come full circle, where you want to keep working and doing well and creating things because it benefits your kids. It’s for the good of your kids. I think that’s the way things are meant to function.

Mm-hmm.

I think you mentioned we could afford another kid. Everyone can afford another kid. You can figure it out. And—

That’s the thing. I took the “we can figure it out” approach when we were considering the second kid. It’s funny—until you mentioned it, I forgot, but I had heard you say, “Man, I wish I had a third kid.” I thought to myself, “Well, if Ben wishes he had a third, I can definitely manage a second.”

I don’t know whether we’ll cross the third-kid bridge, but at the same time—

I’m going to pressure you into it. I’m texting Alice right now.

It’s like—yeah, exactly. Let’s get it done. In any event, it is a good question, and I love having my daughter, particularly as the months pass here and she comes alive. It has been really, really cool.

I think watching her relationship with my son is going to be probably the coolest thing in my life for the next 15 or 20 years.

Oh, it’s my favorite for sure. It is amazing. Not to say it’s always easy—kids fight, and sometimes siblings just end up not getting along.

Yeah.

I do wonder how often siblings not getting along is some sort of function of the overall structure and organization of the family maybe not being what it should be. I actually think parents versus kids is a healthy dynamic.

Mm.

Like, you think about it: in the long run, do you want your kids kissing up to authority and trying to play people off each other? Or—

Yeah.

—standing in solidarity and rebelling together, right? It's actually this whole drive to be friends with your kids—

You want your kids to be the Rebel Alliance, and Mom and Dad are the Empire.

That's right.

And over time, that is character-building for those kids.

Well, that's what actually lets them think for themselves. You are set up as a safe counterparty for them to rebel against.

Yeah.

Right? And it's like a standard to set. It's like the whole drinking thing, right? Do you want your kids to be super strict and never have a drop of alcohol, and then have them go, like me—that was me growing up—to the University of Wisconsin, where you're on your hands and knees in the middle of Park Street picking your guts out? That's not the best way to do it.

Go wild after big road games.

No, you want to—there's a whole bit about how the whole family structure should actually be preparing your kids for the real world. In the real world, guess what? They don't have everyone catering to them all the time. They don't have everyone telling them they're perfect when they're not all the time. They don't have everyone sitting down and reasoning with them all the time.

Sometimes you're told stuff you don't like. Sometimes you're criticized and you don't like it. Sometimes you're told no, and you don't like it. Is it better to learn the coping skills out in the real world as a 25-year-old, when you're competing against kids who were actually raised right, or is it better to learn it as a 5-year-old, when, by the way—

Mm.

—you don't know any better, and Mom and Dad are everything in your life? Like, people—

It's certainly easier to learn at 5 years old than at 25 or 30.

And people are denying their kids this easy opportunity—yeah, and so kids grow up having been catered to their whole life, being little brats, and they learn the lesson in a much more painful and difficult way in their 20s, if they ever learn it at all.

This whole concept—it hurts the parents. Parents who complain, "I have no time. I'm so worn out," blah, blah, blah—stop catering to your kids. Let them learn to be independent. People do too much, and in the long run, it's like you wasted all your good time with your kids with a bunch of 5-year-olds.

Yeah.

Wouldn't you rather instill great qualities in great adults, and they come back, and then you sit down as adults and you're friends and you have conversations? Even with my daughter—she's only 18—we talk about politics, and we talk about sports, and all these sorts of things. It's like I'm talking to another adult. We text all the time. It's great.

Now I have a friend. I have a friend by virtue of not trying to artificially create a friend for the first 15 years of her life.

There you go. Wonderful takes. Always love a good TikTok segment, so thank you, Yiwen. I'm also proud of my deft Star Wars analogy, which I would not have been able to pull off had I not watched Andor. I've never really been much of a Star Wars guy, but that's neither here nor there.

To have more kids, I think, is the takeaway from all of this, and discipline those kids early on in life so that you can reap the rewards later on.

But by discipline, we're not saying—yeah, just treat them as kids.

Yeah.

They're not your friend.

Yeah, no, exactly. And I think what you're laying out is better for the kids. It's also better for the parents over the first 18 years.

Oh, it makes it so much easier to be a parent.

Yeah.

Right?

Exactly.

Just, if you tell them to do something, they do it. There's not this endless negotiation and back-and-forth, and, oh.

You want to feel like you're really losing your mind, try to negotiate with a 2½-year-old, which is a mistake I occasionally make.

You're incapable.

But in any event, we've come too far to get to all the other mail we had. We'll roll that over to next week. Ben, go enjoy yourself for the rest of the week. Touch some grass this weekend, and we'll come back on the other side.

Talk to you later.