Sir Tim Berners-Lee doesn’t think AI will destroy the web

Today, I’m talking with a very special guest: Sir Tim Berners-Lee, the inventor of the World Wide Web. Tim is a legend in the history of the internet.

He created HTML, the standard language for creating and structuring web pages, and the HTTP protocol that allows those pages to communicate using browsers and servers. It doesn’t really get more foundational than that — Tim was there at the very, very beginning of the modern internet.

But right now, in a lot of ways, it feels like maybe we’re sitting at the end of that grand, world-changing project. Tim has been sounding the alarm about where the web has gone wrong for years now. You can go back and read headline after headline to see his increasingly dire warnings about what’s happened to life on the internet, from the concentration of power in Big Tech platforms to the detrimental effects of social media.

Now, Tim is not exactly a pessimist — you’ll hear in our conversation that he still has a lot of optimism about the web and what it can do. But he’s also concerned that we’ve strayed too far from his original vision of the web as a democratizing force for knowledge and creativity.

All of that plays a major thematic role in his new memoir, This is For Everyone, which is about the growth of the web and how he thinks we might be able to salvage its best parts and make something better. You’ll hear Tim explain that the title itself, coined as part of a segment he contributed to the Opening Ceremony of the 2012 Olympics in London, is kind of the purest distillation of what he’s always wanted the web to be. And he sincerely believes in it.

So Tim joined the show to talk about all of that, as well as his current work at the decentralization startup Inrupt which works on the open source Solid standard, and, of course, where AI fits into the conversation about the future of the web. Tim has for a long time been talking about an idea called the Semantic Web, or a web that’s readable and traversable by machines, and so you’ll hear him explain here why he’s excited about generative AI and in particular personalized assistants, including one he helped develop himself at Inrup called Charlie.

We’ve spent a great deal of our time here on Decoder over the last couple of years talking through the implications of AI for the open web generally, and more broadly how closed ecosystems have diminished the power of the web as an information platform, even as it’s become a rich application layer for the internet. Everywhere you look, though, AI is threatening the web in new and interesting ways. There’s the rise of Google’s AI-powered search results, the new browser wars happening between OpenAI and its competitors, and a full-on breakdown of the web’s social contract thanks to AI firms hungry for training data they’d rather not pay for.

So I really wanted to dig into this with Tim to see whether he believes the spirit in which he invented the web could somehow be reborn in the era we live in today. That vision was one where inventors, academics and the open source community collaborated with the tech industry to build something bigger than any one product or platform. And even though they may not have all agreed on what direction the web should take, they were all incentivized to join together on big initiatives like the W3C standards body Tim founded more than three decades ago.

Could something like that ever happen again, and could it happen for an AI-powered web? And is there a future where decentralization wrestles some power away from Big Tech and back to the end user? I think you’ll find Tim’s perspective here really insightful.

Okay: Sir Tim Berners-Lee. Here we go.

This interview has been lightly edited for length and clarity.

Sir Tim Berners-Lee, you are the inventor of the World Wide Web and co-founder and CTO of Inrupt. Welcome to Decoder.

You are also the author of a new memoir called This is For Everyone, which is about the web and the future of the web. I have a lot to talk to you about. I was just reflecting before we began this conversation that my entire career exists because of the web, and the democratic access to publishing that the web afforded me and millions of other people, and that feels like it’s all changing. The frame that I use when I talk to everybody about the web is that if we were starting The Verge today, I don’t know that we would start a website.

In 2011, when we started our site, it was the only choice — that was the thing that we were definitely going to do. We were going to start a big website with lots of functionality, and over the past 14 years, it feels like the thing we would do today is start a video channel on a closed platform, like a walled garden platform.

I just wanted to ask you from the start, do you see that shift, that new information is most often going into closed platforms versus the open web? And how do you feel about that broadly? Because that’s the shift that I think I feel the most.

I see a lot of things end up on YouTube, which is on the web. So, things in general, most of the things I’m involved in, they end up on the web, but some people, like The New York Times, are constantly saying, “Please download the app.” Or the BBC says, “Please use the app.” They always try to persuade you to use the app, because then they have more control, they can track you better. But also with podcasts particularly, I use a podcast program, a generic podcast app, that I can listen to any podcast with.

For me, that’s the web as it should be. You could send me a link to a podcast, or I can search for it and I can keep track of all the hundreds of podcasts I’m interested in. It is a bit like keeping bookmarks on the first original browser. So, to a certain extent, podcasts work well. But if people end up on the app and then are tracked and not using the podcast app, then not so. All that tension is huge right now for the web, yes.

There’s a reason I wanted to start there, because I think understanding how you feel about it will inform so many of the questions I have for you. There is a tension between the fact that YouTube is on the web, but all the videos on YouTube are not available in other players. There’s no ecosystem of open YouTube players that can access that database of videos.

Podcasts are really interesting. Apple effectively maintains a central database of podcasts, and all the podcast players use it for discovery. So, there’s a tension between “where does the centralization lie?” and “where does the open ecosystem live?” — and what’s the relation between the two of them? To me, it feels like the centralized players always end up exerting more and more dominance over the ecosystem. Maybe that has been inexorable over the past few years, but I’m curious how you see it.

When you have a market and a network, then you end up with monopolies. That’s the way markets work. There was a time before Google Chrome was totally dominant, when there was a reasonable market for different browsers. Now Chrome is dominant. There was a time before Google Search came along, there were a number of search engines and so on, but now we have basically one search engine. We have basically one social network. We have basically one marketplace, which is a real problem for people.

It’s a problem for people like you who are trying to just make their way in the world – to be a journalist, be a commentator – and you want that feeling that you’re peers with everybody else. You want to have control of your own destiny. We call it digital sovereignty. In the old days, the early days of the web, anybody used to be able to make a website. So that feeling of sovereignty as an individual being enabled and being a peer with all the other people on the web, that is what we are still fighting for, and in fact, we need to rebuild.

This brings me to your book. It’s called This is For Everyone. That seems like the heart of the book — that you should have more agency as you operate on the internet, that you should be able to publish and consume as you wish. Explain how you came to that phrase and how you think about that phrase in the context of the internet that we all experience today.

That came from when I was given the chance to participate in the London Olympics. That was an amazing moment. I got an email from Danny Boyle saying, “Do you want to come into my office and let’s talk about the Olympics opening ceremony?” That was an amazing time.

I got to sit there and type a few words on an old NeXT computer on the stage in the stadium, and when I hit return then that phrase would go around the stadium in LED lights and then off over the internet. So, we decided that “this is for everyone” was a way of really encapsulating what was most important about the web.

When you think about “this for everyone” now, the argument I hear from so many of the platform CEOs who come onto the show is, “Yes, we are in control of a lot of things; yes, we are in control of a lot of data; yes, we feel like monopolies to a lot of people; but we’ve given so many more people tools to express themselves.”

TikTok might be a closed platform that doesn’t really play all that well with the web, but it’s given many, many millions of people more tools to express themselves and reach an audience. That’s their argument and I hear that argument virtually every day. I’m curious about your perspective on the opposite side of the argument, and whether that has actually closed down a version of the web, and has closed down a type of freedom or a type of digital sovereignty. How would you make the argument in response?

I think there are two things about TikTok, really. One is that it’s not really part of the web; it is an app. But the other is that when you get onto TikTok, the algorithms on TikTok are addictive. And so when they build, you know who you are, you’re building the TikTok back end, you’re building the UX. And you make it so that you optimize the AI, so that people will be kept on the platform. I’m not on TikTok because I get videos on YouTube.

On YouTube, I find that I can stop scrolling through; I don’t end up scrolling through forever. If I was on TikTok, I’d probably end up scrolling through them forever. When you have such great power, if you are a dominant player like TikTok, then you have a lot of responsibility. You have a responsibility to all the people on it to respect them.

At the beginning of the web, you made a lot of people with a lot of power agree to participate in some standards. You made them agree to participate in browser standards. There was a ferocious competition among browsers at that time.

It seems like we’re about to enter a period of ferocious competition in browsers again, which I want to come to. But at that time, at the beginning of the web, there was a lot of competition in browsers. Microsoft was brought to heel in an antitrust case, because there was such ferocious competition among browsers.

How did you go about convincing all of those companies and all of those people to adopt your standards and say, “We actually have to be good stewards of the collective”? Because that doesn’t seem like a thing that could happen today, but you were able to do it at the outset of the web. How’d you do it?

By persuading them that one web was going to be really good. If you have just one web, it would take off exponentially. If we had many little webs, they would each die. People realized that and they managed to persuade the governance within their platforms, their managers and their boards, to make this one web. They knew that if they fought over incompatible versions of HTML, then the web would not take off, like it would if they made one web. If they made one web, then one web would take off and become huge, and then their part of that web would be itself huge.

Could you make that argument today? And what technology would you make it about?

A lot of people would wonder whether you can make it with AI. There is no World Wide Web Consortium (W3C) for AI, which is bringing everybody together in one room. Some people have suggested there should be something like a CERN (The European Organization for Nuclear Research) for AI, some big high-energy physics lab in some big international lab that develops AI. That way you can really optimize both the development of AI, but also prevent the thing running away, so you can build containers around it, for example.

Could that be done? I’m just looking at the companies in the AI space. OpenAI seems unconstrained by any outside forces. Anthropic, I think, is a little more constrained, because it’s very enterprise-focused, but it’s also pushing hard. You can just go down the list. All the big AI companies, they don’t seem constrained or willing to accept constraints.

The personalities driving the development of AI are almost religiously idealistic in nature. Could you get them in a room and say, “You actually have to behave yourselves, and this will be bigger if we work together or constrain the capabilities”?

It’s really hard. I don’t see it happening.

One of the characters in AI now that people are obviously aware of is Marc Andreessen, from Andreessen Horowitz. He is one of the world’s richest people. He is driving a lot of AI development. He’s investing in a lot. He was there at the beginning of the web. He obviously created Mosaic, the browser, and then Netscape.

There was a profile of you in The New Yorker recently, and you said that you felt Andreessen had hijacked the internet, your creation, to do something it wasn’t intended to do, to commercialize it. Do you still see that dynamic playing out? How do you feel about Andreessen’s work today?

No, I don’t think they hijacked it. They did try, originally with Mosaic, to absolutely brand it as Mosaic — “you’ll find it on Mosaic,” not “you’ll find it on the web.” But they lost that battle the moment Microsoft came along, and there was a battle between Microsoft and Netscape, they lost the battle to brand it as Mosaic.

I’ve never been against the commercialization of the web. Advertising has been a really important part of it, and subscription models, and lots of different business models have always been important for the web.

The reason I spoke about Andreessen specifically is that he’s obviously a character now in a big way. He was a character at the beginning. He might understand that making one network, one AI network, just as we made one web, would create exponential growth across the ecosystem.

But I don’t see him making that argument. What do you think would convince some of these players, even the players who’ve had the history of how big the web got, to make that argument? What would tip them over into saying, “Actually, collaboration makes the market bigger”?

I think there are people like Demis Hassabis who have called for it. So some of those people, if you had a government-funded or philanthropically funded non-profit and you invited all of those people who’ve expressed a concern, then you could maybe put them together and build something like a CERN for AI. But I think it’s going to be a very tough job trying to get most of the players out there to slow down for a second.

Demis Hassabis is at Google DeepMind, and he has been on the show before. He is very thoughtful. I have heard him say out loud that he thinks the future of the web is agents going out on the web and doing things for you, and maybe the browser doesn’t have a visual component anymore.

He said that at Google I/O just this past year. I asked Google CEO Sundar Pichai about that comment. He said, “Well, maybe Demis was thinking too far ahead.” I see that version of the web where agentic browsers are going off and using web services for you. They’re browsing the web for you. They’re summarizing information or maybe even using applications for you. And the web itself changes because people are no longer using it, automated systems are using it. That has kicked off another browser war.

Just in the past week, I think three AI browsers were released: OpenAI has Atlas, Google announced some of these features in Chrome, there’s one from Opera, and so on. Do you see that new browser war as a source of innovation and excitement, as somebody who created the first set of browsers?

I think it is exciting and it is innovative. Using AI as your first port of call is exciting. Yes, I do worry about the infrastructure of the web when it comes to the stack of all the flow of data, which is produced by people who make their money from advertising. If nobody is actually following through the links, if people are not using search engines, they’re not actually using their websites, then we lose that flow of ad revenue. That whole model crumbles. I do worry about that.

If I was starting The Verge today, I would look at the monetization available to me on the web. I would look at Google saying that the open web is already in rapid decline. I would look at the monetization available to me on some closed platform like TikTok and say, “Well, that’s better.” I would put the information there.

The split that I see is that the web as an information platform is in decline. It just feels like there’s not new information on the web in the way that there are new ideas and new information on some of the closed platforms. But the web is an application platform, a place where you can deploy an application and get to millions of users outside the strictures of an app store and outside the strictures or whatever an operating system might let you do. That feels like it’s at an all-time high. That’s where all the action is.

That split is really interesting because AI reduces the value of the web as an application platform. It becomes the interface to lots and lots of applications. Maybe they end up without interfaces at all, and that also feels like it will change the web in huge ways. Do you see that split, and how do you think AI factors into the web as an application platform?

Yes, I do see that split, and as an application platform, I’m concerned if the AI does not feed the search engine, or doesn’t feed the blogs or the podcasts. To the extent that AI reads everything on the web and then helps you live your life by using that. One of the things I talked about in the book is that you need an AI that works for you. If it’s only running off the external data out there, if it doesn’t have access to your own personal data, AI won’t be able to do a good job helping you in your life.

At my company, Inrupt, we’ve built a version of AI, a thing called Charlie, that does have access to all your personal data in your data wallet. So an AI that does that is much more powerful. Without that ability to act on your behalf — without promising to work on your best interests, like your doctor or your lawyer, who take oaths and are culturally bound to support the best interests of the client, not of the person who pays them. I feel that AIs that work for you are going to be a really important part of the scene.

How do you think that plays out with the applications on the web, though? If I have an AI that’s working for me, and maybe my personal data is in a wallet using some of the protocols you’ve developed at Inrupt, sure, I get that.

But the AI has to go out onto the web and then if I want to order a sandwich, it has to go to DoorDash — I call this the “DoorDash Problem” — and then I, the customer, never talk to DoorDash. DoorDash has been disintermediated by an AI agent, and maybe DoorDash doesn’t make enough money and now we just have a bunch of commodity sandwich providers and they race to the bottom and go out of business.

How do you see that playing out? Where do you find the balance here? I’ve talked to a lot of these service provider companies over the past year and asked, “How do you solve this problem?” And all of them say, “Well, we’re more valuable than the next one, so of course we won’t get commodified,” which is basically what everyone always says. That’s basically the answer and then it doesn’t play out that way. How would you build that business?

Yes, the model is that at the moment, DoorDash has all that information about the things you’ve ordered on DoorDash, but DoorDash only knows the things that you’ve ordered on its own platform. Suppose your AI, as it did that, made a note of everything that is ordered from DoorDash, but it also had a note of all things it’s ordered for you from DoorDash one and also from DoorDash two, and from DoorDash three and so on, or DoorDash’s competitors.

Then, when you have a world where, in fact, the AI that is doing this work for you will be able to understand more about what it can recommend you buy, what you should eat if it’s got information in your data wallet from the various DoorDashes. It may be more inclined to collaborate with a version of DoorDash that synchronizes with your data wallet. Because sharing that information with the AI becomes more valuable, then there’s incentive in the system, there’s more value for the user if that data is available to the AI. If there’s more value to the user, then generally the markets find ways of arranging incentives.

Describe the architecture of this data wallet to me. So, I have a local store of data that my AI can access that doesn’t go out onto the web or any other system. But my AI is accessing it, and my AI is also running locally? Sketch out the system for me.

When an AI buys something, then at some point it uses something like Visa. At that point when it uses the payment platform, for example, there’s a way Inrupt has been talking to Visa about the future of agentic commerce. There are various architectures around where the payment provider can track what’s going on, what’s being bought, it can check the metadata about what’s being bought, and it can make that data available to the user, for example. There’s a privacy component, too, and a user empowerment component to an architecture in which companies like Visa work with your data wallet.

I was just putting that in comparison to some of the agentic browsers that I see today. The first wave of agentic browsers was all pretty cloud-based. Even Google’s first attempts were like, “We’re going to run Chrome in a data center for you and you have to give us all of your logins; we are going to take all of your logins up to the cloud.” This is pretty obviously a bad idea, but I think, given the AI workload that this involved, they had to do it that way at first.

Now we’re onto OpenAI shipping Atlas, which is an AI-powered browser. It’s going to run on your Mac, and just like Chrome on your Mac, all of your passwords and all of your data will be on your Mac inside of the browser ecosystem, which is a little more secure. Maybe it’s still not as secure as people want with cookies and the like, but so it goes. And then the agent will take action somehow in the browser.

It’s actually unclear to me where all the interference is happening, such that the browser can use itself. There’s some dynamic between OpenAI’s data centers and the browser itself. The next turn is bringing that all the way down, making all of that happen locally. That way, you are totally in control of your data and totally in control of whatever the AI is seeing, without it going to someone else’s data center. Where do you think we are on that journey? Where do you think the right place to end up is?

The right place to end up is where you’re totally in control of your data. As the power of local devices becomes greater, then the amount of stuff which gets stored locally increases… a lot of the things that people are building out there on the web are local-first, meaning that they’re designed to be able to work without the cloud. They store data either on the web, or they store the data on your local device. The good place to be is that the data is all stored on your local device. The inference is happening in your device, and you’re in control.

If that was the right outcome, how come it hasn’t happened in the previous versions of the web? That’s the dynamic that I’m really curious about. I agree that the right outcome is that users should be in control of their data, that we should be able to make our own choices about what’s happening with our data. That we should be able to see and control and restrict that data. The way that has been expressed so far may be effective, but it may not be. For example, cookie sheets in Europe, or Apple constantly prompting people on the iPhone.

But it doesn’t seem like users are actually demanding this. You do the surveys and consumers say they care a lot about privacy, and then they turn that off and they immediately open Instagram. There’s a real disconnect between what people say they want and what they actually look for in the market. And we haven’t gotten into a place in any version of the web so far where the data is private and it is local, because that’s actually what the market demanded. Why do you think that is?

I don’t think users demanded the cookie law. The cookie law should have allowed a lot of cookie treatment to be done without the pop-up. That was done at the wrong point, really. It was a badly written law.

But why do you think the market hasn’t gotten to a place in where users are more in control of their own data? Because we said that in the Web 2.0 era as well — that users should be in control of their own data. We should be able to build a massive application ecosystem that connects everybody’s data.

That was really the heart of the early part of Web 2.0; whatever it became is something different. But there was a big pushback then, too. Users should have a lot of data. You should be able to use a lot of applications and services in an intermixed way. It didn’t play out. Why do you think that didn’t play out?

There are some parts of our lives where it has played out. Like when you use Apple things, you store things in iCloud. Google things are stored in your Google Cloud; Microsoft stores things in its cloud. Each platform stores stuff, and you are in control of that data. One of the problems is that Apple-stored things can’t be stored in Google’s cloud.

But, for example, when you’re using Microsoft software, there’s a Dropbox integration. If you have a Dropbox account, from Microsoft’s point of view, obviously it’s not really worried about where it stores the data. In that area, you can elect to have your Microsoft data stored in Dropbox instead of Microsoft’s cloud stack. In that area, you are in control of your own data.

As somebody who’s reviewed a lot of consumer products, my view of it is that, while some of these things exist, they’re very idealistic. A company like Microsoft will announce a Dropbox storage integration, and then no one uses it because it’s vastly less convenient, or it’s harder to use.

Another example I’ll give you is if you are dead set on having the tightest security possible, you can keep all of your Apple data, you can keep your password, you don’t have to sync it to iCloud. And then Apple will tell you that the vast majority of people do this because they forget their passwords. They go to the Apple Store and they demand a password reset that the centralized authority has to do for them.

There is a real tension between “you should be in control” and “you should give up a lot of control,” because you need the convenience of somebody else being in charge. You can see this broadly in the IT space. Users just want an IT administrator to do a lot of things for them. In the AI space, that control carries a lot of risk. Someone else will let an AI agent make decisions on your behalf across the web. That’s very risky. Do you think that will be enough to push users to say, “Actually, I want less convenience and more control”?

It’s really hard to tell what users are going to do. I think people trust Apple with their passwords, and they trust iCloud to sync them to a certain extent. If Apple did that with open protocols, like if your Apple password sync with iCloud was done using [Inrupt’s] Solid login and Solid data protocols, then that would be open and interoperable. It’s not. They don’t have the interoperability between different platforms, but they do have the functionality, and they do want the functionality of having somebody in the cloud.

In the Solid ecosystem, you’re trusting somebody and you put a lot of stuff in your various data wallets, and you trust the data provider to do the right thing with it. To a certain extent, before and after, the difference isn’t that you trust a big company to look after your data or you don’t, as much as it’s that the data company that does it is doing it in a way that’s standard and interoperable.

Can you do that through market forces alone? Or do you actually need a big set of regulations from the governments around the world?

I cannot see it happening with market forces. Yuval Noah Harari has suggested that we should have regulations for interoperability. I think there’s been talk in Europe about interoperability through regulation.

There’s a big split across the Atlantic. Obviously, we see in Europe that there’s more push towards regulation, there’s more activity around regulating, especially American tech companies. I think we see that very clearly.

In the United States, lately under Trump, it has been very deregulatory, especially around AI. The idea that we shouldn’t regulate AI for a decade has come up over and over again in this country. Do you think there will be a meaningful split, or that European regulation can actually hold some AI companies in check?

It’s very hard to predict the future there. AI is changing so fast. One moment you are regulating AI taking over people’s jobs in one industry, the next moment you’re regulating different industries, and the next moment you’re worrying about AGI and superintelligence. Because AI is changing so fast, I think it’s really hard to tell which way things will go in that way. What do you think?

I think that European governments are very interested in controlling what American tech companies do to their economies, and they’ll pass some regulations. But I’m not sure that American tech companies will listen. I think this is the new danger — that American tech companies are barreling over various kinds of regulation.

Actually, the example I would give to you is the Semantic Web. You spent years of your life working on a Semantic Web, a machine-readable web. I think your rallying cry was to give us the raw data. Open the databases to browsers. Let us build new applications on top of your databases. Don’t make custom front ends. When I was young and I was reading your work on it, this was very exciting. The idea that we would make the world machine-readable and that would create a new class of applications, services, and user experiences.

As a young person in college, reading your work on that was inspiring. Maybe that happened in some ways and maybe it didn’t happen in other ways. Now, what I’m looking at is a bunch of companies with huge training data needs just doing it anyway. They can just horsepower through the front end of the website with an AI tool and take the data, whether or not it was made available, under whatever terms they want.

That dynamic, on the one hand, feels like the vision of the Semantic Web was achieved. Everything is now machine-readable because the machines learned to read. Now, on the other hand, it feels extraordinarily extractive. It feels unfair to so many people in a way that is just obvious to see. I don’t know if anyone will listen. I don’t know that anyone will say, “Actually that trade-off was improper. It made a lot of people feel bad even though you got to build the tools you want.”

I’m curious if you see that dynamic. I’m curious if you see the vision of the Semantic Web achieved with generative AI in this way, and I’m also curious if you see the unfairness of the trade that so many people perceive.

The Semantic Web has succeeded to the extent that there’s the linked open data world of public databases of all kinds of things, about proteins, about geography, the OpenStreetMap, and so on. To a certain extent, the Semantic Web has succeeded in two ways: all of that, and because of Schema.org.

Schema.org is this project of Google. If you have a website and you want it to be recognized by the search engine, then you put metadata in Semantic Web data, you put machine-readable data on your website. And then the Google search engine will build a mental model of your band or your music, whatever it is you’re selling.

In those ways, with the link to the data group and product database, the Semantic Web has been a success. But then we never built the things that would extract semantic data from non-semantic data. Now AI will do that. Now we’ve got another wave of the Semantic Web with AI. You have a possibility where AIs use the Semantic Web to communicate between one and two possibilities and they communicate with each other. There is a web of data that is generated by AIs and used by AIs and used by people, but also mainly used by AIs.

Because AIs find that, once they’ve extracted the data, the most efficient thing is to exchange that data in a semantic way. To a certain extent, AI solves that problem of conversion of non-semantic data into semantic data. Maybe we’ll be in for an exciting time of some of the interoperability that we were looking for from the Semantic Web being available.

What’s interesting about that is your campaign at the beginning of the Semantic Web era was one of persuasion. You were giving talks, you were writing articles saying, “Open your databases to the web in a structured way, so that computers can read them properly and we can have access to all this data and build wonderful things.” That campaign of persuasion is also a campaign of negotiation.

As the holder of the database, I can say, “Okay, that sounds good. I understand the value that would be created if I do this. I understand the value that I might receive. And here’s what I’m willing to give you. And that feels fair.” Maybe it’s good or maybe it’s bad. . Maybe it will succeed, maybe it will fail.

But the person who has the data has agency, and your campaign of persuasion was to convince them to make the decision to give up the data in this way. The AI companies are not in the business of persuasion; they’re in the business of extraction. It feels like a lot of norms on the internet were not ready for that.

By norms, I mean there’s not a system encoded by the W3C about what web browsers can access what websites. There’s robots.txt, which is just a file that everyone has a handshake agreement to abide by, that maybe the companies are abiding by or not. How would you build a system that gives the owners of data control over the extraction from the AI companies? Could you do it today?

You could certainly design it, and semantic web technology allows you to write information about information and to write things like rules. If you wanted, it would certainly be technically possible to design and build that system, whether or not you get anybody using it.

What’s interesting about that, again, is this dynamic between the open ecosystem and the centralized providers. It occurs at every part of this conversation. No one is listening to robots.txt from what I can tell, which dictates, “Can you crawl on my website?” Maybe Google still listens to it, but actually Google has another crawler. If you want to be in the search index, you have to allow Google to crawl your site for AI. They have some leverage that they’re using.

But you have a big centralized service provider in Cloudflare, which has a lot of caching for a lot of websites, and it’s now saying, “We’re just going to block the AI crawlers. We have a lot of leverage in this ecosystem as well. We will block most of them.” Cloudflare CEO Matthew Prince has been very loud about this, and he’s saying, “We have a new format called the Content Signals Policy, and you have to pay. We are creating leverage by just blocking AI crawlers at every scale.”

On the one hand, it’s great that someone has a lot of leverage to push back. On the other hand, it’s a centralized provider on the web saying, “We will control how the web develops. We have the power and we’re going to use it.” That dynamic is, again, the tension that animates this conversation, if not the entire web. Is that appropriate? Do you think that’s good? Is that how you want this system to work?

What do you think of that move?

[Laughs] I’m not the person who invented the web, sir. Again, I see that tension. Most of my conversations on Decoder about the web and the future of the web land on this tension. There are some people with power and some people without, and the web is supposed to democratize power.

But putting the power of implementing robot.txt in the hands of Cloudflare, that’s, in a way, a strange twist.

Yeah, it’s unforeseen. I’m wondering, as somebody who has seen the development of this from the literal beginning, if you anticipated that twist, and if you think Cloudflare having this power over the web is appropriate?

I think Matthew Prince said it himself. He said in a blog before that these are decisions he shouldn’t be making. He said this about the question of blocking hate speech. He said, “It shouldn’t be for me as the CEO of Cloudflare.” He blogged that and I think he may say the same thing — that he would prefer it if this thing was done by the government or by a consortium. I’m guessing, trying to channel him, I imagine that he probably doesn’t feel it’s perfect, but he’s found that he’s only–

Hold on. Matthew has been on the show before. We’ll have him again. I know what he would say, and I think the economic interest for Cloudflare to do this is also very much aligned. Cloudflare, on behalf of its customers, will tell a bunch of AI companies to go away unless you pay. Cloudflare and its customers have a deeply aligned economic interest.

In many ways the customers are paying Cloudflare for that service specifically. I’m wondering about you as the architect of the web. Do you think having one centralized provider that can block or deny access to a number of web browsers on economic terms is the right outcome?

Of course, I don’t feel that a centralized provider at any level — like content distribution networks or at any level of the web being obviously a centralized monopoly — is good for the web.

Could you build that functionality into the architecture of the web itself so that the different websites, different database owners, can say, “Not unless you pay me,” and this is in the standard? And then that would be actually honored by everybody in the ecosystem?

You could write the protocols. One, in fact, is micropayments. We’ve had micropayments projects in W3C every now and again over the decades. There have been projects at MIT, for example, for micropayments and so on. So, suddenly there’s a “payment required” error code in HTTP. The idea that people would pay for information on the web; that’s always been there. But of course whether you’re an AI crawler or whether you are an individual person, it’s the way you want to pay for things that’s going to be very different.

The other thing I’m really interested about — just as we look at the world of browsers being introduced, literally just in the past two weeks — is that there are more new browsers and more big companies interested in browsers in the past two weeks than maybe the past five years. Again, OpenAI has Atlas, Perplexity has Comet, Google is rolling out the features in Chrome. Microsoft is rolling out the features in Edge, on and on down the line.

Atlassian, the big software company, just bought The Browser Company. It has a browser called Dia that does some of these things. All of them are built on the same core browser engine technology, which means these are a bunch of Chromium browsers at the end of the day.

Do you think there needs to be innovation at that part of the stack, that the browser engine itself needs to be where the competition happens? Or is it just the user experience where the competition needs to happen?

In general, both. It’d be nice if there were more than one browser engine around. In fact, there’s Chromium, which is a monopoly. It’d be nice if other browsers were competing, but a browser engine is a big thing to build. So, having one way of doing it… sometimes in software stacks, you have one definitive standard version becomes the standard, because you can’t afford to have more than one implementation with a browser stack. But I think that it’d be nice to have more than one.

There is one other one of note, WebKit, which is what Apple uses in Safari on the iPhone. It is dominant in its way. Building for WebKit is a thing that every mobile developer has to think about all the time, because it is dominant on the iPhone. Apple doesn’t allow other browser engines on the iPhone. It certainly doesn’t allow Chromium.

Now thanks to some EU regulation, it might have to, depending on how some antitrust litigation in the United States goes. But up until now Apple has not allowed anything but WebKit on the iPhone. Even Chrome on the iPhone is a skin over the top of the core WebKit browser engine. Do you think that Apple being made to allow Chromium to run on the iPhone, for example, will actually lead to new browser innovation?

I can’t help but think having competition to allow Chromium to run on iPhone sounds like a good move.

What do you think that would achieve?

Whenever you increase competition between different sections of the layer, it tends to improve innovation. You get more bright ideas out there.

One of the arguments I’ve heard for why Apple will not allow other browser engines is that they can artificially restrict WebKit. It’s not as good of a competitor to the native apps on the iPhone as web apps on the desktop are to native apps on the Mac or Windows.

If you had true web apps being able to run in Chromium on the iPhone, if you had that browser competition and there was a much more capable browser, do you think that would displace how the native apps work?

This is where we began the conversation. Is this push toward apps because apps are more capable on the iPhone? And if you had a browser that was as capable as a desktop browser on the iPhone, do you think that would change the dynamic?

We’ve been chatting to a bunch of experts on tests to find out to what extent this is true. Yes, I’ve heard rumors, but I can’t substantiate them on whether Apple is deliberately slowing down WebKit on the phone, so as not to compete with Apple native apps.

Oh, by the way, just for the record, Apple would tell you, as loudly as it can, that WebKit on iPhone is as good as any other browser. And most users would tell you that it is not. That gap is where I think the theories about Apple’s development priorities come from. But I’m just asking more hypothetically. We’ve discussed the web as an application at an all time high.

On desktop, if you have a Mac, or a Windows PC or Linux machine, you are using web apps. Mostly what you’re using is applications expressed through web technology. Even if they appear to be native, Electron and other wrappers are just presenting web apps to you in ways that feel native. That is not true on mobile.

It just really is not. Even for Google’s efforts on Android, progressive web apps on Android have not taken the world by storm. Do you think a more powerful browser on the iPhone would ever change that dynamic? Or do you think people just want apps on phones?

I think a more powerful browser on the iPhone would change that dynamic.

That to me feels like, for all of the things we’ve talked about, the future of the web has to happen on mobile. That is where the people are, that is the device that everyone carries around every day.

Right now, Apple’s decisions about what the web can and cannot do on mobile are actually make it the gatekeeper in real ways for most people’s experience of the web. So, if you introduce some competition, can you break that or do you need to just switch to Android, where Google wants you to use the web?

Do you have an iPhone or an Android phone?

When you use the web on the iPhone, does it feel like the thing that you set out to build so many years ago?

Pretty much. I suppose if I’m going to read anything serious, I tend to read it on a laptop.

We’ve laid out the state of the web today. You’ve got a company called Inrupt. You’re building digital wallets. Explain to folks what Inrupt is and how those products might actually solve these problems.

Inrupt is a company I started, it’s “In” for innovation, “rupt” for disruption, and inrupt.com was available as the domain name a few years ago. So, Inrupt is designed to do whatever it takes to develop the Solid vision as an ecosystem. What Inrupt has in fact done is to produce the enterprise-grade, secure, scalable version of a data wallet server.

So, all the Solid protocol-compatible servers. It has a protocol, ESS, or Enterprise Solid Server, which is not open source, and it’s engineered to be scalable and secure. Inrupt has worked with a bunch of people. We’ve worked, for example, with the BBC: we worked on a project to give people who logged on to the BBC a little data wallet on the side. They do have a watch party, and the watch party demo worked, which they did with real people, and it worked using a data wallet.

There’s the government of Flanders, the Dutch-speaking part of Belgium, that now interfaces with citizens through these Solid-compatible data wallets. Inrupt also has an interesting connection with Visa at the moment, because it wants to make sure that they do the right thing in terms of personal data when it comes to agentic commerce. There’s a recent announcement, only a few weeks ago, from Visa on that.

If you’re listening to Decoder and you want to go try Inrupt or a Solid-based wallet, can you do it? Or is this all happening at one level up?

Yeah, you go to solidproject.org. Then if you’re a developer, the operating system is developer-ready, but not ready for everybody to get on and get their data wallet yet. But we will be in a year or so.

Well, Sir Tim Berners-Lee, this has been an amazing conversation and thank you so much for joining Decoder. I feel like in a year or two we’ll have you back and we’ll see where the web is at and what you think of it then, because it feels like it’s in a lot of flux right now.

And in a way it feels exciting because it has been in stasis for some time, but it also feels more uncertain, and I’m curious what happens in a year. So, we will have to have you back.

Thank you so much for joining Decoder.

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