This conversation has been edited and condensed for clarity.

Sam: How do you define decentralization? It’s a term that we kick around a lot and it’s a term that it turns out a whole bunch of people haven’t heard about.

Danny: In the real world, when we’re talking to someone, we talk directly to them. And when we’re buying and selling, we buy and sell directly to someone. The internet and systems like it were designed to work like that, where you would be working peer to peer. So it’s got a little weird that now when we talk to people or buy something, everything that we do, everything that we say, every transaction, goes to some mysterious thing in the cloud, gets recorded, filed away, stored, and then comes back. The internet was all about disintermediation. So decentralization in this context for me is really about returning the internet and technology back to how you would kind of expect it to be.

Luke: When I’m thinking about centralization and decentralization, it really comes back to that question of where the power lies in the network. Early bulletin board systems, early web—it was very decentralized in the sense that you as an individual or a user ran your own thing. You as an individual had a lot of autonomy and power in that structure. Whereas now we have an intermediary. If we’re going to go on Facebook, the company Meta is the intermediary between Danny and I. And over time they have grown to have more power and exercise that power in different ways that can impact the way that Danny and I can interact online.

Jay: I think of decentralization as a system that has multiple points of control where agency or the ability to do things is spread out. Centralized systems are really good at doing one thing fast. They concentrate resources. But they’re also fragile because you have a single point of failure, a single point of control. They flatten out differences. If you want experimentation over a large problem space, it’s actually not really good to just do it in a centralized way because then you can only try one thing at a time. Early on there was a ton of experimentation. Every person spinning up a website was in control of their own little domain. And then now that we have a few big sites starting to really dominate, you start to get less and less diversity and variety in the ecosystem overall.

Sam: If you have simply one type of algorithm, one type of object around which you will congregate, your options are severely limited to what types of communities you can even form, let alone whether or not they’ll succeed.

Jay: Where you actually see this is like farming because when you have factory farms that just do one type of crop, we had one variety of banana that we settled on making and then when this banana plague came, we wiped out that variety of banana because we just did one thing too much. It’s not that centralization is bad in itself, it’s just that if you overdo it, you make yourself vulnerable to certain types of systemic failures.

Danny: One of the challenges with explaining decentralization is just because people are a little unused to it in this context. At the beginning of the internet, there were a lot of people going, well, who’s in charge of the internet? I’d like to speak to the person who is managing the internet. A lot of the early engineers were like, we’d love to ideally do this in a decentralized way, because that’s going to be super resilient. But we really need to ship fast. So we’re just going to put all these servers in one room. And so you’ve got to this point where people started asking the question, who’s in charge of the internet, and people would go, it must be Mark Zuckerberg.

Luke: The example that I always go back to is the late Bell system. When voice telephony started, you had a very diverse set of actors. Over the decades that natural centralization ended up with a literal monopoly of the Bell system that was eventually broken up. The engineers at Bell Labs had invented packet switching way earlier than was ever deployed. They built a lot of the technologies that ended up making the internet happen. But because the idea of the internet was competitive to voice telephony, those technological breakthroughs were stifled. When thinking about centralization and decentralization, there’s pretty consistently this trade-off between efficiency and practicality and then room for innovation. Because by its very nature, the centralizers aren’t the ones that want to disrupt themselves.

Danny: I want to push back a little bit about this idea that centralization, while the trains run on time, it’s super efficient, and decentralized systems, you’re making the sacrifice. Centralized systems like Big Bell are very good in relating to and simple for other large systems, centralized systems to interact with. In particular, large states, large governments, large regulatory apparatus. Decentralized systems, it’s a lot harder. Either you have to make yourself the shape of something that can comply with these laws, which means that you have to centralize up, or you have to go to legislators and explain, look, if you want to control this, you’re going to have to control your own citizens.

Jay: The assumption now not just among regulators, but among people in general is that we’re working with just massive scale social companies and that’s what it’s always going to be. I feel like we’re at this inflection point where another generation or so, or even in another five to ten years, we will be in these very solid closed ecosystems where people don’t recall what the open web was like at all.

Danny: People try and start regulating these things as though it’s Big Bell, as though they’re regulating a handful of companies. And then the consequences become apparent. The UK, presuming that the internet was for companies, passed this law, and now small sites are blocking UK users, UK forums have to shut down. So people are going, wait, my vision of how this works is kind of incoherent. A lot of this is technologists going, okay, this is fragile, this is broken, and also it’s not making people happy anymore. People don’t feel happy on social media and they feel happy when they’re talking to their friends. So let’s build something that lets people talk to their friends.

Luke: You did bring in the Communication Decency Act. It was unconstitutional. I was going to bring in the crypto wars in the 90s. For decades, the NSA fought tooth and nail to try and prevent encryption technologies from proliferating. Long story short, the crypto warriors at the time end up winning because this is first amendment, it’s speech. But I think the bigger trend was that essentially this is all just mathematics and people are figuring out how to do it and regulators can try and whack a mole forever. But the technology managed to get its way out.

When Congress was considering comprehensive data privacy laws, there were requirements that users could at any time go to a company and request that their data be deleted. Makes sense on the face of it, right? But when you start to think about, how would that work in a decentralized system? How would that work in a system that is designed specifically to prevent the deletion of data? It was a big educational hurdle. That’s part of the challenge of the initial question about what is decentralization. I’m actually curious, Jay, how y’all have seen the educational component of going out to consumers with Blue Sky and trying to explain to them the benefits.

Jay: From trying to explain decentralization for many years, I actually think that it’s best to approach it not from the architectural, technical side, but really talk about what are the concrete benefits that you’re providing people. If you can show, not tell, that’s the best. That’s one of our internal mottos. If it’s in the product, if it’s a feature or a user benefit that users can immediately use, it is much more straightforward for people to understand. Talking about decentralization is actually much like talking about the architecture of a database to a normal person. At the end of the day, what people do care about is, how do I have a social media experience where I see what I want to see? Where I don’t see the stuff I don’t want to see?

Danny: All of these sorts of questions about decentralization, centralization and how you make that succeed and popular and win people over, a lot of this is downstream of usability. If I was to point to Blue Sky as a really big success in our environment, I would do it less because of the protocol. It’s because it’s great, it’s usable, it’s friendly, you download it and away you go. One of the challenges with having multiple choices, having all this optionality, which is what decentralization offers you, a way out of just these tyrants, these oligopolies, is you’re drowned in choice. When you go to Mastodon, it says, which one of these thousands of servers do you want to hang out on? And you go, I don’t know. I’m going back to Facebook. AI and things like that, as agents that are acting on your behalf in a decentralized way, can make those decisions for you based on what you want.

Sam: I would like to push back on a few things there, Danny. If you create a cocaine machine that always delivers you cocaine, usability level is going to be high. Sometimes these discussions also around efficiency end up being just about, okay, what’s the fastest way to get someone somewhere? Does it matter where? No. All we want to do is make this thing profitable. All we want to do is capture audiences. All we want to do is leverage network effects. I think the question of why people are going to this place is talked about like hanging out at a bar, like wanting to chat to a friend. Those whys can be quite quickly perverted. So you’re always going back for the dopamine rush rather than for the personal engagement.

And the other thing I would just say briefly on AI is I’m even worried that unless we come up with an excellent decentralized AI, now we’re just handing it off to another business that then makes choices on your behalf.

Danny: I totally agree. One of the things that we’ve created in mass social media is these dopamine factories. But a lot of things that I do produce dopamine. And the question is, is whether this is on my terms or not. I want to create something that is as engaging as that, but its purpose is not to get that dopamine by shocking you or making you angry. On AI, I think I have to agree with you. All of us involved in pushing for greater autonomy are fighting on multiple fronts. There are lots of people working on creating locally run AIs that can operate on behalf of you, rather than on behalf of the government, on behalf of a large corporation, on behalf of that large corporation’s likely nutty leader.

Sam: Jay, you’re in this funny in-between place of being a relatively big player and yet being the rebel at the same time. What are the concrete tensions there that you have to deal with? How do you make the calls between what ends up being a decentralized part of your operations and what doesn’t?

Jay: It’s kind of funny that we’re considered a relatively big player now because we only opened up to the public 18, 19 months ago. We’re still a very small team. And we have to constantly juggle this trade-off between protocol and product, between the open decentralized mission of what we’re building and what we need to do to build and serve a mass market social media app. We have to consider that trade-off for every feature we introduce. Are they going to be on protocol or off?

And then the other one for us is just this inherent tension between convenience and control. The easiest thing is to just go to an app that has one really solid algorithm, like TikTok, and just sit there and scroll. But if you want to have a bit more control over your information diet, you have to give slightly more effort into it. Our goal has been really to not just try to create every single interface we think people might need ourselves, but open it up to the ecosystem to build.

Actually the whole custom feeds concept—you can’t even create a custom feed in the Blue Sky app as of today because that’s just something we didn’t implement. And then the third party ecosystem popped up and was so successful at creating no code feed builders that now there’s 100,000 feeds out there and none of them have been created through the Blue Sky app. That’s one of the cool ways we can get around some of these trade-offs by leaning on the strengths of having an open ecosystem.

Luke: We kind of talk about centralization and decentralization typically in almost a binary when in reality it’s like different parts of a technical stack. Some of them should be centralized. Some of them should be decentralized. How do you think about which pieces of the social media stack should be totally decentralized versus which need a little bit more centralization?

Jay: Not everything needs to be decentralized. Centralized systems have benefits. But the places where you really want parallel experimentation and resilience and user control, those are all things that you should open up and decentralize. Parallel experimentation in what makes a good social media feed is actually very important because that’s ultimately what’s delivering information in front of you every day. Yes, maybe we’re going to get one algorithm that’s really good, but it’s just inherently by being one algorithm, it’s not going to serve every single use case in the world. Often the default algorithm is just optimized for engagement. It’s trying to get you to spend longer scrolling so that you can see more ads. This is why social media feeds tend to push towards polarization and outrage, because that’s the most engaging human emotion.

Now, if you have a broader ecosystem of feeds, then you get people experimenting with making feeds that harmonize viewpoints rather than polarizing. Or you can be like, I just want to show cats. Some people just want to see cute animal pictures all day. Other people are like, no, give me the news. You have so many different preferences that it’s good to differentiate and let people have control.

Same with moderation. You need to basically set some standards. That’s why it’s good to have some larger authorities who can invest serious resources in preventing the spam problem. But then there’s other stuff you want to experiment with. If you want to create a moderation labeler that just strips out all screenshots from your experience, which people have done, that’s something that you can build.

Danny: There are two elements for how you make something like this better. The first one is sort of a developer-facing thing, which is just this engine of innovation. If it’s not centrally controlled, then anyone can chip in. The web was not invented by the people who invented the internet. It was invented by somebody 5,000 miles away. So there’s always a natural advantage to having a system like Blue Sky, like these open protocols where anyone can pitch in.

Jay: Just because everything’s open, people are building stuff that is just totally outside the scope of what we could do. They’re building video apps, they’re building whole new content modalities, blogging apps. It’s great because with that level of openness, we can just see what emerges from all the many experiments. If we didn’t have that open foundation of the protocol, we probably wouldn’t see this broader swath of innovation.

Danny: For consumers and users, I think the most important criteria about what should be decentralized and what should be centralized is things that give people a right of exit. The consumer should be able to go, you know, this sucks. I’m just going to go and try something else. Google, back in the day, used to link to all its competitors on the front page. Google to this day argues and says, we can’t be a monopoly because anyone can switch to another search engine. And of course, the truth is that it’s very hard to move from Google because I’ve got a Google login and that gives me YouTube and actually my business runs off my Gmail account. So when you’re designing these systems at the very beginning, if you want to be disciplined about it, you kind of have to go, okay, how do we always provide our customers, our users with a right to leave and how can we stop ourselves from taking that back from them?

Luke: With the right to leave also, it’s not just being able to physically leave, it’s being able to leave and take with you the things that you’ve created there. The difficult part is not setting up the new account, it’s bringing all of those things with you. Being able to take your whole Twitter feed and all your followers and all of the information that is attached to your Twitter username and bring it with you to Blue Sky—I think that’s one of the things that if we’re talking about decentralization on a broader level, that’s a big part of that.

Jay: That’s actually one of the core user benefits that we don’t talk about as much when we’re trying to sell the app itself because it’s not so much an experience you find in the app as the right that the app gives you. It’s important as internet rights, freedom that people should have—the ability to leave services with some of their core identity and data. Otherwise you don’t get competition in the ecosystem. But also as a user, your relationships, this is a central part of your life being mediated by these companies that don’t give you control over this very central thing to your life. If you make a bunch of friends and you connect with them on Instagram or Facebook and then you leave, now you’ve lost access to your friends. Why should Mark Zuckerberg be able to say that if you stop using my product, sorry, you have to figure out some other way to contact your friends?

You have a phone number and you can switch cell phone carriers and keep your phone number. Even email—you can move and you can continue emailing people across providers. It’s not like if I leave Gmail, now I can no longer email anyone at Gmail. That would be the situation we’re in with social companies.

Luke: The example that I’ve given people that seems to really resonate is imagine if you had your Spotify account and you had spent 20 years making really awesome playlists that you really loved and then for XYZ reason you decided that you wanted to move and you can’t bring them. You’d have to physically remake all of those playlists. And that’s kind of what we’re talking about.

Sam: One of the things I’ve found about social media is just how significant it has become to be able to move. If I lose my songs, it doesn’t affect probably my friendships nor my employability. If I’m on Instagram and I’m a wedding planner and that’s how everybody sees my work, that particular platform is now tied to my ability to earn money at all, to pay rent, to feed my kids. It’s not just followers—it’s livelihoods, reputation, status, and friendship all on the line.

Danny: It’s really in those companies’ interests to not make that work well. One of the hardest things about building technology is being able to discipline your company to make something that’s good for the user. And if you’re a little bit terrified that they’re gonna leave, that’s a great disciplining factor. You have adversarial interoperability, where you’re actually empowering these smaller companies to go and get that data on behalf of the users. With Spotify—Spotify doesn’t give you the ability to take your playlist, but there are great websites that will do that matching.

Luke: Policy can sometimes make data portability or interoperability easier. But it can also make it a whole lot harder. When people in Washington, DC are like, what do you want out of the world to try and make the internet better, for me, it’s like, don’t make it worse. But also it’s actually a lot easier to get rid of hurdles and weapons and things that current platforms use in order to prevent those kinds of adversarial interoperability.

Early on in Facebook’s lifetime, Facebook was interoperating with MySpace. Fast forward—there was a company called Power.com that was trying to be a dashboard for all of your social media. Facebook sends Power.com a cease and desist and then ends up suing them under the Computer Fraud and Abuse Act and also the Digital Millennium Copyright Act. After a decade of litigation against Power.com, I think the actual judgment was something like $8,000. But they crushed Power.com through litigation. The lesson for policy makers is the easy thing to do is to just get rid of those things that the companies are currently using to continue cementing centralized control.

Danny: One of the things that’s in the pressure to create these monopolies that people then get very worried about is that they’re manageable for governments. What’s going on is governments are going, okay, you made this problem, you have to fix it. And I think what you will increasingly see as decentralized social media begins to come up with an alternative that doesn’t have those problems is those big companies are gonna go, well, you can’t trust them. You can trust us because we work with you and you know us. And that’s the thing I worry about. It will give those giants an opportunity to say, well, these other people are irresponsible. They’re small. You don’t know them.

Jay: My final word here is that I think it’s gonna be really powerful when people can see the benefits of having more bottoms up solutions to some of these problems. If we can just get more experiments running, happening at the same time, I think we’ll start to see more emergent solutions. Right now the only way to change things is either you lobby the company or you go to the government and say, regulate this big company. People are sort of in this sense of learned helplessness right now. But when you open things up again and say, hey, if you have a problem with the way this is working, you could just submit a PR. Or you could build your own version of the things. Then I think you can pivot towards a better solution faster than if you’re meeting this big institutional resistance. If we have this approach to tackling some of the problems we see with social media, we’ll see a lot faster approximation of reaching something better.

Sam: Thank you very much, Jay, Danny and Luke. I’m looking forward to catching up in a year or two and seeing where we’re at then, because the space is moving quickly and I hope it’s moving in the right direction.