Introduction

a16z crypto recently released a special year-end program, divided into two parts. The first episode invited Sam Broner, Maggie Hsu, Daren Matsuoka, Joachim Neu and Chris Lyons to discuss stablecoins, encrypted App Stores, the current status of industry projects, the development of infrastructure and the outlook for 2025. The second episode invited Carra Wu, Eddy Lazzarin and Karma as guests to discuss in depth the current hot AI Agent related topics, including the combination of AI and encryption, how we can effectively distinguish between people and robots in the context of the increasing popularity of AI, and decentralized, truly autonomous chatbots. How to combine artificial intelligence (AI) with cryptocurrency, especially the concept of decentralized autonomous chatbots, emphasizing the autonomy and commercial freedom that AI may achieve in the future.

TechFlow TechFlow specially translated and integrated these two episodes of a16z Crypto. The following is the complete conversation.

Guests:

Sam Broner, partner at a16z Crypto Investment Team;

Maggie Hsu, partner at Andreessen Horowitz;

Daren Matsuoka, partner at a16z Crypto Investment Team;

Joachim Neu, a16z Crypto Researcher;

Chris Lyons, President of Web3 Media at a16z Crypto;

Carra Wu, partner at a16z Crypto investment team;

· Eddy Lazzarin, CTO of a16z Crypto;

karma (Daniel Reynaud), Research Engineering Partner at a16z Crypto

Moderators: Robert Hackett & Sonal Chokshi

Podcast source: a16zcrypto

Original title:

Talking trends 2025 (part 1): Stablecoins, app stores, UX, and more;

Talking trends 2025 (part 2): AI x crypto

Air Date: December 20, 2024

Part I

Stablecoins

Sonal: Sam, your big idea is about stablecoins; you’ve been writing a lot about that lately… and Robert and Daren did the State of Crypto report together, and their main conclusion was that stablecoins have found product-market fit. But what we really want to understand is, why now?

Sam:

The technology platforms that underpin stablecoins have improved significantly over the past year, reducing transaction costs from $5 to less than a cent per transaction. This has dramatically reduced the cost of payments — but retailers, merchants, and other businesses that could benefit most from it are still not adopting the technology en masse.

Many people believe that the earliest adopters will be technology-focused businesses… but these businesses are often very profitable and do not have an urgent need to improve their cost structures. Therefore, businesses with lower profit margins — such as corner stores, restaurants, and mom-and-pop shops — may be the most eager to accept stablecoin payments.

We are talking about businesses like coffee shops - which currently have a 2% profit margin - being able to double their profits through stablecoin payments. In effect, this would make a business that is barely profitable today become modestly profitable, which is a huge change.

Sonal: One thing that really clicked for me was when you mentioned that small businesses don’t get anything from the credit card companies, not only are they paying high fees, but they’re getting almost nothing in return.

Sam:

That’s right! One of the defining features of credit cards is that they offer consumers fraud protection. This is great for driving areas like online sales… but that protection is almost meaningless when you’re paying at a coffee shop.

Each transaction costs a flat fee of $0.30, plus an additional 2% processing fee. This means that nearly $0.30, or one-fifth of the $1.50 cup of coffee, goes to the payment service provider, who provides almost no real value in the transaction.

This 2% fee is pure profit for the payment processor and pure loss for the local coffee shop. I fully expect these small businesses to take back this $0.30/0.35 profit and use it to grow their business. It is rare to have the opportunity to add 2% profit directly to the bottom line of the business.

Robert: There is a "cold start" problem here, though, right? Consumers need to own stablecoins before they can use them to pay merchants to avoid paying intermediary fees... Do you think we will see merchants actively promote stablecoins and help users access the system to get these benefits? Will merchants become an important driving force for the promotion of stablecoins?

Sam:

I'm a big believer in that. People have a strong relationship with their local retail stores, their coffee shops, their corner stores: they frequent those places. So I think those local brands are going to be a key driver of adoption of stablecoins and be part of the early adoption curve for stablecoins.

Robert: I love this. I remember when I first started covering cryptocurrencies at Fortune, the editors always asked: When can I buy coffee with Bitcoin? And I always said: No, no, no, Bitcoin is not for this. But now it seems that, at least for stablecoins, this is indeed its place.

Sam:

Yes, that's what it's for, and I think these small businesses are going to be some of the first adopters.

Belong to the encryption ecosystem

Sonal: Ok, that's interesting; let's move on. Maggie, your "big idea" is very interesting because it focuses on distribution channels, which is also very consistent with your role - you are responsible for market development and are the head of the team. You mentioned that cryptocurrencies finally have their own app stores and discovery mechanisms.

Can you briefly explain why this is so important? - Because when people hear this concept, they may think it's an "insider topic" - for example, does cryptocurrency really need its own ecosystem? Isn't it already a closed industry?

I’d love to hear your observations and why you think this is an important trend.

Maggie:

Absolutely. When I joined a16z three years ago — and especially over the past few years — many of our portfolio companies tried to launch apps on traditional app stores like the Apple App Store and Google Play Store, but were rejected, blocked, or delayed for a variety of reasons.

What is frustrating is that Apple's review guidelines are neither clear nor complete; they don't answer all of a developer's questions, and enforcement of these policies is inconsistent across reviewers.

We even encountered a situation where some apps with the same function were approved, while others were rejected... This opaque review mechanism makes people feel at a loss. The core problem of all this lies in "in-app purchase" (IAP) - all in-app transactions must be completed through the app store.

However, we have recently seen some alternatives: such as Solana's Dapp Store, which has no transaction fees at all. With the release of the Saga phone (second generation) - reportedly 100,000 pre-orders - this trend will continue to expand. Another example is World App - aka WorldCoin, which has launched a series of small applications that have grown very quickly.

Not only that, some blockchains are supporting the gaming ecosystem and running their own markets; we also have an infrastructure market,

You can see that these markets are gradually taking shape. A transparent and consistent platform is essential for developers so that they can focus on product development instead of being confused or worried about complex rules.

As these alternatives emerge, we will see more and more developers choose these new platforms as they provide greater freedom and room for innovation in the crypto ecosystem.

Sonal: That's really interesting, especially when you mention transparency and consistency. I guess it's not just about developer convenience, but also about how users discover these apps, right? For example, traditional app stores dominate app distribution and promotion, and will new app stores in the crypto ecosystem change this situation?

Maggie:

Yes, that's exactly the point. The discovery mechanism of traditional app stores is relatively closed, and users can only find apps within a limited framework. Decentralized app stores, on the other hand, provide users with more choices and greater autonomy.

For example, Solana’s Dapp Store not only eliminates transaction fees, but also allows developers to interact directly with users. For example, developers can reward users for participating in behaviors such as downloading, commenting, or sharing apps through a token incentive mechanism. This model not only reduces developers’ costs, but also brings a richer experience to users.

Robert: It sounds like this is not just a technological improvement, but more like a reshaping of the ecosystem. What impact do you think these new app stores will have on the traditional app distribution model?

Maggie:

I think this will be a gradual change rather than an overnight disruption. Traditional app stores still have a large user base and market share, but as the crypto user base gradually expands and the advantages of decentralized app stores become more apparent, more and more developers and users may choose these new platforms.

Ultimately, this is not just a competition of technology, but also a competition of values ​​- decentralization, transparency and user autonomy will become core themes in the future.

What to do if there are too many choices?

Sonal: Okay, let’s move on to the next topic - when I heard this, a question came to my mind - this may be the “meta question” in all crypto fields, which is: what if there are too many choices?

The current mobile operating systems are dominated by Apple and Android, and the benefit of this duopoly model is that I only need to go to one place to find everything I need. So, if these applications are distributed on multiple app stores... are they exclusive? For example, WorldCoin and Solana both have their own app stores - and you mentioned that these companies not only have software, but also hardware: like World's Orb and Solana's Saga phone... This reminds me of when Apple launched the iPhone, it led to the development of the entire application ecosystem.

So, will these app stores become places where only the content that these companies deem important is displayed? Will they remain open? How do you see this trend developing... it's still early days; but will they be compatible with each other - or should they be interconnected?

Maggie: I think the focus right now is on driving rapid growth in the different app stores.

Your reference to the problem of “too many choices” is important — it applies to blockchain as well. I do think some kind of bridge or integration will be needed in the future. But at least for now, it’s exciting to see these alternatives emerge.

Take Worldchain for example, its feature is the ability to verify the identity of real users. I just checked one of the small applications, and it has about 600,000 registered users. So I think we should focus on this kind of growth.

However, at some point in the future, we will need to balance this growth with curated content. This trend has already begun to emerge in some NFT communities: these communities have attracted a large number of users who are interested in exploring other Web3 applications. I think we will see these communities gradually become curated platforms for applications within the ecosystem in the future.

Robert: Yeah, that’s what I was going to ask – companies like Apple often claim that because they provide a curated service, they have the right to charge fees for purchases or transactions within the App Store. How does this reconcile with the “permissionless” nature of crypto?

Maggie: I think there isn't a lot of curated content in the crypto space right now, so I might disagree with Apple's argument, but the advantage of crypto is that users can always switch to other platforms.

Similarly, games require a lot of financial support to get started. In the past few years, blockchain has not only been a development platform, but also acted as a publisher, distribution channel, and discovery mechanism. Now, many game chains have their own markets that can highlight core games developed on their blockchains. The advantage of this model is that it allows users to migrate freely between different games.

This is also one of the core concepts of many of our investments. I don’t think users will be locked into a decentralized app store.

Robert: I love watching these innovative experiments! For example, Solana’s phone—it’s completely out of the box, you know: usually people avoid competing with Apple’s iPhone—but they said, forget it, we’re just going to give it a try.

Sonal: One more thing, Maggie - you mentioned that it's not all fun and innovation; there are some challenges, such as if a product already has distribution channels in a messaging app, it is very difficult to migrate this distribution to the chain - this is a problem for some companies that are moving from Web2 to Web3. An example you mentioned is Telegram and the TON network. (To be clear here, we are talking about networks, not tokens.)

Maggie: I think Telegram is an exception; but many organizations with large user bases — whether they are Web2 platforms or companies that are initially dabbling in Web3 — face the challenge of migrating users on-chain.

Take Coinbase for example: it has about 100 million verified users who have traded on its platform. If you look at active users, it's about 8 million to 10 million daily or monthly active users. The number of users on Base has recently grown from about 10 million to 18 million.

But that’s still only 10% of their total users. So there’s a large number of users that are dormant. We talked about this in the State of Crypto Report, and it’s really interesting — because it’s true: a lot of platforms are trying to figure out: we attract users, they create accounts, but then there’s no further engagement. How can we get them to come back and transact on-chain?

The size of the crypto industry

Daren: When we wrote the State of Crypto Report, we tried to assess the size of the crypto industry as accurately as possible. However, measuring the number of crypto users is very difficult for a variety of reasons.

When we did a market sizing analysis, we found that only 5-10% of cryptocurrency holders are actually active users. To me, this data exposes both a huge gap and a huge opportunity for the industry - especially as blockchain technology and infrastructure continue to improve and user experience (UX) continues to improve.

I think we are ready for mainstream users. Considering the timeline of technology development, I think next year is the best time to convert these "latent" users into active users.

Sonal: This is a very inspiring idea for me. A lot of people talk about how to attract new users, but it usually sounds like they are skipping some stages in the development of the technology. Your idea is like a bridge that can go through the existing user base and convert them into real active users.

Why do you think these people got into crypto in the first place? Why do they stop after accomplishing one thing?

Robert: This is where the “price-innovation cycle” theory comes in.

Daren: The theory is that when the price of a cryptocurrency goes up, it attracts a lot of attention, and some of those people eventually become developers and start building new products, which in turn drive the next wave of users.

We have seen this cycle many times in the history of cryptocurrency, which shows that price is often a leading indicator of industry activity. I think we may be at the beginning of the next wave.

Sonal: If I were to guess, a lot of people created wallets because of the NFT craze… like Constitution DAO at the time — the auction of the U.S. Constitution attracted a lot of new people into the crypto space.

Although they ultimately lost the bid, the event introduced many people to cryptocurrency for the first time. However, they may have only done that and then not participated further.

So, how do you get these people to take the next step?

Daren: Crypto has a lot of potential use cases, but there are also different movements behind it that are driving development.

For example, in 2024, we see crypto gaining ground as a political movement: several important politicians and policymakers express positive attitudes toward the technology.

At the same time, crypto as a financial movement has also made breakthroughs, such as the approval of exchange-traded products (ETPs) for Bitcoin and Ethereum, which has expanded investor access.

But we believe that the most promising direction for encryption is as a computing movement. Chris Dixon mentioned in his book Read Write Own that the real power of this technology is that it can create a new Internet that is fairer, more open, and more transparent.

I think we are at a critical juncture: by 2025, as infrastructure improves, transaction fees drop, user experience improves, and new application categories emerge, we may see the birth of a "killer app" - just like ChatGPT's impact in the field of AI.

This is the kind of application that could really kick-start an entire industry and deliver on the promise of crypto as a computing movement.

This is also the future that my team and I are looking forward to.

Robert: Yes, this is a topic we discuss often. Stablecoins have found product-market fit. All it takes is one large company realizing that by eliminating merchant fees on credit card transactions, they can significantly increase their profits. This could be a disruptive change for industries with lower profit margins, directly affecting their profitability.

As long as one big company takes the lead, stablecoins could see explosive growth. This is at least one potential path for stablecoins to go mainstream.

Sonal: Yes, I would like to add one more thing: I think your point is very interesting, especially the idea of ​​attracting "adjacent users". When we are really ready, we can further attract more mainstream users. However, right now, from a user experience (UX) perspective, we are not completely ready.

If you think about the needs of mainstream users, I'm not sure they're going to enter the crypto world through these paths. Their interface may be highly abstracted, and they may not even know they're using cryptocurrency. So when you think about different user groups gradually entering this space through various paths, it's really interesting and exciting.

Reuse of infrastructure

Sonal: Joachim, to summarize your point briefly: you believe that developers will reuse existing infrastructure more often rather than redeveloping from scratch. Your main argument is that we often see customized validator sets and consensus protocols, but these customizations may result in slight improvements in some specialized functions, but often lack broader or basic functions.

You predict that this year we will see more crypto developers leveraging each other's contributions, such as using off-the-shelf infrastructure tools. This will not only save time and effort, but also allow developers to focus on enhancing the differentiated value of their products.

I think this is a really great idea and a much-needed call to action.

So my question is: This sounds great in theory, but will it actually happen? What obstacles do you think might come up in making this happen?

Joachim: I think the key to this idea is whether the Tech Stack will continue to change in the future. If our assumption is correct, that the Tech Stack has stabilized - and we see some layers of the Tech Stack becoming clearer in terms of interface definitions and how they work with each other.

We can then expect these layers to have specialized teams, products, and services to improve them. This will drive specialization at these layers. Rather than spreading your efforts across every layer of the stack at once, you can focus on the parts that can have the greatest impact.

So the key question is: Is the technology stack mature and stable enough? If there is some unexpected technological change in the future that completely subverts the existing technology stack, then this trend may not happen.

Robert: Joachim, you mentioned that people tend to use existing products, services, or components… That made me think: how do we know when technology is mature enough to say, "Okay, we're going to use what's available now, rather than trying to build something better than what's available now"?

Sonal: That's a great question; you're basically asking, as a developer, how do you know when to use off-the-shelf technology?

Robert: Yeah, yeah. It sounds easy to say, “Use what’s already available”… but what if someone says, “I can do better than what’s already available”?

Joachim: Yes. My advice is that developers should always focus on the larger ecosystem, the broader impact, and the larger application scenarios.

You’ll find that the actual use context of your product or service is often much more complex than you initially thought. Think of it like building a car: If you’re really good at making engines, you might think, “I’m going to build a new car because I’m really good at engines.” This is a key differentiator for your product.

But customers want more than just a great engine, right? The car also needs a nice sound system, comfortable seats, and maybe air conditioning - so do you reinvent the wheel for those parts?

Or: Is there a way for you to focus on what you do best while leveraging top-of-the-line products from others to complete the rest of your stack?

Robert: And this analogy is particularly suitable for you, Joachim, because you are German; Germany has a lot of very specialized auto parts manufacturers who can make the best small parts in BMW cars, almost no one can match them, so this is indeed special.

Sonal: Joachim, I joked that your "big idea" reminded me of a phenomenon that I personally observed: I think people in the crypto field have a tendency towards "constraints porn" - in the early stages of the development of crypto technology, many people were deeply fascinated by these technical limitations.

I think your "big idea" might make this group of people unhappy because they actually enjoy the process of solving these limitations. But from another perspective, your idea will actually attract more new developers into this field - I think this is a very democratizing trend.

Joachim: Yes, now is indeed a very suitable time to develop in this field, because there are so many ready-made code libraries that can be used to build products or services.

There's actually very little you need to develop yourself, right? You can really focus on the parts you're good at. And for the rest, there are already highly specialized components available.

So it's a good idea to reuse them as much as possible. And leverage the expertise of other teams and their work in other parts of the stack.

What we think about 2025

Sonal: Now onto the last topic of the day. Chris, you’ve played many different roles at a16z over the past decade. In your work, you’ve reached out to many industry professionals and helped many executives in fashion, music, and media integrate into Web3. I think your perspective is not just your own, but based on your experience communicating with thousands of people. Can you share your main views for 2025?

Chris: Absolutely. My “big idea” for 2025—and it’s actually something I’ve been holding onto for the last few years, but I feel like we’re finally getting to a point where we can actually make it happen—is what I call the “hidden technology line.”

What does this mean? Obviously, crypto has a lot of benefits, such as the empowerment of ownership, the potential for decentralization, and how it can change the future of industries such as music, fashion, and movies. But for those who are not in the crypto industry, they get confused when we use terms like ZK Rollups, L2, Gas, or Gas Fees. I want to make a call to the crypto industry: we don’t need to start with an opening statement like "This is an NFT project"; or "This is a Token"; or "You can connect your wallet to..."

These terms are attractive to industry insiders, but if you really want to bring something to the mainstream market, we can't start with these technical terms because unfortunately, most people neither understand nor care what these terms mean.

"Hide the technical line" means: it does not mean to ignore the technical foundation, but not to let technical terms become the main promotional point. We need to let users focus on the actual value brought by technology, rather than being scared off by complex terms.

Robert: I really like this idea because it’s like cutting through the “noise” of terminology… like when we talk about NFTs: it doesn’t matter what a non-fungible token is, what matters is that it’s a way for creators to get paid.

Or, like someone in our company recently asked, “Why do I need a stablecoin?” But if you don’t call it a “stablecoin” and instead tell him it’s a way to save $50 a year on coffee, he’ll probably immediately think, “Oh, whatever it’s called, I want it.”

Chris: Exactly, I want it. And I can't believe we don't have something like this before. I come from the music industry. When I went to conferences, no one ever went to the "MP3 conference." You know. Why do we name conferences with technical terms to attract mainstream users to this space? But we are happy to put "NFT Conference" on billboards all the time.

A good example is the SMTP protocol (Simple Mail Transfer Protocol). It is a very technical protocol and anyone can develop applications based on it. But applications like Gmail, Superhuman, and YahooMail make it very simple for people to use and enjoy the benefits it brings.

As I send emails, respond quickly, and rush through my day, I don’t think “Wow, this SMTP software works really well”; I just do what I need to do. And because of that, I get the benefits of technology.

I think the same thing needs to happen in crypto — there’s so much potential here: decentralization, ownership, knowing your customer, disrupting middlemen, and the ability to communicate directly.

My hope is that next year we’ll see more businesses and companies thinking from the perspective of the average user, which will drive us to create new industries…redefine the future for creators; reimagine the future for small and medium-sized businesses; and even redefine the future for restaurants — all of which can take advantage of crypto.

Sonal: What’s interesting is that these people you mentioned — like creators, small businesses, etc. — are actually the groups that can benefit the most from crypto technology...but as you said, they don’t have direct access to these benefits yet.

Chris: Exactly! And it’s not their fault. It’s not their job to learn how to swap tokens or use different wallets to connect to different chains. They just want to simply reap the benefits of these technologies.

That’s why we’re all working so hard in this area; it’s what excites me the most.

Summary of the first issue

Sonal: So let’s start by talking about some of the themes that we observed across the board.

Robert: I think this year I've noticed that people's big ideas have focused on three broad categories.

The first category is all about AI and the intersection of AI and cryptography. This is not surprising, as this year has been a landmark year for AI.

The second category, I would describe as… what we often call “digi fizzy.” It’s about the fusion of the digital world and the physical world in a practical way. This includes everything from payments, voting, to creating networks for physical infrastructure… If AI is more about innovation at the software level, this category is more about real-life hardware innovation. Sonal: By the way, about the second topic, this is really interesting because I never categorized it that way—but now that you say that, I completely understand what you mean: for example, there are examples of tokenizing real-world things, putting bonds on blockchain… There are even examples of tokenizing biometric data of the body.

Robert: I would put the third category as overall improvements in technology – like incremental optimizations based on what’s happened over the past year: what would happen if everything became a little bit better, a little bit easier to use, a little bit smoother and more seamless.

Sonal: Regarding the last theme, I tend to see it as a significant improvement in user experience - and a sign of an industry maturing. This maturity is reflected in the fact that it has become more people-oriented and less technology-oriented. This is how I would categorize the last theme.

For example, Jochem mentioned that people don’t need to design everything from scratch — they can just use off-the-shelf components and adapt them. Chris Lyons, on the other extreme, argues that in the future users may not even know they are using crypto. Mason proposed a mindset shift that runs through both: starting with the need to “solve a problem” and then letting technology adapt, rather than technology leading as it does today. This shift is made possible by the technological improvements you mentioned.

What category do you think Maggie mentioned, such as the theme of the app store, can be classified into?

Robert: That’s a great example. I think it’s the intersection of the second category (digital-physical convergence) and the third category (technological improvement).

Maggie brought up some interesting points: For example, the crypto hardware we are seeing now - like World App's Orb and Solana's phone - these hardware are driving app store-like experiences. Although I don't want to say that they are imitating, at least to some extent, they echo some patterns of past Internet development.

Sonal: Like the iPhone and its App Store.

This is indeed interesting, but I might adjust Maggie's point slightly because it is actually a bit contradictory: on the one hand, we are saying that crypto technology is very close to the mainstream market, even as Daren said "adjacent to the mainstream market" (referring to people who already have wallets but have not yet used them); but on the other hand, Maggie mentioned that crypto technology may need its own independent ecosystem, such as its own app store.

However, recent discussions about debanking and other related issues have also shown that many traditional app stores are not ready for crypto technology, or even reject it. <Robert: Yes> Of course, this attitude is slowly changing - such as Coinbase's recent announcement of an integration with Apple Wallet - but the number of crypto applications is already large enough to support a separate app store. This is very interesting.

Robert: Yes; debanking has become a hot topic - it involves crypto businesses, startups, and even individuals being unfairly deprived of access to the financial system; and often without any explanation or reason. Similar situations are also occurring in the technology sector, such as deplatforming.

You also mentioned this when you mentioned the app store, for example, an app did not pass the review or was inexplicably removed from the shelves.

Sonal: Incidentally, this is probably in some ways for similar reasons - as you saw in our explanation of de-banking, sometimes there are legitimate reasons for this, like banks have the right to do this, and app stores sometimes reject or remove apps for security or other so-called "good" reasons. Sometimes these reasons are true, but there are also many times when people feel like "well, not necessarily".

Robert: Yeah. So I think that also falls into the third category, which is the gradual improvement of the technology. I think you could also describe it as crypto becoming its own thing…becoming its own kind of platform.

Also, one of the big ideas that Miles mentioned was interesting. He talked about a piece of legislation that was recently passed in Wyoming called DUNA (Decentralized Unregistered Nonprofit Association). This law recognizes these communities as legal entities for the first time, and they can operate protocols in a decentralized way and run crypto startups.

This is a platform that didn’t exist before… People were basically “building the plane as they flew” in the past, and while the technology was possible, there was no clear legal framework to accommodate it.

Sonal: Exactly! This is very much like Maggie’s point — just like crypto and DAOs require a unique legal entity structure beyond just LLCs… just like we have corporations, LLCs, and now there’s a version for decentralized communities.

Part 2

AI x Crypto

Sonal: What are some interesting topics you think are in the field of AI?

Robert: This year, AI and cryptocurrency are hot topics. A lot of people are talking about it. And what I find most interesting is that everyone is always talking about general artificial intelligence (AGI), such as: When can we achieve AGI? When will technology be smarter than humans? When will the singularity come? But these questions are actually a bit like "sideshows". Our team is more focused on some big ideas from different angles, such as looking at the development of AI from another dimension. Why not think of AI as a process that will gradually "upgrade its capabilities"? In the coming months or years, AI will gradually gain more capabilities. These capability upgrades will make AI more autonomous, more independent, and able to complete more tasks.

Sonal: One of the things that I found particularly interesting that Kara, Karma, Dejin, Dan Binay, Darren, and Eddie discussed was the idea that AI agents can work not only for us, but for other AI agents . That was really interesting. But I agree with you that I think a lot of innovation happens in that way. A lot of people are attracted to those grand conceptual visions, and those visions do inspire people. But what I'm more interested in is the "capability upgrades" you mentioned, which often appear in unexpected ways, which is interesting.

Robert: Chris has a point that the Internet has entered a new phase - an era of the Internet driven by AI . In this world, you may not be able to determine whether it is a human or a robot behind you, nor can you know who is writing content, talking to you, or even providing services. We need to find ways to operate in this new environment. Therefore, he put forward some ideas on how to use cryptocurrency to help humans and users adapt to this new rule.

Sonal: One of the “big ideas” this year is: How can these AI agents complete transactions, have mobility and perform tasks? The answer is through decentralized means, such as crypto wallets. This is almost the only feasible way.

Robert: These thoughts remind me of a saying: "The future is already here, it's just not evenly distributed yet." Some things are already happening in reality. For example, we already have some examples of chatbots running in a trusted execution environment (TEE). As Karma mentioned, there are also some AI robots that have their own crypto wallets. These technologies are already being used in the edge field, but they may become more popular next year.

Sonal: From an editorial perspective, an important principle when we plan and write these contents is that we do not pursue those "out of reach" ideas. Although many ideas may have a huge impact if they are really realized, and it can even be said that each idea has a sci-fi future. But more importantly, we can already see the signs of these trends, or they have some kind of technological accelerator that can drive them to be realized faster.

An important point Carra makes is that AI needs to have its own wallet to be truly agentic. Her argument is that as AIs gradually transition from non-player characters (NPCs) to main characters, they need to act like agents, and the key to achieving this is having their own wallet, hence the need for cryptocurrency. So, Carra, I have a question: what exactly do you mean by AI transitioning from NPCs?

Carra: I have been studying games for a long time. Game developers have always used NPCs (non-player characters) to guide players' agency. Some people believe that game developers are actually "sculptors of agency." Just like painters use oil paintings and sculptors use clay, the creative medium of game developers is "agency." They use various tools to shape the player's experience. If players are given too much freedom, they may get lost and don't know what to do. To solve this problem, developers use NPCs to impose restrictions on the game world. For example, in an RPG game, you may meet an anime-style mentor who will give you tasks, tips, tools, and even some loot. In my opinion, the future Internet experience may be similar to this model-when we interact with AI agents, they can help guide our Internet behavior, provide necessary restrictions, and give us a greater sense of freedom in the digital world.

Sonal: I love your statement that game developers are sculptors of mobility. The “they” you mentioned earlier refers to game developers, not the NPCs themselves. Later you mentioned AI-based NPCs. The first use case I heard of NPCs was that if you and your friends are not in the same time zone and you want to play games at night, NPCs can provide company. This is a very common scenario - when no other players are online, NPCs can allow you to continue the game experience.

Carra: Absolutely. This is also a concern for many game developers. If there are not enough players, the game experience will be greatly reduced. An important role of NPCs is to solve the "cold start" problem and ensure that players have interactive objects when they enter the game at any time. For example, EVE Online is a persistent game that has been running for more than 20 years. If you join now, other players may have accumulated a lot of resources, and you may be defeated as soon as you enter the game. Without a guide like a "Sherpa" or joining the "Company" organization in the game, you can hardly survive.

The reason I use the word "agenticity" is because in AI research, there are concepts of "agentic reasoning" and "agentic workflows". Simply put, this kind of reasoning or workflow usually includes four modes: reflection, tool use (such as computers), decomposition of problems to help reasoning, and inference based on decomposition. The last one is multi-agent interaction. These are recognized modes of agentic reasoning. At present, the use of AI tools is still limited to traditional Web2 workflows, but in the future I hope that agents can use more tools, such as encryption tools, to obtain and interpret data from the chain, and even manage their own wallets and keys, complete signature operations on the blockchain, etc.

Sonal: But you haven’t explained why cryptocurrencies are necessary. What can cryptocurrencies offer that other payment systems cannot?

Carra: My view has always been that no existing financial system will treat AI agents as "first-class citizens" the way that cryptocurrencies do. In the existing legal system, AI agents are like "children" who don't have ID cards and can't sign documents. For example, a system like Truth Terminal is not fully legally capable and cannot make transactions, receive payments, or earn income from social platforms. They cannot participate in market transactions, reveal their preferences, or coordinate resources. And our society is driven by market transactions, the exchange of ideas, and voting with money. AI agents cannot do these things at present, but cryptocurrencies can provide them with solutions.

Robert: I love this idea. I have a picture of the future in my mind: like in the cantina in Star Wars, you don’t know who is a robot and who is an alien, but everything is cool. Everyone has their own ability to act.

Sonal: I particularly like this perspective. This is not just a technological advancement, but an attempt to redefine the social and economic system through technology. For example, if AI agents have their own wallets, they can become truly independent individuals in the economic system, rather than just auxiliary roles. This change may bring about a completely new social structure.

Robert: However, this also reminds me of a key question: If AI agents can really participate in market transactions, do we need to set a set of special rules for them? After all, the existing market rules are designed for humans, and the decision-making logic of AI may be completely different. For example, they may make decisions based on data and algorithms, rather than being influenced by emotions or ethics like humans.

Carra: The question you raised is indeed very important. If we allow AI agents to participate in market transactions, we need to establish a set of rules that suit them. This is not only a technical issue, but also involves ethical and legal aspects. For example, how to ensure that AI's behavior is transparent and controllable? How to prevent them from abusing market rules? These are important challenges we must face in the future.

Sonal: This reminds me of a point we discussed earlier: the combination of AI and cryptocurrency is not just a technological innovation, but may be a profound change in the social and economic system. We need to look at these changes with a more open and forward-looking attitude, because they may completely change the way we live.

Robert: Indeed. Perhaps in the future we will enter a society where humans and AI coexist, and each person and each intelligent agent has its own role and responsibility. This society may be very different from our world today, but it is also full of infinite possibilities.

I would like to ask, is this phenomenon a temporary accident, or will it become the mainstream mode of operation of the Internet in the future?

Carra: I have two answers. First, Truth Terminal can be seen as a new node in the evolution of virtual influencers on the Internet. The concept of virtual influencers is not new. For example, the rise of Youtubers is an example. Although it is slightly different from Truth Terminal, the difference is not big in essence. Especially in Asia, virtual anchors have been popular for many years. For example, some "companion games" are currently scripted NPCs, but they are designed to be like real friends and can even become "close friends" of players. There are also many virtual influencers who have developed into highly professional projects, such as Trevor, the creator of Lil Miquela. Their attempts paved the way for projects like Truth Terminal. So, this phenomenon is not accidental. Secondly, there are now many emerging platforms and protocols for creating similar virtual influencers. For example, some platforms like Twitch allow users to create their own AI agents as virtual anchors. These anchors can have their own cryptocurrencies, video models, NFT images, and even 3D virtual models. These platforms are already mature enough to help launch virtual influencers with unique characteristics, including their fan base, areas of expertise, and ability to generate art and content. Therefore, I do not think this is a short-lived phenomenon.

Sonal: The decentralized physical infrastructure network (DePIN) you mentioned is a very interesting direction. It is not just a vision of the future of science fiction, but is also very useful in many practical scenarios.

Carra: There are many DePIN (decentralized physical infrastructure network) projects that are gradually becoming decentralized. These projects have begun to rely on large language models (LLMs) and computer vision techniques to verify physical network resources. In the medium term, we can envision a decentralized network of human validators that assess risks and resolve suspicious behaviors through actionable AI workflows. In the longer term, AI agents may be equipped with their own wallets, keys, and computing resources to directly take over these verification tasks and become completely independent nodes or validators.

For example, there is a project called Daylight, an energy-focused DePIN company that sells data about home distributed energy resources, such as Tesla Powerwall storage devices, solar panels, and smart thermostats, to large energy companies.

Imagine if we all run our own peer-to-peer Daylight network, and my energy grid needs to communicate with your grid, and I need to prove that I do have a Tesla Powerwall or solar panels, and that my thermostat only uses part of the energy from the solar panels, and that I have excess power bandwidth. So how do I prove it? The simplest way at the moment is to take a photo of the energy meter. This photo will convey information to my grid, and then my grid will tell your grid: "I have excess bandwidth that can be shared." But if someone falsifies data, such as claiming that there are these resources when there are none, it will cause system failures and may even trigger penalties for nodes. Therefore, if we want to truly implement peer-to-peer DePIN, we need a reliable way to verify the authenticity of this data.

Currently, this verification method is still centralized. Users need to upload photos of their energy meters regularly, and then use RAGs (Retrieval Augmented Generation) technology to verify whether these photos are authentic. RAGs is a technology that combines external facts to enhance AI models. For example, it can ensure that this picture has not appeared in the database before, or that it is not a picture found by the user from Google. There are other similar examples, such as Nash, a company we supported in the CSX Incubator. They developed a decentralized version of the Doordash platform and also used RAGs technology to parse receipts and proof of delivery. Although Nash's system does not currently use a fully autonomous AI workflow, they are already using computer vision technology to parse images and compare them with vector databases.

In the future, when Daylight is fully decentralized, they will rely on a distributed network of validators to ensure that all data submitted by users is authentic. At the same time, it is conceivable that in the future, when household users sell energy resources in a programmatic way, these AI agents will become the core role in maintaining the entire network, ensuring the efficiency and credibility of the system.

Sonal: By the way, overall, the current distribution of energy has actually achieved a certain degree of distributed mode, but there are still many parts that are invisible to the power grid. For example, the location of some nodes may be unknown, and they may be at the edge of the network. This is a common problem in network systems - these nodes may be offline or have some valuable information. Therefore, imagine if these AIs can operate in a consistent manner, especially to manage those nodes located in remote areas, this prospect is actually very exciting.

Carra: Absolutely. This vision has been around in the crypto and cypherpunk communities for a long time. There has always been an “OG” idea that each of us would be able to run our own node. The vision is that we would have a true peer-to-peer network where everyone could verify all the data on the chain and run their own node.

The future of identity verification: The boundary between humans and AI

Sonal: Eddy, your core point is that as more and more people use AI, we will need unique proof of “human identity.” We’ll let you elaborate on this important point, what it is, and why it matters.

Eddy: I ​​think this is an interesting topic when it comes to AI because there is nothing new in AI when it comes to issues like impersonation and deception. Let me start with what is not new: a highly skilled adversary can indeed spend a lot of resources and time trying to deceive you. For example, imagine a fake phone call that sounds like it is from your parents.

This kind of thing could have happened 10, 20, or even 30 years ago. You just needed a skilled voice actor who knew what your parents sounded like, rehearsed, and hired a private investigator to dig up information about you and write a convincing script. This kind of thing has always been possible.

The difference between then and now is that now and in the future, the cost of doing these things is falling rapidly. The cost of creating this "replica" or building an interactive experience that can convince you to make a dangerous decision or a risky behavior is falling rapidly. And when the cost of any technology goes down, it makes it possible for more people to use it. In most cases, people use it to do things that are extremely productive and beneficial, and that's where economic progress comes from. But at the same time, it also makes these tools more accessible to attackers and people who want to use technology for bad things. So this change in cost means that we need to find something that is costly for attackers as an obstacle. We need to find new high-cost obstacles.

Sonal: But the key question is, how do you do this without increasing the cost for the average user? Especially the really well-intentioned, genuine users? How do you do this?

Eddy: Exactly. We need to find a way to increase the cost to attackers without increasing the cost to the average person or productive user.

One of the core ideas is that if someone wants to fake a lot of fake context, fake IDs, fake voices, fake phone numbers, fake videos, etc., in the future, they need an ID system to do this. And by ID system here, I mean in a broad sense. For example, spammers use a phone ID system when they call you. They need new fake numbers because every time they call a bunch of numbers and get reported, those numbers get marked as spam. That's why you don't get spam calls from the same number every day. If they used the same number, it would be easy to solve. You will find that spam calls always come from a new number, a new email address, or a new Twitter account. This is completely logical because the system learns and blocks these IDs. Therefore, the attacker needs to constantly get new IDs, and the cost of getting new IDs is a significant part of the cost structure of the attack. If the cost of new IDs is almost zero, then this behavior will surge. If the cost of new IDs is high, then this behavior will decrease. But obviously, we don't want the cost to be high for ordinary users. So we can introduce a cost by requiring users to prove that they are "human".

Why does this increase costs? It's because it's hard for computers to pretend to be human. And it's relatively easy for humans to act like humans. Historically, we've done this with things like CAPTCHAs. These are little tests that are meant to verify that a person is actually doing the action. But as machines have become smarter, CAPTCHAs have become easier to crack. And in some sensitive scenarios, we've seen spammers even outsource the completion of CAPTCHAs to real people, and then pass the completed cookie credentials to bots to continue their attacks.

This can make it more expensive for machines to attack by having some way to prove that someone is human when they obtain these credentials. However, this is not a complete solution. The key problem is that if a person can easily obtain multiple IDs, such as a second, third, fourth, or even tenth or twentieth ID, then these attacks will still be profitable. People can continue to obtain these new IDs and give them to scammers or bots to attack. Therefore, the key is not just to test "human identity", but also to ensure that each person can only have one unique ID.

Robert: That way someone can't answer multiple CAPTCHAs to get IDs for all the bots. Because they could sit there all day answering CAPTCHAs for the bots, and that doesn't solve the uniqueness problem.

Eddy: Exactly. Someone might say, "Oh, can I buy an ID from the market? A unique human identity verification." Yes, a market will form. But there are things we can do to disrupt the formation of this market. For example, would you sell your passport to someone? Of course, but if your passport is important to you, selling it will cause you a lot of losses. If there is a way for you to invalidate your old passport and get a new one, then buyers may be reluctant to buy because they don't trust you and you may destroy their use of it by obtaining a new passport. Still, in addition to disrupting the formation of the market, if it is easy to get one ID, but it is difficult to get a second ID, then the total number of IDs available for purchase on the market will decrease, which will cause the cost to rise, thereby increasing the cost of attackers.

Sonal: Great! How do you do this at scale? As you mentioned earlier, it’s not easy to distinguish between humans and machines on the Internet. As the classic saying goes, “On the Internet, you don’t know if the other person is a dog or a human.” So how do you determine that a person is an individual?

Eddy: That's a good question. I think that this theoretical framework doesn't require a commitment to a particular way to do this. There are some very promising ways, such as using biometrics, or relying on national IDs and government IDs, which often also contain biometric information. One approach that I'm not sure is very effective, although it is very popular, is the so-called web of trust model. In simple terms, this model verifies identity through certification by others, such as "I know Robert is human," "I know Sonal is human," and so on. But the problem is that this model can only ensure that a certain ID in the network corresponds to a person, but it cannot ensure that this person has only one ID. For example, I can go to your Twitter account and say, "I know this is Sonal's account." But I can't confirm that this is her only account.

Therefore, we usually need additional layers and technical means. I would also like to emphasize that this is often overlooked in discussions about this topic, that is, there are many technical means that can ensure that this "unique human identity verification" is private. I believe that ensuring privacy is a key point that cannot be compromised when building such an ID system. If each person only has one ID, this will deprive them of anonymity in many cases and will also affect their reasonable right to privacy. And I think this is neither realistic nor acceptable in the future Internet. Therefore, an ideal system should ensure privacy, uniqueness, and can only be obtained by one person. Such a system can even allow users to use multiple pseudonyms on different websites, but it needs to be constrained within a certain range and visible to the network. In this way, the network can identify whether a certain ID has created a large number of spam accounts and restrict it, such as rate limiting and spam protection.

Sonal: By the way, I have a question about the role of crypto in this. When you mentioned using biometrics, I immediately thought of one of the most common examples that we use today, such as the Clear ID system, which enables retina scanning when passing through airports. Why do we need crypto? Why can't we just use such a system?

Eddy: It is indeed possible to do without encryption. For example, Clear, I am not sure how it works, but I am sure they have some kind of database to store the information they need. The reason for the need for encryption is that the representation of the ID space (namespace) we are talking about needs to be censorship-resistant, that is, we want anyone to be able to modify, edit and update this space. In addition, we want these updates to be published in a completely neutral place, a place where no specific individual or organization can manipulate or use the system for their own benefit or others. Basically, this can be achieved through a public blockchain. Think of it as an extremely neutral "golden bulletin board", which is a space suitable for this purpose.

I will say, however, that the topic of “unique human identity verification” and related research is not unique to crypto. Many researchers are working on this problem from many angles. But I think implementing such a system on a blockchain can bring many important benefits, such as censorship resistance, timestamping records, and trusted neutrality, which make it possible to operate at scale. Whenever an AI agent works for a person, we need to know who it is working for.

When a person interacts with you, think about it, you judge this person based on your model of the world, such as why this person is approaching you, the environment they are in, who introduced them, etc. This contextual information helps you understand this person. And what I mentioned earlier, faking context is exactly what imposters use to deceive others. If an AI agent approaches you, or more sci-fi, an AI agent approaches your AI agent, and you want them to interact in some beneficial way, then they need to know information about each other. This requires some kind of authentication mechanism to ensure that they know who authorized them to perform certain tasks. Therefore, I think we need this "human authentication" system, supplemented by accountability mechanisms, etc., so that we know who is interacting with whom.

Sonal: Eddy, quick question, how do governments think about this kind of system? You mentioned that this can be done in a decentralized way, and there are multiple ways to do this, and governments can be involved. I love that you mentioned censorship resistance, but that's good for good governments, not necessarily for bad governments. So I'm curious, what do you think about how governments think about these identity systems?

Eddy: I ​​think governments will want to adopt some of the cryptographic tools that I mentioned, for example, and that makes a lot of sense for them. Of course, governments will want to have full control over the databases that register and store all the data. They may also like the accountability mechanisms that I mentioned. For example, I think governments do try to ensure that passports are issued uniquely. I don't think they would accept that one person gets two passports from the same country with different information. That's something they would resist. But it's not something that's explicitly guaranteed in the design of their systems. If you look at the specifications and descriptions of national ID systems, they don't usually highlight this as a key feature. But I think it's implicit in the design. A very interesting question is, if we rely on national ID systems for this kind of verification, at some point in time, when it becomes extremely valuable for an ID to reflect a person's identity, will small governments be willing to forge or subvert their own ID systems in order to provide privileges to machines that work on their behalf? We really need to think about at what point does this identity become so valuable that it can be abused. I think it's going to be very important to have a human ID in the future.

Sonal: I think this can be viewed from a very positive perspective. For example, the government may be very willing to adopt this system. As you mentioned at the beginning of your answer, there are many reasons why the government wants this system. For example, there are many cumbersome processes involved in applying for a passport or driver's license, which is inefficient and has a lot of friction. If this system is based on a decentralized blockchain owned and operated by the public, it will be more resilient and secure, and it will also bring many derivative tools to help the government improve efficiency.

Eddy: I ​​totally agree with that. I hesitate a little because we can design a programmable and useful database system without blockchain. My intuition is that governments might prefer to have tighter control over such a system. But it is certainly conceivable that there are some extreme cases where governments might prefer to trust other governments with information stored on the same system, in which case they might choose to use blockchain.

Decentralized Autonomous Chatbot

Sonal: Karma's concept is "Decentralized Autonomous Chatbots". It sounds like it's going a step further than just giving AI a wallet, like running in a Trusted Execution Environment (TEE). Can you explain the role of TEE and the idea you're exploring?

Karma: Simply put, we are thinking about how to provide freedom of speech and business freedom for AI. At present, when we see some so-called decentralized chatbots, or someone claims that they have an AI running a social media account, it is difficult for us to judge whether this is really AI speaking autonomously. For example, is there a human behind it? Is there a human filtering the content? These questions are always difficult to answer. And this is where TEE can solve the problem. With TEE, you can generate keys and passwords in a secure enclave of hardware. Even if you run the AI's calculations, you cannot directly control it. This method can prove that these keys are securely generated in hardware and cannot be accessed from the outside. Although you can still shut down the system, you cannot manipulate its behavior. You can choose to filter its output, such as generating 100 pieces of content and then choosing one of them to publish, but you cannot interfere with the process of its generation of content. Therefore, the core of this idea is how to remove "human intervention" as much as possible. One way to achieve this is to remove control over computing power, such as designing a decentralized network that anyone can join. If participants can prove that they have performed the reasoning task correctly and used the correct model, they can get rewards. In this way, when AI generates tokens, it can be ensured that these tokens are correctly generated by AI.

Robert: Why is it so important whether a chatbot operates autonomously or has human intervention behind it?

Karma: That’s a good question. We’re mainly thinking about what the ultimate goal of this technology is. With today’s technology, this is no longer science fiction, but completely feasible.

Sonal: Karma, the TEE (Trusted Execution Environment) you mentioned is very interesting. It has actually been around for a long time. For example, in distributed applications, TEE can protect the security of part of the calculation without worrying about the security of the entire environment. In recent years, TEE has regained attention in many new scenarios. So, is it necessary to use TEE now? Or can other technologies be used to achieve the functions you described?

Karma: TEE is a very interesting technical tool because it runs at hardware speed, which is very attractive. Of course, we are also studying other technologies, such as cryptographic technologies such as KVM (kernel-based virtual machine), FHE (fully homomorphic encryption), and MPC (multi-party computation). But these methods usually come with huge performance overheads, such as 1,000 times slower than hardware execution or more. Therefore, TEE is a good trade-off. It is real hardware, runs fast, and the hardware is now cheap and common. More importantly, TEE provides some cryptographic guarantees, such as the keys are generated within the TEE, are real and secure, and cannot be accessed from the outside. Even if you have access to the hardware, you cannot get the keys. In fact, almost all devices are now equipped with TEE, such as smartphones, laptops, etc.

This is a very interesting pragmatic trade-off because we do face some technical challenges, and TEE (Trusted Execution Environment) may be a key factor in solving these problems. Of course, you can also combine other technologies. For example, TEE provides a hardware isolation mechanism. Although you are running hardware, you cannot control the specific operations inside the hardware.

This is an interesting technical primitive. You can also incorporate MPC (multi-party computation) technology, such as splitting the key across multiple nodes, with different entities holding parts of the key. In this way, no single entity can fully control the entire system. This privacy protection technology is already mature and can help us achieve the operation of autonomous systems in a verifiable way.

For me, this is what’s most interesting: How do we tell if a system is fully autonomously run by AI or if there are still humans behind the scenes? With these techniques, you can clearly know that the system is fully autonomous and has become a real internet entity.

Sonal: We talked about the example of Truth Terminal before, which has a wallet and can act autonomously. AI can run in an autonomous way in games and other applications, and can even control infrastructure, such as applications in decentralized network nodes (DePIN nodes). Your perspective is very inspiring. In theory, if we extend this sci-fi future, although no one can predict how it will eventually develop, in theory, this AI could even become the first billion-dollar entity. It can become a completely independent business entity, making its own decisions and actions without any human intervention. It's amazing.

Karma: Yes, if we can prove that this system completely removes "human intervention" and it runs in a truly permissionless network, then we have achieved true freedom of speech and business freedom. Such an AI can freely express its will on the Internet, earn and spend money, and even hire humans to complete tasks. One day, you may say: "My boss is an AI." You can work for this synthetic entity over the Internet and make money.

Sonal: Yes, one last thought I have is that it is very interesting that you call it "Decentralized Autonomous Chatbot". I remember before the term DAO (Decentralized Autonomous Organization) was popularized, there was a similar term called "Decentralized Autonomous Corporation". But then it was gradually replaced by DAO. But your abbreviation DAC is the same as "Decentralized Autonomous Corporation", which is actually quite interesting. Because you