Original Title: AI x crypto crossovers
Original Author: a16z Crypto, Scott Duke Kominers, Sam Broner, Jay Drain, Guy Wuollet, Elizabeth Harkavy, Carra Wu, Matt Gleason
Original Translation: BUBBLE, BlockBeats

The economic model of the Internet is changing. As the open network gradually collapses into a "prompt bar," we have to consider: Will AI lead to an open Internet, or will it descend into a new paywall labyrinth? And who will control all of this—large centralized companies or a broad user community?

This is where cryptographic technology can intervene. We have discussed the intersection of AI and cryptography many times; in short, blockchain is a new way to build Internet services and networks that are decentralized, structurally neutral, and user-owned. They provide a counterbalance to the increasingly centralized trend in current AI systems by renegotiating the economic relationships behind the system, helping to achieve a more open and robust Internet.

The idea that "crypto can help build better AI systems, and vice versa" is not new, but it often lacks a clear definition. Some intersection areas, such as verifying "proof of humanity" in the context of the widespread adoption of low-cost AI systems, have begun to attract the attention of builders and users. Other use cases may take several years, or even decades, to materialize. Therefore, in this article, we have compiled 11 practical use cases of the intersection of AI and cryptography, aimed at promoting in-depth discussions on questions such as "what is possible" and "what challenges still need to be addressed." These use cases are all based on the technology currently being developed, whether it is dealing with massive micropayments or ensuring that humans can have a relationship with future AI.

Identity

Persistent Data and Context in AI Interactions

Author: Scott Duke Kominers

Generative AI relies on data-driven approaches, but in many applications, context (i.e., the state and background information relevant to a particular interaction) is equally important, if not more critical.

Ideally, an AI system (be it an agent, an LLM interface, or another application) should be able to remember the type of project you are working on, your communication style, your preferred programming language, and many other details. However, in reality, users often have to repeatedly rebuild this context in different sessions of the same application (such as when you start a new ChatGPT or Claude session), not to mention when switching between different systems.

Currently, the context between different generative AI applications is essentially non-transferable.

However, with the help of blockchain, AI systems can preserve key contextual elements as persistent digital assets, load them each time a session is initiated, and seamlessly transfer them across multiple AI platforms. Moreover, blockchain may be the only technology solution that is both forward-compatible and naturally emphasizes interoperability, these features being core attributes of blockchain protocols.

A natural application scenario is in AI-participating games and media, where user preferences (from game difficulty to key bindings) can remain consistent across different games and environments. But the more valuable scenarios lie in knowledge-based applications, where AI needs to understand the knowledge users already possess and their learning methods; or in more professional AI usage contexts, such as programming. Although some companies have developed custom AI bots that can maintain context within a certain range, this context often cannot be transferred between different AI systems within the same enterprise.

Companies are just beginning to realize this issue, and the closest thing to a general solution currently is custom robots with fixed contexts. Internally on platforms, practices of sharing context between different users are starting to emerge off-chain, for example, the Poe platform allows users to rent out their custom bots to others.

If this kind of behavior is migrated onto the chain, we can allow AI systems we interact with to share a "context layer" composed of key elements of all our digital behaviors. AI systems will be able to instantly understand our preferences, better fine-tune for us, and optimize the interaction experience. In return, just like registering intellectual property on-chain, allowing AI to reference persistent on-chain context will also inspire a new market interaction around keywords and information modules. For example, users can directly authorize or monetize their expertise while retaining ownership of the data. Of course, shared context will unlock many possibilities that we have not yet imagined.

Universal Identity for AI Agents

Author: Sam Broner

Identity, the authoritative record of "who" or "what," is the silent foundational structure supporting today's digital discovery, aggregation, and payment systems. As platforms abstract this infrastructure behind the scenes to operate, we only experience its presence in the end products: Amazon assigns unique identifiers (ASIN or FNSKU) to products, pools them for display, and aids in discovery and payment; Facebook is similar: user identities form the basis of its content recommendations, Marketplace product displays, and organic content and ad discovery.

But as AI agents evolve, this situation will change. Enterprises are deploying AI agents in multiple scenarios like customer service, logistics, and payments, and the form of platforms is transitioning from a single interface to a distributed system across platforms and endpoints. These agents will accumulate rich context to perform more tasks for users. If an agent's identity is tied to only one platform or marketplace, it will struggle to operate effectively in other critical contexts – such as email conversations, Slack channels, or other products.

Therefore, an AI agent needs a unified, portable "passport." Otherwise, we won't be able to recognize its payment methods, confirm its version, query its capabilities, know who it is acting on behalf of, or track its cross-platform reputation. An agent's identity should encompass the functions of a wallet, API registry, update log, and social proof—all interfaces (be it email, Slack, or another agent) should be able to identify and interact with it consistently.

Without a unified "identity" primitive, each integration has to build underlying structures from scratch, discovery mechanisms still rely on happenstance, and users lose context when switching between different platforms.

We are in a phase where we can redesign agent infrastructure "from first principles." So, how do we build a richer, trusted neutral identity layer than DNS records? We should not rebuild those "monolithic platforms" that bundle identity, discovery, aggregation, and payment together; instead, agents should be able to freely receive payments, list their capabilities, and coexist across multiple ecosystems without the fear of being locked into one platform.

This is where AI intersects with crypto: the "permissionless composability" provided by blockchain networks can help developers build more useful agents and a better user experience.

Of course, currently, those vertically integrated platforms (like Facebook or Amazon) still offer a better user experience—because one of the complexities of building a high-quality product is ensuring all modules work together seamlessly. But this convenience comes at a high cost. Especially as the costs of building, aggregating, monetizing, and distributing agents continue to decrease, and the reach of agent applications expands, a trusted neutral identity layer will grant entrepreneurs a true sovereignty "passport" and encourage more exploration and innovation in distribution and design.

Future-Proof "Proof of Personhood"

Authors: Jay Drain Jr. and Scott Duke Kominers

With the widespread penetration of AI—whether driving bots and intelligent agents in network interactions or creating deepfakes and manipulating social media—it is becoming increasingly difficult for people to discern whether online interactants are human or algorithmic. This erosion of trust is not a distant possibility; it is already here. From astroturfing on X (formerly Twitter) to bots on dating apps, the boundary between reality and virtuality is becoming blurred. In this environment, "Proof of Personhood" (PoP) is gradually becoming a critical infrastructure.

Currently, one way to verify a person is human is through the use of digital identity (such as the centralized identity systems used by the U.S. TSA). Digital identity includes various pieces of information that a user can use to verify their identity—username, PIN, password, third-party authentications (such as citizenship or credit records), and other credentials. The value of decentralization in this context is evident: when this data is centrally managed, identity issuers can revoke access, charge fees, and even assist in monitoring; decentralization reverses this structure, putting users, not platforms, in control of their identity, making it more secure and less susceptible to scrutiny.

Unlike traditional identity systems, a decentralized "Proof of Personhood" mechanism (such as Worldcoin's World ID system) allows users to independently manage and secure their identity data, verifying themselves as real individuals in a privacy-friendly, trust-agnostic manner. Similar to a driver's license, once issued, PoP can be universally accepted on any platform, anytime, anywhere. This blockchain-based PoP thus exhibits future-forward compatibility, specifically in two aspects:

 Portability: PoP follows an open protocol, allowing integration on any platform. Since users control their identity on a public infrastructure, it is fully portable, enabling access on any existing or future platform.

 Permissionless Accessibility: Any platform can independently choose to recognize this PoP identity without requiring centralized API authorization, thus mitigating the risk of certain use cases being denied.

The primary current challenge in this field is user adoption: while we have not yet seen widespread scalable real-world use cases of Proof of Personhood, we believe that once the user count reaches a critical mass, with several key partners and certain "killer apps" driving the momentum, the adoption of PoP will accelerate rapidly. Every application adopting a PoP standard enhances the practical value of that identity, attracting more users to claim it, which, in turn, incentivizes more applications to integrate the standard, creating a network effect of rapid growth. (Given that on-chain identity is inherently designed for interoperability, this effect will be even more explosive.)

We have already seen some mainstream consumer applications and services, especially in gaming, social, and dating sectors, announcing partnerships with World ID to assist users in verifying that they are interacting with a real person, perhaps even the specific individual they expect. Simultaneously, new identity protocols continue to emerge, such as the Solana Attestation Service (SAS). While SAS itself is not a PoP issuer, it enables users to associate off-chain data (such as KYC verification required for compliance or investment qualifications) with their Solana wallet in a privacy-preserving manner, laying the groundwork for building a decentralized identity framework.

All these signs indicate that the tipping point for decentralized PoP may be on the horizon.

The significance of PoP goes beyond merely "banning bots"; it is a crucial mechanism for delineating a clear boundary between the human network and the AI network. It enables users and applications to distinctly differentiate between "human-to-human interaction" and "human-to-machine interaction," thereby fostering a more secure, authentic, and wholesome experience in the digital world.

Decentralized Infrastructure for AI

Decentralized Physical Infrastructure for AI (DePIN)

Author: Guy Wuollet

While AI is a digital service, its development is increasingly constrained by physical infrastructure. The Decentralized Physical Infrastructure Network (DePIN) as a new model for constructing and operating real-world systems is helping to democratize the compute infrastructure on which AI innovation relies, making it cheaper, more resilient, and more resistant to censorship.

Why is this the case? The two main bottlenecks in AI development are energy and chip access. Decentralized energy can help unlock more power resources, and developers are leveraging DePIN to aggregate idle chip resources from sources like gaming PCs and data centers. These computing devices can collectively form a permissionless compute market, creating a more equitable environment for AI product development.

Other use cases include distributed training and fine-tuning of Large Language Models (LLMs) and a distributed network for model inference. Decentralized training and inference not only reduce costs (by utilizing originally idle compute power) but also provide censorship resistance, ensuring developers are not shut down for relying on hyperscalers.

The highly centralized nature of AI models in the hands of a few companies has long been a concern; decentralized networks help build a more cost-effective, censorship-resistant, and scalable AI ecosystem.

Providing Infrastructure and Guardrails for Interaction Between AI Agents, Endpoint Service Providers, and Users

Author: Scott Duke Kominers

As AI tools become increasingly adept at handling complex tasks and multi-step chains of interaction, they will increasingly need to interact autonomously with other AIs rather than relying on human controllers.

For example, an AI agent may need to access data related to specific computations or recruit another AI agent specialized in a certain task—such as enlisting a statistical robot for model simulation or invoking an image-generating bot in the process of creating marketing materials. AI agents will also create significant value for users by carrying out entire transaction flows or activity sequences—such as finding and booking flights based on user preferences or discovering and purchasing a new book tailored to their taste.

Today, there is no mature, general agent-to-agent market—such interactions are mostly confined to explicit APIs or closed ecosystems that support a limited number of internal agent calls.

A more common issue is that currently, most AI agents are in isolated systems with closed interfaces and lack of architectural standards. However, blockchain technology can help establish open standards for protocols, which is crucial for short-term adoption. In the long term, this also supports "forward compatibility": as new AI agents continue to evolve and emerge, they can still access the same underlying network. Due to blockchain's interoperable, open-source, decentralized, and easily upgradable architecture, it can more quickly adapt to the transformative innovation of AI.

Several companies are already building a blockchain "track" for agent-to-agent interaction: for example, Halliday has introduced a protocol that provides a standardized cross-chain architecture for AI workflows and interactions, with guardrails at the protocol level to ensure AI does not deviate from the user's intent. Catena, Skyfire, and Nevermind support payment interactions between AI agents without human intervention. Coinbase has also begun to provide infrastructure support for such projects.

Keeping AI / vibe-coding Applications in Sync

Authors: Sam Broner and Scott Duke Kominers

In recent years, the explosive development of generative AI has made software development easier than ever. Coding efficiency has increased by several orders of magnitude, and more importantly—natural language programming is now possible, allowing even non-programmers to fork existing programs or build entirely new applications from scratch.

However, while AI-assisted programming brings new opportunities, it also introduces a significant amount of "entropy" into and between programs. So-called "vibe coding" simplifies the complex network of underlying dependencies but can also lead to functional or security issues when underlying components are updated. Additionally, as more and more people use AI to create personalized applications and workflows, interactions between different user systems become more challenging. In fact, even if two vibe-coded programs have the same functionality, their operational logic and output structure may vary widely.

In the past, the standard ways to ensure consistency and compatibility were file formats and operating systems, and more recently shared software libraries and API interfaces. But in a world where software is constantly evolving, morphing, and forking in real time, these standardization layers need to have widespread accessibility and continuous upgradability while maintaining user trust. Moreover, relying solely on AI cannot solve the challenge of incentivizing people to maintain these connections and compatibility.

Blockchain provides a solution to address both of these issues simultaneously: embedding a "protocolized sync layer" into users' custom software and ensuring cross-application compatibility through dynamic updates. Previously, a large enterprise might have had to pay millions of dollars to hire a system integrator (like Deloitte) to customize a Salesforce system. Today, an engineer might create a sales data visualization interface in just a weekend. But with the proliferation of personalized software, developers will also need help to keep these applications in sync and running smoothly.

This is somewhat akin to how current open-source software libraries operate, but with real-time updates instead of periodic ones—and with an incentive mechanism. All of this can be achieved through crypto. Like other blockchain-based protocols, shared ownership encourages participants to actively contribute to protocol improvements. Developers, users (or their AI agents), and other consumers can be rewarded for introducing, using, enhancing new features, and integrations.

In turn, shared ownership also aligns every user with the overall success of the protocol, creating a kind of "anti-tragedy" mechanism. Just as Microsoft would not readily disrupt the .docx file format standard because it would affect users and brand reputation, the collective owners of the protocol would also not easily introduce bad or malicious code.

As with various software standardization architectures we've seen in the past, there is enormous network effect potential here. As the AI programming software "Cambrian Explosion" continues to advance, the number of heterogeneous systems needing to communicate with each other will skyrocket.

In short: vibe programming needs to stay in sync, relying not just on vibe. Crypto is the key.

New Economic and Incentive Models

Micro-Payment Mechanism Supporting Revenue Sharing

By Liz Harkavy

AI agents and tools like ChatGPT, Claude, and Copilot provide us with a new and convenient way to navigate the digital world. But for better or for worse, these technologies are disrupting the economic system of the open internet. We have already seen the initial signs of this trend—for example, some educational platforms have experienced a significant drop in traffic as students have turned to using AI tools; and several U.S. newspapers have sued OpenAI for copyright infringement. If we cannot readjust the incentive mechanisms, the internet will become more closed: more paywalls, fewer content creators.

Of course, the problem can also be addressed through policy, but alongside the legal process, some technological solutions have begun to emerge. Among the most promising (and challenging technically) is perhaps embedding a revenue-sharing mechanism directly into the internet architecture. When a transaction is facilitated by AI-driven behavior, the content creator providing the source of that behavior should receive a corresponding share. This has been reflected to some extent in the affiliate marketing system, which can track sources and share revenue; a more advanced version could automatically track and reward all contributors in the information chain. Blockchain can evidently play a crucial role in such a "source traceability" mechanism.

However, such systems require building new infrastructure—especially micro-payment systems that can handle extremely small transactions, attribution protocols that can fairly evaluate different types of contributions, and governance models that ensure transparency and fairness. Some blockchain-based tools have shown potential in this "source traceability" mechanism, such as rollups, L2 scaling solutions, AI-native financial institution Catena Labs, and the financial infrastructure protocol 0xSplits—they can achieve almost zero-cost transactions and more granular revenue splits.

AI Represented by Webcrawler Should Compensate Content Creators

Author: Carra Wu

Currently, among AI agents, the most market-demand-friendly are not programming assistants or entertainment tools, but Webcrawlers — they automatically browse the web, collect data, and autonomously decide which links to visit.

It is estimated that nearly half of the current internet traffic comes from non-human sources. Bot programs often ignore the robots.txt file (theoretically used to indicate whether the crawler is allowed to crawl that site) and use the collected data to support the core competitiveness of the world's largest tech companies. What's worse is that websites themselves end up footing the bill for these "uninvited visitors," bearing the cost of bandwidth and server resources. Therefore, CDN providers like Cloudflare have had to introduce a series of blocking services. This is now a fragmented, cumbersome adversarial mechanism, but in reality, it could be replaced by a more reasonable system.

We have pointed out that the internet's original "economic contract" — the mutually beneficial relationship between content creators and platforms — is on the verge of disintegration. This is also reflected in the data: in the past year, an increasing number of websites have begun actively blocking AI crawlers. In July 2024, only 9% of the top 10,000 websites that blocked AI scrapers did so; now, this proportion has risen to 37% and is rapidly increasing.

So, can we stop indiscriminately blocking all suspicious robot requests and instead seek a balance in the middle ground? A new model is: AI crawlers no longer "freeload" web content but pay for data retrieval behavior. Blockchain can serve as the execution layer for this model: each crawler agent holds cryptocurrency and, when accessing a website, initiates on-chain negotiation with the site's "gatekeeper agent" or paywall system through the x402 protocol.

The problem is that robots.txt (also known as the "robots exclusion standard") has been an industry default practice since the 1990s, and overturning it would require large-scale industry coordination or the intervention of CDN providers like Cloudflare. However, on the other hand, we can open a separate channel for human users: they can prove their "human identity" through World ID (as mentioned above) and can continue to access content for free.

As a result, the behavior of AI content collection can achieve compensation for creators at the collection point, while human users can still enjoy an "information-free" internet.

More Privacy-Preserving Advertising: Precise Yet Non-Intrusive

Author: Matt Gleason

AI has already been changing the way we shop, so can advertising also be more "useful"? Many people dislike ads because they are either irrelevant or too intrusive. Even with "personalized ads," if they are too precise and based on a large amount of personal data, they can feel like a "violation of privacy."

Some applications are trying to monetize through a paywall (such as watching videos or unlocking game levels). Cryptographic technology can help us reshape this logic. When combined with blockchain, a personalized AI agent can deliver ads based on user-set preferences without exposing user privacy data; at the same time, users can be rewarded with cryptocurrency after voluntary interaction.

Technically, this model requires:

 Low-cost digital payment system: Ad interaction rewards must support high-frequency micropayments, the system must have high-speed, low-cost characteristics;

 Privacy-preserving data validation mechanism: The AI ad agent needs to verify if the user meets certain demographic features but cannot expose specific data, zero-knowledge proof (ZKP) technology can achieve this;

 New incentive model: If the micro-payment (\<$0.05) ad revenue model becomes popular, users can actively choose to view ads and benefit from them, shifting from "passive harvesting" to "voluntary participation."

Humans have long tried to make ads more useful, whether online or offline. Reshaping the advertising system into an "AI + Blockchain"-driven one finally holds the promise of making ads truly useful: non-intrusive yet profitable.

This will also make ad space itself more valuable, while having the potential to overturn the highly invasive "ad exploitation economy" today, and instead, build a human-centered system: where users are no longer the "product" but the "participants."

Shaping the Future of AI

AI Companions Owned and Controlled by Humans

Author: Guy Wuollet

Today, many people spend more time with devices than in face-to-face communication, and this time is increasingly spent interacting with AI models and AI-curated content. In fact, these models have already been providing some form of companionship—whether in entertainment, information retrieval, niche interest fulfillment, or children's education. It is easy to imagine that in the near future, AI companions will be widely used in fields such as education, healthcare, legal consultation, and social companionship, becoming a common form of interaction for humans.

Future AI companions will have infinite patience and can be highly customized based on individuals and their specific needs. They will not just be assistants or "robotic servants" but are more likely to become highly valued relationship objects for people. Therefore, who will own and control these relationships—whether it is the users themselves or companies and other intermediaries—becomes critically important. If you have been concerned about content curation and censorship by social media over the past decade, this issue will become more complex and personal in the future.

In fact, similar views have long been proposed (see here and here): Anticensorship-enabled custodial platforms like blockchain may be the clearest path to achieving an uncensorable, user-controllable AI. While individual users can run a local model and purchase GPUs themselves, most either cannot afford to do so or simply do not know how.

Although we are still some way off from the widespread adoption of AI companions, the technology to make all this happen is rapidly advancing: AI companions for text interactions have already shown great promise, and visual avatars have also seen significant improvements; blockchain performance is also gradually increasing. To make it easier for users to interact with uncensorable AI companions, we also need to continue improving the user experience (UX) of encryption applications. Fortunately, blockchain wallets like Phantom have already made on-chain interactions simple, while embedded wallets, passkeys, and account abstraction technologies allow users to have self-custodial wallets without having to manage their own seed phrases.

Furthermore, high-throughput, trustless computing technologies like Optimistic and zero-knowledge coprocessors will also enable us to establish meaningful and enduring relationships with digital companions.

In the near future, we will move from discussing "when will anthropomorphic digital companions and virtual identities appear" to "who has the right to control them, and in what manner."

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia