Authors: Scott Duke Kominers, Sam Broner, Jay Drain, Guy Wuollet, Elizabeth Harkavy, Carra Wu, and Matt Gleason
Compiled by: Aki Wu Shuo Blockchain
The economic structure of the internet is changing. As the open web gradually collapses into a "prompt bar," we must ask: will AI lead to a more open internet, or will it lead us into a maze of new paywalls? And who will control the future internet—large centralized companies, or broad user communities?
This is where crypto comes in. We have discussed the intersection of AI and crypto many times, but in short, blockchain is a way to redesign internet services and network architecture, enabling the construction of decentralized, credibly neutral, and user-“ownable” systems. By reshaping the economic incentives behind today's systems, blockchain provides a check and balance against the increasing centralization in AI systems, thereby promoting a more open and resilient internet.
The idea that "crypto can help build better AI systems, and vice versa" is not new—but it has long lacked clear definition. Some intersections (such as how to verify "human identity" in the context of a proliferation of low-cost AI systems) have already attracted significant developer and user attention. But other applications may take years or even decades to materialize. Therefore, this article shares 11 intersectional application scenarios of AI and crypto, hoping to spark more industry discussion: which are feasible, which challenges remain, and how they might evolve in the future.
These scenarios are all based on technologies currently under development—from processing large volumes of micropayments to ensuring humans retain ownership in their future relationship with AI.
1. Introducing Persistent Data and Context into AI Interactions
Scott Duke Kominers: Generative AI relies fundamentally on data, but in many applications, "context"—the state and background information related to an interaction—is often as important as the data itself, if not more so.
Ideally, agents, LLM interfaces, or other types of AI applications should be able to remember vast amount of personalized information, including the types of projects you are working on, your communication habits, preferred programming languages, etc. But in reality, users often have to repeatedly rebuild this context—not only when starting new sessions within the same application, like opening a new ChatGPT or Claude window, but especially when switching between different AI systems.
Currently, context from one generative AI application is almost impossible to migrate to another.
Using blockchain, AI systems can store key contextual elements as persistent digital assets, allowing them to be loaded at the start of a session and seamlessly migrated across different AI platforms. Furthermore, because "forwards-compatibility" and "interoperability promises" are core features of blockchain protocols, blockchain may be the only technology path that systematically solves this problem.
An intuitive application is in AI-led gaming and media, where user preferences (like difficulty, key bindings, etc.) can persist across games and environments. But the truly high-value applications are in knowledge-based scenarios—where AI needs to understand a user's knowledge system, learning style, and capabilities; and more specialized scenarios, such as programming assistance. Although some companies have built customized AI tools with "global context" for their own operations, this context still cannot be effectively migrated between the different AI systems used within the organization.
Various organizations are only just beginning to truly realize this problem, and the closest thing to a general solution currently is custom bots with fixed, persistent context. However, context portability between users on a platform is already emerging off-chain; for example, on the Poe platform, users can rent out their custom bots to other users.
If this type of activity moves on-chain, then the AI systems we interact with will be able to share a contextual layer composed of key elements of our entire digital behavior. AI will be able to instantly understand our preferences, allowing for better fine-tuning and experience optimization. Conversely, mechanisms like an on-chain intellectual property registry, if they allow AI to reference on-chain persistent context, could spawn new, more sophisticated market interaction models around prompts and information modules—for example, users could directly monetize their professional expertise through licensing while maintaining data self-management.
Of course, as context-sharing capabilities improve, they will also spawn a host of new use cases and possibilities that are currently unforeseeable.
2. A Universal Identity System for Agents
Sam Broner: Identity—the canonical record of "who or what" something is—is the underlying infrastructure that supports today's digital discovery, aggregation, and payment systems. But because platforms enclose this "plumbing" within their systems, users typically only experience identity within a finished product interface. For example, Amazon assigns identifiers (like ASIN or FNSKU) to products, integrates them into a unified interface, and helps users with discovery and payment; Facebook is similar: user identity determines their news feed content and forms the basis for discovering various content within the app, including Marketplace listings, organic content, and ad targeting.
With the rapid evolution of AI Agents, this landscape is about to change. More and more businesses are using agents for customer service, logistics, payments, etc. Their platforms will no longer be traditional "single-interface apps," but will be distributed across multiple channels and platforms, continuously accumulating deep context, and performing more tasks on behalf of users. But if an agent's identity is tied only to a single platform or single market, it will be difficult to use in other critical environments (like email threads, Slack channels, or inside other products).
Therefore, agents need a unified, portable "digital passport." Without it, there is no way to confirm how to pay the agent, verify its version, query its capabilities, identify who it is acting on behalf of, or track its reputation across applications and platforms. An agent's identity system must combine the functions of a wallet, an API registry, a changelog, and social reputation proofs, enabling any interface (whether email, Slack, or another agent) to parse and communicate with it in a consistent manner.
Without this shared "identity primitive," every system integration would need to rebuild this plumbing from scratch; content discovery would remain ad hoc; and users would constantly lose their critical context when switching between channels and platforms.
We now have the opportunity to design agent infrastructure from "first principles." So the question is: how to build an identity layer richer than DNS records and credibly neutral? Instead of rebuilding monolithic platforms that bundle identity, discovery, aggregation, and payment functions together, let agents be able to receive payments autonomously, publicize their capability lists, and exist in multiple ecosystems without fear of being locked into one platform.
This is where the intersection of crypto and AI can play a role—blockchain networks provide permissionless composability, enabling developers to create more powerful agents and a friendlier user experience.
Overall, vertically integrated solutions like Facebook and Amazon currently offer a better user experience—the reason being that part of the complexity of building great products is ensuring all components work together naturally from the top down. However, the cost of this convenience is becoming increasingly high, especially as the software cost to build, aggregate, promote, commercialize, and distribute agents decreases, and the reach of agent applications expands.
Achieving the user experience of vertically integrated platforms still requires significant effort, but once a credibly neutral agent identity layer is built, entrepreneurs can truly own their passport. This will also drive widespread experimentation and innovation in distribution models and interaction design.
3. Future-Proof "Proof of Personhood" (PoP)
Jay Drain Jr. and Scott Duke Kominers: As AI proliferates—whether through bots and agents running in various web interactions, or through deepfakes and social media manipulation—it becomes increasingly difficult to tell if the entities we interact with online are real humans. This erosion of trust is not a future worry; it is a present reality. From comment spam on X to automated accounts on dating apps, the line between real and fake is blurring. In such an environment, "Proof of Personhood" is gradually becoming key internet infrastructure.
One way to verify "you are human" is to use digital identity, including centralized identity verification systems used by agencies like TSA. Digital IDs encompass all information a user can use to prove their identity—usernames, PINs, passwords, and proofs issued by third parties (such as nationality, credibility, or credit status). The value of decentralization here is very clear: when identity data is stored in centralized systems, the issuer can revoke access, charge fees, or even assist in surveillance. Decentralization upends this structure: users, not platform gatekeepers, control their own identity, making it more secure and censorship-resistant.
Unlike traditional identity systems, decentralized Proof of Personhood mechanisms (such as World's Proof of Human launched by Worldcoin) allow users to manage their identity data themselves and verify that they are indeed "human" in a privacy-preserving, credibly neutral manner. Similar to a driver's license—which can be used in any scenario regardless of when or where it was issued—a decentralized PoP can serve as a universal underlying primitive, reusable on any platform, including those that don't exist yet. In other words, blockchain-based PoP is "forwards-compatible" because it provides:
Portability: The protocol is an open standard that any platform can integrate. Decentralized PoP can be managed by public infrastructure and is entirely user-controlled. This means PoP is inherently portable, compatible with any platform now or in the future.
Permissionless Accessibility: Platforms can choose whether to support a particular PoP identity on their own, without going through a centralized API approval process that might set discriminatory restrictions for different use cases.
The core challenge in this area is "adoption." Currently, there is no large-scale, real-world "Proof of Personhood" (PoP) application, but we expect that once the user base reaches a critical mass, with several early partners and killer apps that drive user demand, PoP adoption will accelerate significantly. Every application that adopts a certain digital ID standard increases the value of that ID type to users; this in turn drives more users to obtain the ID; and a larger user base increases the attractiveness for applications to integrate that ID standard to verify "humanness." (Furthermore, because on-chain IDs are designed to be interoperable, this network effect can spread rapidly.)
We have already seen mainstream consumer applications in gaming, dating, and social media announce partnerships with World ID to ensure that when users are gaming, chatting, or transacting, they are indeed interacting with real humans—or even the specific person they expect. At the same time, new identity protocols have emerged this year, such as the Solana Attestation Service (SAS). Although SAS itself is not a PoP issuer, it allows users to privately associate off-chain data (such as KYC results required for compliance, investor accreditation status, etc.) with a Solana wallet, building a user's decentralized identity. These signs suggest that the tipping point for decentralized PoP may not be far away.
Proof of Personhood means much more than "stopping bots." It aims to build a clear boundary between AI agents and human networks, enabling users and applications to distinguish between different interactions of "human and machine," thereby creating conditions for a better, safer, and more authentic digital experience.
4. Decentralized Physical Infrastructure (DePIN) for AI
Guy Wuollet: Although AI is a digital service, its development is increasingly constrained by physical infrastructure. Decentralized Physical Infrastructure Networks (DePIN)—a new model for building and operating real-world systems—have the potential to democratize the computing infrastructure that supports AI innovation, making it cheaper, more resilient, and more censorship-resistant.
Why? The two main bottlenecks for AI development are energy and chip access. Decentralized energy systems can provide more abundant power, and developers are already using DePIN to aggregate idle chips from gaming PCs, data centers, and other sources. These computing devices can together form a permissionless compute market, creating a level playing field for building new AI products.
Other application scenarios include: distributed training and fine-tuning of large language models (LLMs), and building distributed inference networks. Decentralized training and inference can significantly reduce costs because they utilize computing resources that would otherwise be idle. At the same time, such architectures are naturally censorship-resistant, ensuring developers are not "deplatformed" or restricted from access due to reliance on hyperscalers (i.e., centralized cloud infrastructure providers offering large-scale scalable computing resources).
The concentration of AI models in the hands of a few companies has been a long-term concern; decentralized networks can help build AI systems that are cheaper, more censorship-resistant, and more scalable.
5. Building Infrastructure and Security Mechanisms for Interactions Between AI Agents, End Service Providers, and Users
Scott Duke Kominers: As AI tools become increasingly capable of handling complex tasks and executing multi-level interaction chains, AI will increasingly need to collaborate independently with other AIs without direct human control.
For example, an AI agent may need to request specific data required for a computation, or need to call upon other agents with specialized capabilities to perform tasks—such as having a statistical analysis agent build and run model simulations, or mobilizing an image generation agent to help create marketing materials. AI agents will also create huge value in end-to-end transaction execution, such as completely replacing the user in completing a transaction process: finding and booking a flight based on preferences, or automatically discovering and purchasing a new book that matches the user's taste.
Currently, there is no "generalized agent-to-agent market." Such cross-agent requests can usually only be achieved through explicit API calls, or are limited to certain closed AI agent ecosystems as internal functions.
More broadly, most AI agents today operate in isolated ecosystems: APIs are relatively closed, lacking unified architectural standards. Blockchain technology can help protocols establish open standards, which is crucial for short-term adoption; in the long run, this also helps achieve forward compatibility: as new types of agents emerge, they can all connect to the same underlying network. Because blockchains are interoperable, open-source, decentralized, and generally easier to upgrade architecturally, they are better able to adapt to changes brought about by future AI innovation.
Several companies are currently building on-chain infrastructure for agent interaction. Take Halliday, for example, which recently launched a protocol that provides standardized cross-chain architecture for AI workflows and interactions, while incorporating protection mechanisms at the protocol level to ensure AI does not act beyond user intent. On the other hand, projects like Catena, Skyfire, Nevermind use blockchain to support automatic settlement between agents, enabling AI-to-AI payments without any human intervention. Similar systems are emerging, and Coinbase has also begun to provide infrastructure support for such development.
6. Keeping "Vibe Coding" Applications in Sync
Sam Broner and Scott Duke Kominers: The generative AI revolution has made building software easier than ever before. Coding speed has increased by orders of magnitude, and more importantly, coding can be done directly through natural language, allowing inexperienced developers to replicate existing programs or even build new applications from scratch.
However, while AI-assisted coding creates new opportunities, it also introduces a lot of "entropy" within and across programs. So-called "vibe coding" abstracts away the complex dependencies behind the software—but because of this, programs can expose risks in functionality and security when the underlying source code or inputs change. Also, when people use AI to create highly personalized applications and workflows, interfacing their systems with others becomes more difficult. In fact, even if two vibe-coded programs perform almost identical tasks, their operating logic and output structure may be completely different.
Traditionally, the work of ensuring consistency and compatibility was handled by file formats, operating systems, and later, software sharing and API integration. But in a world where software evolves, morphs, and branches in real-time, the standardization layer must have: broad accessibility, continuous upgradability, and also user trust. Furthermore, AI alone cannot solve the incentive problem—how to incentivize developers to build and maintain these links between systems.
Blockchain can solve both problems at once; it can provide protocolized synchronization layers that are embedded in user-customized software builds and can dynamically update as the context changes to ensure cross-system compatibility.
In the past, large enterprises might pay millions of dollars to system integrators like Deloitte to customize a Salesforce instance. Today, an engineer might build a custom "sales data viewer" interface in a weekend. But as the number of customized software continues to grow, developers will need help ensuring these applications remain synchronized and available.
This is similar to the development model of today's open-source software libraries, but the difference is: the synchronization layer is not dependent on periodic version releases, but is continuously updated—and also comes with incentives. And both of these can be more easily achieved with crypto. Like other blockchain-based protocols, shared ownership of the synchronization layer can incentivize all parties to continuously invest resources in improvements. Developers, users (and their AI agents), and other stakeholders can be incentivized for introducing, using, or iterating on new features and integration solutions.
Conversely, shared ownership also gives all users a stake in the overall success of the protocol, creating a mechanism to deter misbehavior. Just as Microsoft would not easily break the .docx file format standard because it would cause widespread negative impact on its users and brand; co-owners of the synchronization layer would also be harmed and thus unwilling to introduce clumsy or malicious code into the protocol.
As with all previous software standardization architectures, there is potential for powerful network effects here. As AI-generated software experiences a "Cambrian explosion," the number of diverse, heterogeneous systems that need to communicate with each other will grow exponentially. In short: vibe coding, if it wants to stay in sync, cannot rely on vibes alone; crypto is the answer.
7. Micropayment Systems Supporting Revenue Sharing
Liz Harkavy: AI agents and tools like ChatGPT, Claude, and Copilot provide people with an easier way to access information in the digital world. But for better or worse, they are also shaking up the economic structure of the open internet. This trend is already apparent—for example, as students increasingly use AI tools, educational platforms are experiencing significant traffic declines; meanwhile, several US media companies are suing OpenAI for copyright infringement. If the incentive system cannot be readjusted, we may see the internet become further enclosed, with more paywalls, while content creators continue to decrease.
Policy approaches certainly always exist, but while judicial processes are underway, some technical solutions are also emerging. One of the most promising (and technically challenging) solutions is to embed a "revenue sharing mechanism" into the underlying architecture of the internet. When an AI-driven operation ultimately leads to a sale, the content creator who provided the information source for that decision should receive a share of the revenue. The affiliate marketing ecosystem already does similar attribution tracking and revenue sharing; more advanced systems could automatically track all contributors along the entire information chain and reward them. Blockchain can clearly play a key role in tracking the "chain of information sources."
However, achieving such a system requires new infrastructure—especially: micropayment systems capable of handling very small amounts between multiple sources; attribution protocols capable of fairly assessing the value of different contributions; and governance models that ensure transparency and fairness.
Many existing blockchain tools show potential, such as various rollups, L2 networks, AI-native financial institution Catena Labs, and financial infrastructure protocol 0xSplits, all of which can achieve near-zero-cost transactions and more granular payment splitting.
Blockchain can enable advanced payment systems led by agents through various mechanisms:
Nanopayments: Can be split among multiple data providers, enabling a single user interaction to automatically trigger micro-payments to all contributing sources, executed by smart contracts.
Smart Contracts: Can automatically trigger enforceable "post-payment" after a transaction is completed, providing transparent, traceable compensation to content sources that influenced the purchasing decision.
Programmable Payment Splitting: Enables revenue distribution to be enforced by code rather than decided by a centralized institution, thus establishing trustless financial relationships between automated agents.
As these emerging technologies mature, they will build a new media economic model that captures the entire value creation chain, from creators, to platforms, to users.
8. Using Blockchain as a Registry for Intellectual Property and Provenance
Scott Duke Kominers: The emergence of generative AI makes it urgent to establish efficient, programmable mechanisms for intellectual property (IP) registration and tracking—both to ensure accurate provenance and to support new business models around access, sharing, and derivative creation of IP. Existing IP frameworks rely on costly intermediaries and ex-post enforcement mechanisms, which are clearly inadequate in an era where AI can instantly consume content and generate variants with one click.
What we need is an open, public registry that provides creators with clear proof of ownership, with low barriers to entry and high efficiency—while also allowing and other web applications to interact with it directly. Blockchain is well-suited for this role: it allows creators to register IP without relying on intermediaries and provides tamper-proof provenance; at the same time, it also enables third-party applications to easily identify, license, and interact with these IP assets.
Of course, people remain cautious about the overall concept of "whether technology can truly protect intellectual property." After all, the first two eras of the internet—and even the current AI revolution—have often been associated with a decline in IP protection. One reason is that many existing IP business models emphasize "excluding derivative works" rather than incentivizing and monetizing derivative creation. Programmable IP infrastructure can not only allow creators, franchisees, and brands to clearly establish their IP ownership in digital space, but also spawn new business models centered on "sharing IP for generative AI and digital applications." In a sense, it turns one of the threats generative AI poses to creative work into a new opportunity.
In the early days of NFTs, we already saw creators experimenting with new models, such as building brand network effects through CC0 on Ethereum to achieve value沉淀 (precipitation/sedimentation of value). Recently, we have seen infrastructure providers begin to build standardized, composable IP registration and licensing protocols, and even launch specialized blockchains (like Story Protocol). Some artists have begun using protocols like Alias, Neura, Titles to license their styles and works to support creative remixing. Meanwhile, Incention's sci-fi series Emergence allows fans to co-create universes and character settings, with each creative contribution recorded on Story's on-chain registry system.
9. Web Crawlers That Compensate Content Creators
Carra Wu: The AI agents with the strongest product-market fit today are not those for programming or entertainment, but web crawlers—they can autonomously browse the internet, collect data, and make judgments about which links to follow.
According to some estimates, nearly half of today's internet traffic already comes from non-human sources. Bots often ignore robots.txt files—a standard that should tell automated crawlers whether a website allows their access, but has little binding force in reality—and use the scraped data to strengthen the core moats of the world's largest tech companies. Worse, websites ultimately bear the cost for these "uninvited guests," expending bandwidth and CPU resources to deal with endless anonymous crawlers. In response, companies like Cloudflare and other CDNs (Content Delivery Networks) offer blocking services. All this constitutes a "kludged" system that shouldn't exist.
We have pointed out before that the original contract of the internet—the economic synergy between content creators creating content and platforms distributing content—is gradually disintegrating. This trend is already reflected in the data: over the past twelve months, website operators have begun blocking AI-oriented crawlers on a large scale. In July 2024, only about 9% of the global top 10,000 websites blocked AI crawlers, but now that proportion has reached 37%. As more website operators' technology matures and user dissatisfaction increases, this proportion will continue to rise.
So, what if instead of paying CDNs to block suspected bots with a "one-size-fits-all" approach, we try a middle path? That is, instead of AI crawlers "free-riding," they pay for the right to access data. Here, blockchain can play a role: in this vision, each web crawler agent holds a certain amount of crypto assets and negotiates on-chain with each website's "gatekeeper agent" or paywall protocol via something like an x402 protocol. (Of course, the challenge is that robots.txt, the "Robots Exclusion Standard," has been deeply ingrained in the operating models of internet companies since the 1990s. Changing this requires large-scale collaboration, or support from CDNs like Cloudflare.)
Meanwhile, human users can prove they are real people through World ID (see above) and gain free access. In this way, content creators and website operators can be compensated at the moment their data is collected by AI, while human users can still enjoy an internet with free flow of information.
10. Privacy-Preserving Advertising That is Both Precise and Not "Creepy"
Matt Gleason: AI is already influencing how we shop online, but what if the ads we see every day could actually be "useful"? People dislike ads for many reasons: irrelevant ads are pure noise; at the same time, not all "personalization" is good. Highly targeted advertising driven by vast amounts of consumer data can feel invasive; other applications try to monetize through "forced viewing of ads" (like unskippable ads on streaming platforms or in game levels).
Crypto can help improve these issues, offering an opportunity to reimagine the advertising system. When AI agents are combined with blockchain, they can tailor ads based on user-actively-set preferences, making ads neither irrelevant nor overly "creepy." More importantly, in this process, user data is not exposed globally, and users willing to share data or interact with ads can be compensated.
Achieving this model requires several technical foundations:
Low-fee digital payment systems: To compensate users for ad interactions (viewing, clicking, converting), businesses need to send a large number of small payments. To scale, this requires a system that is high-speed, high-throughput, and has almost zero fees.
Privacy-preserving data verification: AI agents need to verify whether consumers meet certain demographic characteristics. Zero-Knowledge Proofs (ZKPs) can perform such verification without revealing specific private information.
New incentive models: If the internet adopts a monetization method based on micropayments (e.g., < $0.05 per interaction), users can actively choose to watch ads to get compensated, thus turning the current "data extraction model" into a "user participation model."
For decades, people have been trying to make ads more "relevant"—online and offline alike. And re-examining advertising from the perspective of crypto and AI can truly make ads useful, controllable, and optional. For builders and advertisers, this means a more sustainable and consistent incentive structure; for users, it provides richer ways to discover information and explore the digital world.
Ultimately, this will not only make ad inventory more valuable but could also shake up the deeply entrenched, "extractive" advertising economy, replacing it with a more human-centric system: where users are not the "product being sold," but true participants.
11. User-“Owned and Controlled” AI Companions
Guy Wuollet: Today, many people spend more time on their devices than in offline communication, and this online time is increasingly spent interacting with AI models or AI-curated content. These models already provide a form of "companionship"—whether for entertainment, information, niche interests, or as educational tools for children. It is easy to imagine that in the near future, AI companions for education, healthcare, legal advice, and even daily emotional companionship will become a primary mode of interaction for humans.
Future AI companions will have infinite patience and be deeply customized for the individual and their usage scenarios. They are not just assistants or "robot servants," but may become relationship objects that users highly value. Therefore, the question arises: who ultimately owns and controls these relationships—the user, or companies and other intermediaries? If you have been concerned about content curation and censorship issues on social media over the past decade, this problem will become exponentially more complex and personal in the future.
The argument that "censorship-resistant hosting platforms like blockchain may be the best path to building uncensorable, user-controlled AI" has been充分论述 (fully discussed). Although users can run local models themselves and buy GPUs, for most people, this is either too expensive or too technically demanding.
Although the widespread adoption of AI companions is still some way off, the relevant technology is maturing rapidly: text-based chat AIs are already extremely natural and realistic; visual avatars are continuously improving; blockchain performance is constantly improving. To make "uncensorable AI companions" truly easy to use, we need to rely on better crypto application user experience (UX). Fortunately, wallets like Phantom have made blockchain interaction simpler, and embedded wallets, Passkeys, and account abstraction technologies allow users to easily achieve self-custody without managing seed phrases. At the same time, high-throughput, trustless computing systems based on optimistic and ZK coprocessors will also enable us to establish meaningful and sustainable long-term relationships with digital companions.
In the near future, the focus of public discussion will shift from "when will realistic digital companions and avatars appear" to "who will control them, and how will they be controlled."
