a16z: 11 Application Scenarios of the Fusion of Crypto and AI

Authors: Scott Duke Kominers, Sam Broner, Jay Drain, Guy Wuollet, Elizabeth Harkavy, Carra Wu, Matt Gleason

Compiled by: Saoirse, Foresight News

The economic logic of the internet has quietly changed. As the open web gradually shrinks into a "command input bar," we must consider: Will artificial intelligence lead us toward an open internet, or will we be trapped in a maze constructed by new types of paywalls? Will control rest in the hands of large centralized companies or a broad user base?

This is where Crypto comes into play. We have explored the intersection of AI and Crypto multiple times—simply put, blockchain is a new paradigm for reconstructing the architecture of internet services, capable of building a decentralized, trust-neutral, and user-owned network system. By redefining the economic rules that underpin existing systems, blockchain provides an effective means to counteract the centralization trends in the AI field, aiming for a more open and resilient internet ecosystem.

The concept of mutual empowerment between Crypto and AI systems has long existed, but the ways in which they combine have lacked clear definition. Some overlapping areas (such as "human identity" verification in the context of the proliferation of low-cost AI tools) have attracted the attention of developers and users, while other application scenarios may take years or even decades to materialize. Therefore, this article will share 11 cross-border application scenarios of AI and Crypto, aiming to promote relevant discussions: exploring the potential possibilities and challenges of combining AI and Crypto, and looking forward to more innovative directions. These scenarios are based on current technological levels and cover diverse fields, from massive micropayment processing to ensuring human dominance in future AI interactions.

Identity and Data Management

1. Persistent Data and Context in AI Interactions

Written by: Scott Duke Kominers

The development of generative AI heavily relies on data, but in many application scenarios, the importance of context (i.e., the state and background information relevant to the interaction) is just as crucial, if not more so, than the data itself.

Ideally, whether it’s an agent, a large language model interface, or other applications, AI systems should remember various details about the user, such as their work type, communication style, and preferred programming languages. However, in reality, users often need to reset this context in different sessions of the same application (for example, starting a new ChatGPT or Claude session), let alone when switching between different systems. Currently, the context of a generative AI application is almost impossible to transfer to other applications.

With the help of blockchain technology, AI systems can convert key contextual elements into persistent digital assets, allowing them to load upon session initiation and achieve seamless transfer across different AI platforms. Moreover, due to its characteristics, blockchain may be the only solution that simultaneously meets the requirements of "forward compatibility" and "interoperability."

This application is particularly suitable in AI-mediated gaming and media fields—user preferences (from difficulty settings to key bindings) can remain consistent across different games and scenarios. However, the real value lies in knowledge application scenarios (where AI needs to understand the user's knowledge base and learning patterns) and specialized AI applications (such as programming assistance). Of course, some companies have developed custom bots tailored to specific business contexts, but in such scenarios, context often cannot be transferred across systems, and even within different AI tools in the same company, sharing is difficult.

Organizations are just beginning to recognize this issue, and the current general solution is custom bots with fixed backgrounds. However, context migration among users within platforms has already begun to emerge off-chain: for example, on the Poe platform, users can rent out custom bots for others to use.

Once such scenarios are brought on-chain, our interactive AI systems will be able to share a context layer that includes all key elements of digital activities. They will immediately understand user preferences and optimize the user experience more accurately. Conversely, a blockchain-based intellectual property registration system that allows AI to reference persistent on-chain context also creates possibilities for new market interactions around prompts and information modules—users can directly authorize or commercialize their expertise while retaining control over their data. Of course, shared context will also give rise to many possibilities we have yet to foresee.

2. Universal Identity System for Agents

Written by: Sam Broner

Identity (i.e., "an authoritative record of the essential attributes of something") is the underlying architecture supporting today’s digital discovery, aggregation, and payment systems. As platforms enclose this architecture within ecological walls, identity has become part of the product functionality in the eyes of users: Amazon assigns unique identifiers (ASIN or FNSKU) to products, centrally displays them, and assists users in discovery and payment; Facebook operates similarly, where user identity is the core foundation of its information flow and overall application discovery features, including product listings, native posts, and paid advertisements.

As AI agents evolve, this landscape is about to change. As more companies adopt agents (for customer service, logistics management, payment processing, etc.), their platforms will no longer be limited to single interface applications but will span multiple platform ecosystems, accumulating deep context and executing more diverse tasks for users. However, if agent identity is only tied to a single market, it will lose usability in other critical scenarios (such as email threads, Slack channels, and other products).

Therefore, agents need a single, portable "digital passport." Without this passport, it will be impossible to determine how to pay agents, verify their version information, inquire about their functional attributes, know their service targets, or trace their reputation across different applications and platforms. Agent identity needs to serve multiple functions, such as a wallet, API registry, update log, and social proof, to ensure that any interface (email, Slack, or other agents) can parse and interact according to a unified standard. Lacking this shared information of "identity," every system integration would need to build the underlying architecture from scratch, and the discovery mechanism would always remain temporary, causing users to lose contextual information every time they switch channels or platforms.

We have the opportunity to design agent infrastructure from foundational principles. So, how do we build a trusted neutral identity layer that is more comprehensive than DNS records? Agents should not repeat the mistakes of "binding identity with discovery, aggregation, and payment functions" in monolithic platforms, but should be able to accept payments and showcase functionalities across multiple ecosystems without worrying about being locked into a specific platform. This is precisely where the value of the cross-border integration of Crypto and AI lies—the permissionless composability provided by blockchain networks can help developers build more practical agents and better user experiences.

Overall, vertically integrated solutions (like Facebook or Amazon) currently offer a superior user experience. One of the inherent challenges of creating excellent products is ensuring that all components work together seamlessly from top to bottom. However, this convenience comes at a high cost, especially as the costs of building agent aggregation, marketing, commercialization, and distribution software continue to decline, and the coverage of agent applications expands. While achieving the user experience level of vertically integrated platforms still requires effort, building a trusted neutral identity layer for agents will empower entrepreneurs to take control of their "digital passports" and drive innovation in distribution and design.

3. Forward-Compatible "Human Identity" Proof Mechanism

Written by: Jay Drain Jr., Scott Duke Kominers

As AI technology permeates various online interaction scenarios (including deepfakes and social media manipulation), determining "whether one is interacting with a real human online" has become increasingly difficult. The collapse of this trust system has already occurred—from comment bots on the X platform (formerly Twitter) to bot accounts in dating apps, the boundaries between the virtual and the real are gradually blurring. In this context, "human identity" proof has become a core infrastructure of the digital ecosystem.

One way to prove "I am human" is through digital IDs (including centralized IDs used by the TSA). Digital IDs encompass all elements used for identity verification, such as usernames, PINs, passwords, and third-party proofs (like citizenship or credit ratings). Decentralization shows significant value here: when data is stored in centralized systems, issuers may revoke access, charge additional fees, or implement surveillance; whereas a decentralized model reverses this logic—users (rather than platform administrators) control their identity, making it more secure and resistant to censorship.

Unlike traditional identity systems, decentralized "human identity" proof mechanisms (like World’s Proof of Human) allow users to manage their identity information autonomously and verify "human attributes" in a privacy-preserving and trust-neutral manner. Just as a driver's license can be used in any region (regardless of when or where it was issued), decentralized "human identity" proof can serve as a universal underlying protocol across platforms, even applicable to emerging platforms that have yet to be born. In other words, blockchain-based "human identity" proof possesses forward-compatible characteristics, as it has the following advantages:

• Portability: Relevant protocols belong to open standards, and any platform can integrate them. Decentralized "human identity" proof can be managed through public infrastructure, controlled by users, and possesses complete portability, allowing any platform now or in the future to achieve compatibility;
• Permissionless accessibility: Platforms can autonomously choose to recognize "human identity" IDs without going through APIs that may discriminate against different use cases.

The challenge in this field is the practical application: although there has yet to be a large-scale "human identity" proof application scenario, we anticipate that as user numbers reach a critical mass, early partnerships form, and killer applications emerge, adoption rates will accelerate. Each application adopting a specific digital ID standard will enhance the value of that ID to users, thereby attracting more users to acquire IDs, creating a positive feedback loop (and since on-chain IDs are designed to be interoperable, network effects can accumulate quickly).

We have seen mainstream consumer applications in gaming, dating, and social media announce partnerships with World ID to help users confirm they are interacting with real humans (rather than programs) while gaming, chatting, and trading; this year has also seen the emergence of new identity protocols like Solana Attestation Service (SAS)—although SAS is not an issuer of "human identity" proof, it allows users to privately associate off-chain data (such as compliant KYC or investment qualifications) with their Solana wallets, helping to build a decentralized identity system. All these signs indicate that the tipping point for decentralized "human identity" proof may not be far off.

The significance of "human identity" proof lies not only in banning bots but also in delineating clear boundaries between AI agents and human networks. It enables users and applications to distinguish between human and machine interactions, thereby creating a higher quality, safer, and more authentic digital experience.

Decentralized Infrastructure

4. Decentralized Infrastructure in AI (DePIN)

Written by: Guy Wuollet

While AI belongs to digital services, its development is increasingly constrained by physical infrastructure. Decentralized infrastructure networks (DePIN) offer a new model for building and operating real-world systems, helping to democratize the computational infrastructure behind AI innovation, making it more economical, resilient, and resistant to censorship.

How can this be achieved? The two core challenges facing AI development are computing power supply and chip acquisition. Decentralized computing networks can provide more computing power, while developers are leveraging DePIN to aggregate idle chip resources from gaming PCs, data centers, and more. These computing devices can form a permissionless computing market, creating a fair competitive environment for developing new AI products.

Other application scenarios include distributed training and fine-tuning of large language models, as well as distributed networks for model inference. Decentralized training and inference (utilizing idle computing resources) can significantly reduce costs while providing resistance to censorship, ensuring that developers are not cut off from service by oversized cloud service providers (like centralized cloud giants).

The long-standing issue of a few companies monopolizing AI models can be addressed by decentralized networks, helping to build a more economical, censorship-resistant, and scalable AI ecosystem.

5. Infrastructure and Rule Framework for Interactions Between AI Agents, Endpoint Service Providers, and Users

Written by: Scott Duke Kominers

As AI tools continue to enhance their capabilities in solving complex tasks and executing multi-layered interaction chains, AI systems will increasingly need to interact with other AI systems without human intervention.

For example, a particular AI agent may need to request specific data related to a computational task or recruit specialized AI agents to complete specific tasks (such as assigning statistical bots to develop and run model simulations or invoking image generation bots when creating marketing materials). AI agents will also create significant value in completing the entire transaction process or other activities on behalf of users, such as searching for and booking flights based on user preferences or discovering and ordering new books of their favorite genres.

Currently, there is no mature general market for interactions between agents, and such cross-system queries can mostly only be achieved through explicit API connections or by implementing agent calls as internal functions within AI ecosystems.

Overall, most AI agents today operate in siloed ecosystems, with API interfaces relatively closed and lacking architectural standardization. However, blockchain technology can help protocols establish open standards, which is crucial for short-term application implementation; in the long run, this also supports forward compatibility— as new AI agents evolve and emerge, they can connect to the same underlying network. Due to its interoperable, open-source, decentralized, and generally more easily upgradeable architectural characteristics, blockchain can flexibly adapt to the innovative demands of the AI field.

As the market develops, several companies have begun to build blockchain infrastructure for interactions between agents: for example, Halliday recently launched a related protocol to provide a standardized cross-chain architecture for AI workflows and interactions, offering protective mechanisms at the protocol layer to ensure AI behavior does not exceed user intent; Catena, Skyfire, and Nevermind are leveraging blockchain technology to support automatic payments between AI agents without human intervention. More such systems are under development, and Coinbase has even begun to provide infrastructure support for these explorations.

6. Ensuring Synchronization of AI/Custom Programming Applications

Written by: Sam Broner, Scott Duke Kominers

The innovation of generative AI has led to a qualitative leap in software development efficiency: coding speed has increased by several orders of magnitude, and most importantly, it can be accomplished through natural language— even inexperienced programmers can replicate existing programs or build new applications from scratch.

However, AI-assisted coding introduces a significant amount of uncertainty both within and outside the program while creating new opportunities. "Custom programming" abstracts the complex network of dependencies underlying software, but this also makes programs susceptible to functional and security vulnerabilities when changes occur in source libraries and other inputs. Moreover, as people use AI to create personalized applications and workflows, interactions with others' systems become increasingly challenging— in fact, even two "custom programming" programs with the same functionality may exhibit significant differences in operational logic and output structure.

Historically, ensuring software consistency and compatibility has been the responsibility of file formats and operating systems, and in recent years, it has relied on shared software and API integration. However, in this new era of real-time evolution, iteration, and branching of software, the standardization layer needs to be widely accessible and continuously upgradable while maintaining user trust. Additionally, relying solely on AI cannot solve the problem of "incentivizing people to build and maintain these connections."

Blockchain technology addresses both of these issues: a protocolized synchronization layer can be embedded within users' custom software architectures and dynamically updated to ensure cross-system compatibility as environments change. Historically, large companies might pay system integrators like Deloitte millions of dollars to customize Salesforce instances. Today, engineers can create custom interfaces to view sales information over a weekend, but as the number of custom software increases, developers will need professional support to keep these applications synchronized. (Note: Salesforce is a customer relationship management (CRM) software service provider founded in March 1999 in the United States.)

This is similar to the development model of today’s open-source software libraries but with continuous updates (rather than periodic releases) and incentive mechanisms— both of which are made easier by Crypto technology. Like other blockchain-based protocols, the shared ownership mechanism of the synchronization layer incentivizes all parties to actively invest in improvements: developers, users (and their AI agents), and other consumers can be rewarded for introducing, using, and optimizing new features and integrations.

Conversely, shared ownership tightly binds all users to the overall success of the protocol, forming a buffer against malicious behavior— just as Microsoft would not easily undermine the .docx file standard (as it would have a cascading impact on users and the brand), the co-owners of the synchronization layer would also be disinclined to introduce inefficient or malicious code into the protocol.

As with all standardized software architectures we have seen, there is tremendous potential for network effects here. As the "Cambrian explosion" of AI coding software continues to evolve, the network of heterogeneous systems that need to maintain communication will expand exponentially. In short, "custom programming" requires not only a "coding style" but also Crypto technology to maintain system synchronization.

New Economy and Incentive Models

7. Micropayment Systems Supporting Revenue Sharing

Written by: Liz Harkavy

AI agents and tools like ChatGPT, Claude, and Copilot provide new ways to navigate the digital world conveniently, but whether good or bad, they are shaking the economic foundations of the open internet. We have seen concrete manifestations of this trend— for example, educational platforms have seen a sharp decline in traffic due to students' extensive use of AI tools, and several newspapers in the U.S. are suing OpenAI for alleged copyright infringement. Without a realignment of incentive mechanisms, we may face an increasingly closed internet: more paywalls and fewer content creators.

Of course, policy solutions always exist, but while they progress through judicial processes, a series of technological solutions are also emerging. The most promising (and technically challenging) solution may be to embed revenue-sharing mechanisms within the network architecture: when AI-driven actions facilitate transactions, the content sources that provide informational support for those decisions should receive corresponding shares. The affiliate marketing ecosystem has already been conducting similar attribution tracking and revenue sharing, while a more advanced version could automatically track and reward all contributors in the information chain— blockchain technology can clearly play a key role in tracking this provenance chain.

However, such systems also require new infrastructure with other functionalities— particularly micropayment systems capable of handling cross-source microtransactions, attribution protocols that fairly assess the value of different contributions, and governance models that ensure transparency and fairness. Many existing blockchain tools (such as Rollups and Layer2, AI-native financial institutions like Catena Labs, and financial infrastructure protocols like 0xSplits) have already demonstrated application potential, supporting near-zero-cost transactions and more granular payment splits.

Blockchain can implement complex agent payment systems through various mechanisms:

• Nanopayments can be split among multiple data providers, allowing a single user interaction to trigger micropayments to all contributing sources via automated smart contracts.
• Smart contracts can support executable retrospective payments triggered after a transaction is completed, transparently and traceably compensating sources that provided information for purchasing decisions.
• Additionally, blockchain supports complex and programmable payment split allocations, ensuring that revenue is fairly distributed according to code-enforced rules rather than relying on centralized decision-making, creating trustless financial relationships among autonomous agents.

As these emerging technologies mature, they will create new economic models for the media industry, capturing the entire value creation chain from creators to platforms to users.

8. Blockchain as an Intellectual Property and Provenance Registration System

Written by: Scott Duke Kominers

The development of generative AI has generated an urgent need for efficient, programmable intellectual property registration and tracking mechanisms— both to clarify rights ownership and to support business models around access, sharing, and re-creation of intellectual property. Existing intellectual property protection frameworks rely on expensive intermediaries and post-facto enforcement measures, which are ill-suited to the demands of an era where AI consumes content instantly and generates new variants at the click of a button.

What we need is an open public registration system that provides clear proof of ownership, allowing intellectual property creators to participate in interactions efficiently, and enabling AI and other web applications to connect directly. Blockchain technology is an ideal choice: it allows for intellectual property registration without intermediaries, provides immutable provenance proof, and enables third-party applications to easily identify, authorize, and use that intellectual property.

Some are skeptical of the view that "technology can protect intellectual property"— after all, the first two eras of internet development (and the ongoing evolution of AI) have often been associated with weakened intellectual property protection. Part of the reason is that many current intellectual property-based business models focus on "prohibiting derivative works" rather than incentivizing and commercializing derivative creations. However, programmable intellectual property infrastructure not only allows creators, brands, and IP owners to clearly establish ownership in the digital space but also opens the door to "business models around IP sharing (for generative AI and other digital applications)"— effectively transforming the primary threat that generative AI poses to creation into an opportunity.

We have seen creators early on experimenting with new models in the NFT space: companies leveraging NFT assets on Ethereum to support network effects and value accumulation under CC0 branding; recently, infrastructure providers are also building protocols for standardized and composable IP registration and licensing (like Story Protocol) or even dedicated blockchains. Some artists have begun using these tools to authorize their styles and works for creative re-creation through protocols like Alias, Neura, and Titles. Invention's Emergence series invites fans to co-create a sci-fi universe and its characters, with a blockchain registry built on Story Protocol tracking the creators of each element.

9. Compensation Mechanism for Content Creators via Web Crawlers

Written by: Carra Wu

Today, the most market-relevant AI agents are not programming or entertainment tools but web crawlers— they autonomously browse the web, collect data, and decide where to scrape from.

It is estimated that nearly half of current web traffic comes from non-human entities. Crawlers often ignore the robots.txt protocol (which is supposed to inform automated crawlers whether they are allowed to access a site, but in practice has weak enforceability) and use the scraped data to reinforce the market barriers of tech giants. Worse still, websites have to bear the costs of providing bandwidth and CPU resources for these uninvited guests. In response, CDNs (Content Delivery Networks) like Cloudflare offer blocking services, but this is merely a patchwork solution that should not exist.

We have pointed out that the native protocols of the internet (the economic agreements between content creators and distribution platforms) may be collapsing, and data is confirming this trend. Over the past 12 months, website owners have massively blocked AI crawlers: in July 2024, only 9% of the top 10,000 websites globally prohibited AI crawlers, but this figure has now risen to 37%, and as more website operators upgrade their technology and user dissatisfaction grows, this number will continue to climb.

Is there a compromise solution that does not rely on CDNs to completely block suspected crawler access? AI crawlers should not exploit systems designed for human traffic for free but should pay for data scraping rights. This is where blockchain comes into play: each web crawler agent could hold Crypto and negotiate on-chain with each website's "access agent" or paywall protocol via the x402 protocol (of course, the challenge lies in the fact that the robots.txt protocol has been deeply embedded in internet business logic since the 1990s, requiring large-scale collaboration or participation from CDNs like Cloudflare to break through).

Meanwhile, humans can access content for free by proving their identity through World ID (see Chapter 3). In this model, content creators and website owners can receive compensation during the AI dataset collection process, while human users can still enjoy an "information-free" internet.

The Future Ownership of AI

10. Privacy-Preserving Personalized Advertising

Written by: Matt Gleason

AI has begun to influence online shopping experiences, but what if the ads we see daily could be "truly useful"? The reasons people dislike ads are obvious: ineffective ads are mere noise, while overly precise AI ads based on massive consumption data feel invasive to privacy. Other applications profit by imposing "non-skippable ads" on content (such as streaming services or game levels).

Crypto offers the potential to reconstruct advertising models. Personalized AI agents combined with blockchain can find a balance between "irrelevant ads" and "overly precise ads," delivering ads based on user-defined preferences. More importantly, this model does not require exposing users' global data and can directly compensate users who actively share data or interact with ads.

Achieving this goal requires meeting the following technical requirements:

• Low-fee digital payments: To compensate users for their ad interactions (views, clicks, conversions), businesses need to frequently send small payments, which requires systems to have high processing capabilities and near-zero fees;
• Privacy-preserving data verification: AI agents need to prove that users meet specific demographic attributes, and zero-knowledge proofs can accomplish attribute verification while protecting privacy;
• Incentive model: If the internet adopts a profit model based on micropayments (such as interactions costing less than $0.05, as discussed in Chapter 7), users can choose to "watch ads for small rewards," transforming the existing "exploitative" model into a "participatory" model.

For decades, online (and even offline for hundreds of years) advertising has pursued "relevance." Reconstructing advertising from the perspective of Crypto and AI will ultimately make it more practically valuable— customized yet non-intrusive, benefiting all parties: unlocking more sustainable and incentive-compatible business models for developers and advertisers, and providing users with more pathways to explore the digital world.

This will not only enhance the value of ad placements but may also disrupt the deeply entrenched "exploitative" advertising economy, building a more human-centered system— users are no longer commodities being traded but active participants.

11. Human-Owned and Controlled AI Companions

Written by: Guy Wuollet

Today, people spend more time on devices than in offline interactions, increasingly engaging with AI models and AI-generated content. These models have begun to provide companionship value, whether for entertainment, information retrieval, satisfying niche interests, or educating children. It is not hard to imagine that in the near future, AI companions used for education, healthcare, legal consultation, and emotional support will become mainstream interaction methods.

Future AI companions will possess infinite patience and be tailored to individual needs— they will no longer just be tools or robotic servants but may become highly valued relationships. Therefore, the question of "who owns and controls these relationships" is crucial (is it the user or intermediaries like companies?). If you have been concerned about content filtering and censorship on social media over the past decade, this issue will become more complex and personal in the future.

The viewpoint that "censorship-resistant blockchain hosting platforms are the most viable path to user-controlled AI" has been discussed multiple times (as mentioned earlier). In theory, individuals could run device-side models or purchase GPUs themselves, but most people either cannot afford it or lack the technical skills.

While the widespread adoption of AI companions will take time, the relevant technologies are rapidly iterating: human-like text interaction companions are quite mature, visual avatar technology has significantly advanced, and blockchain performance continues to improve. To ensure that censorship-resistant companions are user-friendly, we need to rely on better user experiences to realize crypto-driven applications. Fortunately, wallets like Phantom have greatly simplified blockchain interactions, and embedded wallets, password keys, and account abstraction technologies allow users to hold self-custodial wallets without memorizing seed phrases. High-throughput trusted computing technologies based on Optimistic and ZK co-processors will also help build deep, lasting relationships with digital companions.

In the near future, the focus of discussion will shift from "when will realistic digital companions appear" to "who will control them."