a16z on the Second Half of AI+Crypto: Identity Verification, Infrastructure, and New Economic Models

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 undergoing a transformation. As open networks are increasingly collapsing into a "prompt bar," we are forced to consider: will AI lead to an open internet, or will it degrade into a maze of new paywalls? And who will ultimately control this—large, centralized corporations, or widespread user communities?

This is precisely where crypto can intervene. We’ve frequently discussed the intersection of AI and crypto; in simple terms, blockchain represents a new way to build internet services and networks that are decentralized, structurally neutral, and owned by users. They provide a counterbalance to the increasingly centralized tendencies of current AI systems by renegotiating the underlying economic relationships, thereby contributing to a more open and resilient internet.

The idea that "crypto can help build better AI systems, and vice versa" isn’t new—but it often lacks clear definition. Some intersection areas, such as verifying "proof of humanity" in the context of widespread deployment of low-cost AI systems, have already begun to attract the attention of developers and users. Other use cases may take years or even decades to materialize. Therefore, in this article, we’ve compiled 11 practical use cases at the intersection of AI and crypto, aiming to facilitate deeper discussions around "what's possible" and "what challenges remain to be solved." These use cases are grounded in technologies currently under development, from handling vast micropayments to ensuring humans can truly own their relationships with future AI systems.

Identity

Persistent Data and Context in AI Interactions

Author: Scott Duke Kominers

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

Ideally, an AI system—whether an agent, LLM interface, or other application—should remember the type of projects you’re working on, your communication style, your preferred programming language, and numerous other details. However, in practice, users often need to reconstruct this context repeatedly within different sessions of the same application—like starting a new ChatGPT or Claude session—not to mention the added difficulty of switching between different systems.

Currently, context transferability between different generative AI applications is almost non-existent.

However, leveraging blockchain technology, AI systems can preserve critical context elements as persistent digital assets, which can be loaded at the start of each session and seamlessly transferred across multiple AI platforms. Additionally, blockchain is arguably the only solution that is both forward-compatible and inherently emphasizes interoperability—traits fundamental to blockchain protocols.

A natural application scenario lies in gaming and media involving AI, where user preferences (ranging from game difficulty to key bindings) can be consistently maintained across different games and environments. However, the even more valuable use cases surface in knowledge-related applications, where AI needs to understand the knowledge a user has already mastered and their learning style; or in more specialized AI scenarios, such as programming. Although some enterprises have already developed custom AI bots capable of maintaining context within a limited scope, these contextual elements often cannot be transferred across different AI systems within the organization.

Enterprises are only beginning to recognize this issue, and the closest solution so far lies in custom bots with fixed context. Meanwhile, within certain platforms, practices of sharing context between different users have started to emerge off-chain. For example, the Poe platform allows users to rent out their custom-designed bots to others.

By taking such practices on-chain, we could establish a "context layer" composed of all the key elements of our digital behavior, shared among the AI systems we interact with. AI systems would then be able to immediately understand our preferences, improving and optimizing the interaction experience for us. Conversely, just as blockchain facilitates intellectual property registration, allowing AI to reference persistent on-chain contexts could stimulate new types of market interactions centered around prompts and informational modules. For instance, users could directly license or monetize their expertise while retaining ownership of their data. Naturally, this kind of context sharing would 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 underlying infrastructure supporting today's systems of digital discovery, aggregation, and payment. Since platforms operate this foundational layer behind the scenes, we typically only experience its effects in the final product. For example, Amazon assigns unique identifiers (ASIN or FNSKU) to products, aggregates them for centralized display, and facilitates both discovery and payment for users; Facebook does something similar—user identities underpin its content recommendations, Marketplace product listings, organic content, and ad discovery.

But as AI agents evolve, this status quo is poised for change. Enterprises are deploying AI agents across customer service, logistics, payments, and numerous other areas, while platform architectures are transitioning from single interfaces to cross-platform, cross-device distributed systems. These AI agents will accumulate deep contextual knowledge and perform increasingly complex tasks on behalf of users. If an AI agent's identity is tethered to a single platform or marketplace, it will struggle to function effectively in other critical scenarios—be it within email conversations, Slack channels, or other tools and environments.

Therefore, AI agents need a unified, transferable "passport." Without it, we won’t be able to identify their payment methods, verify their versions, query their capabilities, determine who they represent, or track their reputation across platforms. An agent's identity should simultaneously function as a wallet, API registry, update log, and social proof—this way, any interface (whether email, Slack, or another agent) can consistently identify and interact with it.

Without a unified "identity" primitive, every integration starts from scratch, discovery mechanisms remain serendipitous, and users lose context when switching between platforms.

We are currently in a phase where we can redesign agent infrastructure "from first principles." So, how do we construct a richer, trust-minimized, and neutral identity layer that’s better than a DNS record? We shouldn’t rebuild "monolithic platforms" that bundle identity, discovery, aggregation, and payments together; instead, agents should freely receive payments, list capabilities, and coexist across multiple ecosystems without fear of being locked into a single platform.

This is where the synergy between AI and crypto shines: the "permissionless composability" provided by blockchain networks enables developers to build more useful agents and deliver superior user experiences.

Of course, today’s vertically integrated platforms (like Facebook or Amazon) still offer better user experiences—because one of the complexities of building great products is ensuring that all modules work seamlessly from top to bottom. But this convenience comes at a high cost. Especially as the costs of building, aggregating, monetizing, and distributing agents continue to decline, and the reach of agent applications expands, a trusted and neutral identity layer will grant entrepreneurs a true sovereign "passport" and encourage more exploration and innovation in distribution and design.

A Future-proof "Proof of Personhood"

By: Jay Drain Jr. and Scott Duke Kominers

As AI becomes more pervasive—whether powering bots and intelligent agents in online interactions or creating deepfakes and manipulating social media—it’s becoming increasingly difficult to distinguish between humans and programs in online interactions. This erosion of trust isn’t some far-off possibility; it’s already quietly happening. From comment farms on X (formerly Twitter) to bots on dating apps, the line between reality and virtuality is becoming blurred. In this environment, "Proof of Personhood" (PoP) is steadily emerging as a critical piece of infrastructure.

Currently, one way to verify that someone is human is through digital identity systems (such as the centralized identity systems used by the US TSA). Digital identities encompass the information a user can provide to verify their identity—usernames, PINs, passwords, third-party attestations (like citizenship or credit records), and other credentials. The value of decentralization becomes clear here: when this data is centrally managed, identity providers can revoke access, charge fees, or even enable surveillance. Decentralization flips this model, empowering users, not platforms, to control their identities, making them more secure and censorship-resistant.

Unlike traditional identity systems, decentralized "proof of personhood" (PoP) mechanisms (such as Worldcoin's World ID system) allow users to manage and store their identity data autonomously, enabling them to verify they are real humans in a privacy-friendly, trust-neutral manner. Similar to a driver's license, once a PoP is issued, it can be used universally across any platform, anytime and anywhere. This blockchain-based PoP thus offers forward compatibility, reflected in two key aspects:

 Portability: PoP adheres to open protocols, allowing integration on any platform. Identities are user-controlled and built on public infrastructure, making them fully portable. Any existing or future platform can connect seamlessly.

 Permissionless accessibility: Any platform can independently choose to recognize the PoP identity without requiring authorization through a centralized API. This eliminates the risk of certain use cases being denied.

The primary challenge in this domain currently lies in user adoption: while we have not yet seen large-scale practical use cases for proof of personhood, we believe that adoption will accelerate rapidly once a critical mass of users is reached, supported by key partnerships and certain "killer applications." Each application integrating a PoP standard enhances the utility of the identity, attracting more users to claim it. In turn, a growing user base incentivizes more applications to integrate the standard, creating a powerful network effect. (And because on-chain identities are inherently designed for interoperability, this effect will be even more explosive.)

We already see some mainstream consumer applications and services, particularly in gaming, social networking, and dating, announcing collaborations with World ID to help users ensure that they are interacting with real people — or even the specific individuals they expect. Meanwhile, new identity protocols continue to emerge, such as the Solana Attestation Service (SAS). Although SAS is not itself a PoP issuer, it enables users to privately link off-chain data (e.g., KYC verification or accreditation required for compliance) with their Solana wallets, laying a foundation for building decentralized identity systems.

All these indicators suggest that the tipping point for decentralized PoP may be imminent.

The significance of PoP extends beyond just "blocking bots"; it is a crucial mechanism for drawing a clear boundary between human networks and AI networks. It allows users and applications to explicitly distinguish "human-to-human interaction" from "human-to-machine interaction," enabling a safer, more authentic, and healthier experience in the digital world.

Decentralized Infrastructure for AI

Decentralized Physical Infrastructure for AI (DePIN)

Author: Guy Wuollet

While AI is fundamentally a digital service, its development is increasingly constrained by physical infrastructure. Decentralized Physical Infrastructure Networks (DePIN) represent a groundbreaking model for building and operating real-world systems, democratizing the compute infrastructure necessary for AI innovation. This approach makes resources cheaper, more resilient, and more censorship-resistant.

Why is this important? The two major bottlenecks in AI development are energy and chip availability. Decentralized energy can help unlock more electricity resources, and developers are leveraging DePIN to aggregate idle chip resources from sources like gaming computers and data centers. Together, these computing devices create a permissionless compute marketplace, fostering a more equitable environment for the development of AI products.

Other use cases include distributed training and fine-tuning of large language models (LLMs) and creating distributed networks for model inference. Decentralized training and inference not only reduce costs (by utilizing previously idle compute power) but also provide censorship resistance, ensuring that developers are not at risk of being deplatformed by relying on hyperscale centralized cloud service providers.

The high concentration of AI models in the hands of a few companies has long been a concern. Decentralized networks can play a critical role in building a more cost-effective, censorship-resistant, and scalable AI ecosystem.

Enabling Infrastructure and Guardrails for Interactions Among AI Agents, Service Providers, and Users

Author: Scott Duke Kominers

As AI tools grow increasingly proficient at handling complex tasks and multi-layered chains of interaction, they will need to autonomously interact with other AIs more frequently, rather than relying on human controllers.

For instance, an AI agent might need to access data essential for a specific computation or recruit an AI agent skilled in a particular task—such as arranging a statistics bot to perform model simulations or invoking an image generation bot for crafting marketing materials. These AI agents could also deliver immense value by executing complete transaction or activity workflows on behalf of users—for example, identifying and booking a flight based on user preferences, or discovering and purchasing a book that aligns with the user's tastes.

At present, a mature and general agent-to-agent marketplace does not exist—such interactions are mostly limited to explicit API interfaces or a small number of closed ecosystems that maintain internal agent calls.

A more pervasive issue is that most current AI agents exist in isolated systems with closed interfaces and a lack of standardized architecture. However, blockchain technology can help protocols establish open standards, which are essential for short-term adoption. In the long term, this also supports "forward compatibility": as new AI agents continue to evolve and emerge, they will still be able to connect to the same underlying network. Thanks to blockchain's interoperable, open-source, decentralized, and easily upgradable architecture, it can adapt more rapidly to the transformative changes brought by AI innovation.

Several companies are already building blockchain "rails" for agent-to-agent interaction. For instance, Halliday has introduced a protocol that provides a standardized cross-chain framework for AI workflows and interactions, with guardrails embedded at the protocol layer to ensure AI aligns with user intent. Meanwhile, Catena, Skyfire, and Nevermind support payment interactions between AI agents without requiring human intervention. Coinbase has also begun to provide infrastructure support for such projects.

Keeping AI / Vibe-Coding Applications in Sync

By Sam Broner and Scott Duke Kominers

The explosive growth of generative AI in recent years has made software development easier than ever before. Coding efficiency has increased by several orders of magnitude, and more importantly—now it’s possible to program using natural language. Even individuals unfamiliar with programming can fork existing programs or build entirely new applications from scratch.

However, while AI-assisted programming brings new opportunities, it also introduces significant amounts of "entropy" within and across programs. So-called "vibe coding" simplifies the intricate web of dependencies in underlying systems, but it can also make programs more prone to functional or security issues when foundational components are updated. Additionally, as more people use AI to create personalized applications and workflows, interactions between different user systems become increasingly challenging. In fact, even if two vibe-coded programs have the same functionality, their operating logic and output structures might differ significantly.

Historically, standardized mechanisms such as file formats and operating systems were used to ensure consistency and compatibility. Over time, this evolved into shared software libraries and API interfaces. However, in a world where software continuously evolves, morphs, and forks in real time, these standardization layers need to be widely accessible, continuously upgradable—and maintain user trust. Moreover, AI alone cannot solve the challenge of motivating people to maintain these connections and compatibility standards.

Blockchain offers a solution to address both issues simultaneously: embedding a "protocolized synchronization layer" within users’ custom software and ensuring cross-application compatibility through dynamic updates. In the past, large enterprises might have had to spend millions of dollars hiring system integrators (like Deloitte) to customize Salesforce systems. Today, an engineer might create a sales data visualization interface over a weekend. Yet, as personalized software proliferates, developers will also need tools to keep these applications synchronized and operational.

This is somewhat akin to the operational mechanism of current open-source software libraries, but its updates are real-time rather than periodic—and it comes with built-in incentive structures. All of these can be implemented through crypto. Like other blockchain-based protocols, shared ownership encourages participants to actively contribute to improving the protocol. Developers, users (or their AI agents), and other consumers can be rewarded for introducing, using, and enhancing new features and integrations.

Conversely, shared ownership also ensures that every user has a vested interest in the overall success of the protocol, creating an anti-abuse mechanism. Just as Microsoft wouldn't recklessly disrupt the .docx file format standard because it would harm users and damage its brand reputation, co-owners of a protocol are unlikely to introduce bad or malicious code.

As we've seen with various software standardization frameworks in the past, there is tremendous potential for network effects here as well. As the "Cambrian explosion" of AI programming software continues, the number of heterogeneous systems needing to communicate with each other will grow rapidly.

In short: vibe programming needs to stay aligned—it can't rely on vibe alone. Crypto is key.

New Economic and Incentive Models

Micro-Payment Mechanisms for Revenue Sharing

Author: Liz Harkavy

AI agents and tools like ChatGPT, Claude, and Copilot offer us entirely new and convenient ways to navigate the digital world. For better or worse, however, these technologies are disrupting the economic structure of the open internet. We're already seeing early signs of this trend—for example, some educational platforms are experiencing significant traffic declines as students increasingly turn to AI tools; meanwhile, several U.S. newspapers have sued OpenAI for copyright infringement. If we fail to realign incentive systems, the internet will become more closed off: more paywalls, fewer content creators.

Of course, policy could address these issues, but while the legal processes unfold, some technical solutions have already begun to emerge. Among the most promising (and technically challenging) is embedding revenue-sharing mechanisms directly into the fabric of the internet. When an AI-driven action facilitates a transaction, the content creators who provided the information for that action should receive a proportional share. This concept already exists in affiliate marketing systems, where sources can be tracked and revenue shared; more advanced versions could automate the tracking and reward every contributor in an information chain. Blockchain technology is clearly well-suited to play a critical role in this "source attribution" mechanism.

However, such systems require new infrastructure—especially micro-payment systems capable of handling 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 are already showing potential in this space, such as rollups, Layer-2 (L2) scaling solutions, AI-native financial institutions like Catena Labs, and financial infrastructure protocols like 0xSplits—which enable near-zero-cost transactions and finer-grained revenue sharing.

Blockchain can enable complex micropayment systems to become a reality through the following mechanisms:

Nanopayments can be automatically split among multiple data providers, allowing a single user interaction to trigger micropayments to all contributors of information;

Smart contracts can enable enforceable post-transaction payments, ensuring that the sources of information influencing user decisions are compensated after the transaction is completed, all while maintaining transparency and traceability throughout the process;

Blockchain can also facilitate complex, programmable revenue-sharing rules that are enforced through code, eliminating subjective judgments from centralized entities and enabling trustless financial relationships among autonomous agents.

As these emerging technologies continue to mature, they are expected to foster a new economic model for the media industry—one that encompasses the entire value chain, from content creators to platforms and finally to end users.

Blockchain as a Registry System for Intellectual Property and Source Attribution

Author: Scott Duke Kominers

The rise of generative AI has created an urgent need for an efficient, programmable mechanism to register and track intellectual property (IP)—one that can verify the origins of creative works while supporting business models involving IP access, sharing, and adaptation. The current IP system relies on costly intermediaries and post-hoc enforcement, which is no longer viable in a world where AI can consume content instantly and generate derivatives with a single click.

What we need now is an open, public registry system that provides clear proof of ownership and allows IP creators to operate efficiently, while enabling AI systems and other web applications to easily integrate with it. Blockchain is the ideal solution: it supports decentralized IP registration, offers tamper-proof proof of creation, and allows third-party applications to easily identify, authorize, and interact with these IPs.

Of course, some remain skeptical about whether technology can truly protect IP. After all, Web 1.0, Web 2.0, and now the AI revolution have often coincided with the weakening of intellectual property protections. But the primary issue lies in the fact that many existing IP business models are still focused on excluding derivative works rather than incentivizing and monetizing them. Programmable IP infrastructure not only helps creators, brands, and other stakeholders establish ownership in the digital realm but also has the potential to give rise to a new model—one that builds new businesses around frameworks for "licensed and legitimate use of IP in generative AI and other digital applications."

We've already seen early experiments in this direction in the NFT space, such as using CC0 licensing to amplify brand network effects and accumulate value. Taking it a step further, infrastructure developers are now building protocols and dedicated blockchains specifically designed to standardize and composite IP registration and licensing (e.g., Story Protocol). Some artists have already started leveraging protocols like Alias, Neura, and Titles to license their styles and works for creative reuse. The science fiction series Emergence by Incention, for instance, involves fans in co-creating characters and worldbuilding while retaining a persistent record of each contributor’s input through Story Protocol's registration system.

AI Represented by Webcrawlers Should Compensate Content Creators

By Carra Wu

Currently, among AI agents, the ones most aligned with market demand are not coding 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. Bots often ignore the robots.txt file (which is theoretically designed to instruct crawlers on whether they are allowed to scrape a site) and use the scraped data to fuel the core competitiveness of the world's largest tech companies. Worse yet, websites themselves end up "footing the bill" for these "uninvited visitors," bearing costs related to bandwidth and server resources. As a result, CDN providers like Cloudflare have been forced to introduce a range of blocking services. This, today, forms a fragmented and cumbersome adversarial mechanism, one that could be entirely replaced by a more reasonable system.

We have previously 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 trend is also evident in the data: over 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 blocked AI scrapers; now, that figure has soared to 37% and continues to rise rapidly.

So, instead of persistently blocking all requests suspected of being from bots, could we find a middle ground? One potential new model: AI crawlers no longer "freeload" on web content but pay for their data scraping activities. Blockchain could serve as the enforcement layer for this model: every crawler agent would hold cryptocurrency and initiate on-chain negotiations for access with a website's "gatekeeper agent" or paywall system using the x402 protocol.

The challenge lies in the fact that robots.txt (aka the "Robots Exclusion Standard") has become an industry default practice since the 1990s, and overturning it would require large-scale industry coordination or intervention from CDN providers like Cloudflare. On the other hand, a separate pathway could be developed for human users: by proving their "human identity" through something like World ID (see above), they could continue to access content for free.

In this way, AI's content collection activities can compensate creators at the source of the collection, while human users can still enjoy an internet that embodies "freedom of information."

More Privacy-Friendly Ads: Precise Without Overstepping

By Matt Gleason

AI is already transforming the way we shop, but can advertising become a bit more "useful"? Many people hate ads because they are either irrelevant or overly intrusive. Even "personalized ads" can feel unsettling when they are too accurate and come from extensive personal data usage, leaving individuals feeling "spied on."

Some applications attempt to monetize through paywalls, like video viewing or unlocking game levels. Cryptographic technologies can help reshape this logic. When integrated with blockchain, personalized AI agents can deliver ads based on user-defined preferences without exposing their private data. At the same time, these systems can reward users with cryptocurrency for voluntary engagements.

Technically, this model requires:

 A low-fee digital payment system: Ad engagement rewards must support high-frequency, micro-transactions, necessitating a system with high speed and low cost;

 Privacy-protecting data verification mechanisms: AI ad agents need to verify whether users fit certain demographic profiles without exposing specific data—Zero-Knowledge Proof (ZKP) technology can enable this;

 A new incentive model: If micro-payment (<$0.05) ad revenue models become mainstream, users can proactively choose to watch ads and earn rewards, shifting from "passive exploitation" to "voluntary participation."

Humans have long sought to make ads more useful, whether offline or online. By reshaping advertising systems to be "AI + blockchain"-driven, there's finally hope for ads to become truly beneficial: non-intrusive yet rewarding.

This could also make ad placements themselves more valuable while potentially disrupting today’s highly invasive "ad-exploitation economy," replacing it with a human-centric system: one where users are no longer the "product" but rather "participants."

Shaping the Future of AI

AI Companions Owned and Controlled by Humans

Author: Guy Wuollet

Today, many people spend more time on their devices than in face-to-face interactions, and this screen time is increasingly dominated by engagements with AI models and content curated by AI. In practice, these models are already offering a form of companionship—whether for entertainment, information retrieval, niche interests, or educating children. It’s easy to imagine a near future where AI companions are widely used in education, healthcare, legal advice, and social companionship, becoming a common mode of human interaction.

Future AI companions will exhibit infinite patience and can be highly customized to suit the needs of an individual. They will not just be assistants or "robot servants" but could become deeply valued relationship entities. That raises a critical question: who will own and control these relationships—the users themselves, or companies and other intermediaries? If concerns around social media content curation and moderation have troubled you over the past decade, this issue will grow even more intricate and deeply personal in the future.

In fact, this idea has been proposed before (see here and here): blockchain and other censorship-resistant custody platforms might represent the clearest path to achieving censorship-proof, user-controlled AI. While individual users could run local models and purchase GPUs on their own, the majority either can’t afford it or simply don’t know how to do it.

Although we are still some distance away from the widespread adoption of AI companions, the technology to make this a reality is advancing rapidly: text-based AI companions are already performing exceptionally well, and visual avatars have seen significant improvement. Blockchain performance is also steadily improving. To make censorship-proof AI companions more accessible to users, further improvements in the user experience (UX) of crypto applications are needed. Fortunately, blockchain wallets like Phantom have already simplified on-chain interactions, while embedded wallets, passkeys, and account abstraction technologies are enabling self-custodied wallets without requiring users to manage seed phrases themselves.

Additionally, high-throughput, trustless computing technologies like Optimistic and zero-knowledge co-processors will allow us to build meaningful and lasting relationships with digital companions.

In the near future, the discussion will likely shift from “When will we see anthropomorphic digital companions and virtual avatars?” to “Who gets to control them, and in what ways?”