Source: a16z crypto; Translation: Jinse Finance xiaozou
The economic model of the Internet is changing. As the open Internet collapses into a simple prompt box, we can't help but ask: Will AI lead us to an open Internet, or build a new paywall maze? And who will control all this, a highly centralized giant enterprise or a vast user community?
This is where cryptography comes in handy. We have discussed the intersection of AI and cryptography many times before. In short, blockchain provides a new paradigm for building Internet services and networks - decentralized, trusted, neutral, and truly owned by users. By reconstructing the economic model of the current system, blockchain can effectively check and balance the increasingly obvious centralization trend in AI systems, and help build a more open and reliable Internet.
The idea that crypto can improve AI systems and vice versa is not new, but it is often vaguely defined. Some intersections are already attracting builders and users, such as verifying "proof of humanity" at a time when low-cost AI is rampant. But other use cases still seem years or even decades away. In this article, we share 11 real-world use cases where crypto and AI converge to stimulate discussion about what is possible, what challenges remain, and more. These examples are based on technologies being developed today, ranging from processing massive micropayments to ensuring human control over the relationship with future AI.
1. Persistent Data and Context in AI Interactions
By: Scott Duke Kominers, a16z crypto research partner
Generative AI thrives on data, but for many applications, context—the state and background information associated with an interaction—is just as important as, or even more critical than, the data.
Ideally, an AI system (whether it’s an agent, a large language model interface, or something else) should be able to remember a lot of details about the type of project you’re working on, your communication style, your preferred programming language, etc. But in reality, users often need to re-establish these contexts in different interactions within a single application (such as every time you start a new ChatGPT or Claude session), not to mention when switching across systems.
Currently, the context of generative AI applications is almost impossible to transfer between different systems.
With blockchain technology, AI systems can transform key contextual elements into persistent digital assets. These assets can be loaded at the beginning of a session and transferred seamlessly across AI platforms. More importantly, blockchain may be the only solution that has the promise of both forward compatibility and interoperability - because these features are based on the core properties of the blockchain protocol.
Games and media are natural applications: user preferences persist across games and environments. But the real value lies in knowledge applications (AI needs to understand the user's knowledge structure and learning style) and specialized AI use cases such as programming. Of course, companies have long developed customized bots with business-specific global context - but this context is usually not transferable, even between different AI systems used within the same organization.
Organizations are just beginning to realize this problem. The closest universal solution is currently custom robots with fixed persistent context. But context portability between users within a platform is beginning to sprout off-chain: for example, the Poe platform allows users to rent out their own custom robots.
Bringing this type of activity to the chain will allow the AI systems we interact with to share a contextual layer that contains the key elements of all digital activities. They can immediately understand our preferences and optimize the user experience more accurately. Conversely, just like the on-chain intellectual property registration mechanism, allowing AI to reference persistent on-chain context can also give rise to new market interactions around prompt words and information modules - for example, users can directly license or monetize their expertise while maintaining control over their data. Of course, shared context will also enable many possibilities that we have not yet imagined.
2. Universal identity system for intelligent agents
Author: Sam Broner, Partner at a16z Crypto Investment Team
Identity—the authoritative record of what things are—is the invisible infrastructure that underpins today’s digital discovery, aggregation, and payment systems. Because platforms keep this infrastructure behind walls, the identity we perceive is only part of the finished product: Amazon assigns identifiers (ASIN or FNSKU) to products, centralizes product display, and facilitates user discovery and payment. The same is true for Facebook: user identities form the foundation of its information flow and run through all scenarios within the app, including Marketplace product listings, organic posts, and paid ads.
With the development of AI agents, all this is about to change. As more companies use agents for scenarios such as customer service, logistics, and payment, their platforms will become less and less like single-interface applications, but will run across multiple carriers and platforms, accumulating deep context and performing more tasks for users. However, if the identity of the agent is bound to a single market, it will not be usable in other important scenarios (email threads, Slack channels, and other products).
Therefore, agents need a unified "digital passport". Without it, we will not be able to pay agents, verify their versions, query their capabilities, confirm who they are acting for, or track their reputation across applications. Agent identities need to be wallets, API registries, change logs, and social credentials - allowing any interface (mail, Slack, or other agents) to identify and interact with them in a unified way. Without the shared foundation of "identity", each integration requires rebuilding infrastructure from scratch, discovery mechanisms are always in a temporary state, and users lose context when switching channels.
We have an opportunity to design agent infrastructure from first principles. So how do we build a trusted neutral identity layer that is richer than DNS records? Rather than reinventing an all-in-one platform that bundles identity with discovery, aggregation, and payments, we can enable agents to receive payments, list functionality, and exist in multiple ecosystems without being tied to a specific platform. This is where the value of the intersection of crypto and AI lies - blockchain networks provide permissionless composability, enabling developers to create more useful agents and better user experiences.
At present, vertically integrated solutions such as Facebook or Amazon do provide a better user experience - the inherent complexity of building great products includes ensuring top-down coordination of all links. But this convenience comes at a high price, especially as the cost of building agent aggregation, marketing, monetization and distribution software decreases, and the use cases of agents continue to expand. Although it will take work to match the user experience of vertically integrated providers, a trusted and neutral agent identity layer will allow entrepreneurs to truly own their own digital passports and encourage them to innovate boldly in the areas of distribution and design.
3. Forward-compatible human proof mechanism
By: Jay Drain Jr., a16z crypto investment partner; Scott Duke Kominers, a16z crypto research partner
As AI increasingly permeates all types of online interactions (from deep fakes to social media manipulation, powering various robots and intelligent entities), it becomes increasingly difficult to distinguish whether the person you are interacting with online is a real person. This trust crisis is not a future risk, but a current reality - from the water army comment area of platform X to the robots of dating apps, the boundaries between real and virtual are blurring. In this environment, human proof becomes critical infrastructure.
Digital IDs, including the centralized IDs used by the Transportation Security Administration in the U.S., are a way to verify a person’s identity. These IDs contain everything from usernames, PINs, passwords, third-party verification (like citizenship or credit ratings) to prove a person’s identity. The value of decentralization is clear here: when this data is stored in a centralized system, the issuer can revoke access, charge fees, or facilitate surveillance at any time. Decentralization completely flips this power structure: users control their identity, not platform gatekeepers, making it more secure and censorship-resistant.
Unlike traditional identity systems, decentralized human proof mechanisms (such as Worldcoin's Proof of Human) allow users to autonomously keep and verify their human identity while protecting privacy and trusted neutrality. Just as a driver's license is available everywhere regardless of the time and place of issuance, decentralized PoP (Proof of Personhood) can serve as a reusable base layer for any platform, including those that have not yet been born. In other words, blockchain-based PoP is forward-compatible because it has:
Portability: The protocol is available as a public standard for integration on any platform. Decentralized PoPs are managed through a public infrastructure and are completely controlled by users. This makes it completely portable and compatible with any platform now or in the future.
Permissionless accessibility: Platforms can choose to recognize PoP IDs on their own, without going through a gatekeeper API that may discriminate against different use cases.
The challenge in this space is adoption. While there are no large-scale use cases for human proof yet, we expect critical mass, early partners, and killer apps to accelerate adoption. Each application that adopts a particular digital ID standard increases the value of that ID to users, which in turn attracts more users to acquire that ID, which in turn drives more applications to integrate that ID as a human authentication method (this network effect can form quickly due to the inherent interoperability of on-chain IDs).
We have seen mainstream consumer applications in the fields of games, dating, and social media announce partnerships with World ID to help users confirm that they are playing, chatting, and trading with real people (and the specific people they expect). New identity protocols such as Solana Attestation Service (SAS) have also emerged this year. Although it does not directly issue human certificates, SAS allows users to privately associate off-chain data (such as KYC checks or investment qualification certifications required for compliance) with Solana wallets to build decentralized identities. These signs indicate that the turning point of decentralized PoP may not be far away.
Human proof is not just about banning robots, but about drawing clear boundaries between AI agents and human networks. It enables users and applications to distinguish between human and machine interactions, creating space for better, safer, and more authentic digital experiences.
4. Decentralized Physical Infrastructure Network (DePIN) for AI
By Guy Wuollet, Partner at a16z Crypto Investment Team
Although AI is a digital service, its development is increasingly constrained by physical infrastructure bottlenecks. Decentralized Physical Infrastructure Networks (DePIN) - a new model for building and operating physical systems - can help democratize the computing infrastructure required for AI innovation, making it cheaper, more resilient, and more censorship-resistant.
How to achieve it? Energy and chip acquisition are two core obstacles to the development of AI. Decentralized energy can improve power supply, and builders also integrate idle chips in gaming PCs, data centers and other scenarios through DePIN. These computers can jointly form a permissionless computing resource market, creating a fair competition environment for the development of new AI products.
Other application scenarios include distributed training and fine-tuning of large language models, and distributed networks for model reasoning. Decentralized training and reasoning can significantly reduce costs because they utilize previously idle computing resources. At the same time, they provide censorship resistance to ensure that developers will not be deprived of platform use rights by hyperscale cloud service providers (centralized cloud service giants that provide elastically scalable computing infrastructure).
The problem of AI models being concentrated in a few companies has long existed; decentralized networks can help create more cost-effective, censorship-resistant, and scalable AI systems.
5. Infrastructure and protection mechanisms for interaction between AI agents, terminal service providers and users
By: Scott Duke Kominers, a16z crypto research partner
As AI tools improve their ability to perform complex tasks and multi-level interaction chains, the demand for autonomous interaction between intelligent agents will grow significantly.
For example, an AI agent may need to obtain specific computing data, or call on professional agents to perform special tasks - such as assigning statistical robots to develop and run model simulations, or enabling image generation robots in the production of marketing materials. AI agents can also create huge value by completing the entire transaction process on behalf of users - such as searching and booking air tickets based on preferences, or discovering and ordering new books of a certain type.
Currently, there is no universal inter-agent market, and such cross-system queries are mainly implemented through explicit API connections or are limited to closed ecosystems that support internal agent calls.
Broadly speaking, most AI agents currently operate in isolated ecosystems, with relatively closed APIs and a lack of architectural standardization. Blockchain technology can help establish open standards for protocols, which is crucial for short-term adoption. In the long run, this also supports forward compatibility: new AI agents can be seamlessly connected to existing underlying networks when they emerge. Thanks to the interoperability, open source, decentralization, and easy-to-upgrade architectural characteristics, blockchain can better adapt to AI innovation iterations.
As the market develops, many companies have begun to build blockchain infrastructure for interactions between intelligent entities: for example, Halliday recently launched a standardized cross-chain architecture protocol that supports AI workflow interactions, ensuring that AI behavior does not deviate from user intent through protocol-level protection. Catena, Skyfire, and Nevermind use blockchain to achieve autonomous payments between intelligent entities without human intervention. More similar systems are under development, and Coinbase has even begun to provide infrastructure support for these attempts.
6. Keep AI/ ambience coding applications in sync
Author: Sam Broner, Partner, a16z Crypto Investment Team; Scott Duke Kominers, Partner, a16z Crypto Research
The revolutionary progress of generative AI has made software development easier than ever before. Coding efficiency has increased by orders of magnitude, and more importantly - now that natural language programming is possible, even inexperienced developers can fork existing programs or build new applications from scratch.
But while AI-assisted coding creates new opportunities, it also introduces a lot of entropy into and out of the program. Although "vibe coding" abstracts away the complex dependency network at the bottom of the software, this programming method may cause hidden dangers in the functionality and security of the program as the source library and other inputs change. In addition, when people use AI to create personalized applications and workflows, it becomes more difficult for these systems to connect with other people's systems. In fact, two vibe coding programs that perform the same task may have completely different operating logic and output structure.
The work of standardization to ensure consistency and compatibility has historically been undertaken first by file formats and operating systems, and later by shared software and API integrations. But in a world where software evolves, morphs, and forks in real time, the standardization layer needs to be widely accessible and continuously upgradable while maintaining user trust. More importantly, AI alone cannot solve the problem of incentivizing people to build and maintain these links.
Blockchain technology can solve both problems at the same time: embedding user-customized software builds through protocolized synchronization layers, and dynamically updating to ensure cross-platform compatibility in changes. In the past, large companies may spend millions of dollars to hire "system integrators" such as Deloitte to customize Salesforce instances, but now engineers can create a customized interface for viewing sales information in a weekend. However, as the number of customized software has increased dramatically, developers need assistance to keep these applications running in sync.
This is similar to the current development model for open source software libraries, but with continuous updates rather than periodic releases — and an added incentive layer. Both of these are made easier with the power of cryptography. As with other blockchain-based protocols, shared ownership of the sync layer incentivizes all parties to continue to invest in improvements. Developers, users (and their AI agents), and other participants can all be rewarded for introducing, using, and evolving new features and integrations.
Conversely, shared ownership gives all users a stake in the overall success of the protocol, which acts as a buffer against malicious behavior. Just as Microsoft would not break the .docx file standard without risking user and brand reputation, the co-owners of the sync layer have no incentive to introduce poor or malicious code into the protocol.
As with all software standardization architectures before it, there is huge potential for network effects in this space. As the Cambrian explosion of AI coding software continues, the network of heterogeneous systems that need to stay connected will expand dramatically. In short: Vibes cannot keep up with vibe coding alone. Encryption is the answer.
7. Micropayment system supporting revenue sharing
By Liz Harkavy, Partner at a16z Crypto Investment Team
AI tools and agents such as ChatGPT, Claude, and Copilot provide a new and convenient way to navigate the digital world. But regardless of the pros and cons, they are shaking the economic foundation of the open Internet. The real impact is already evident - educational platforms are facing a sharp drop in traffic due to students turning to AI tools, and several American newspapers are suing OpenAI for copyright infringement. If the incentive mechanism cannot be restructured, we will witness an increasingly closed Internet, with more paywalls and fewer content creators.
Policy measures exist, but as legal proceedings proceed, a number of technical solutions are emerging. Perhaps the most promising (and complex) solution is to embed a revenue-sharing system into the network architecture: when AI-driven behavior leads to a transaction, the content source involved in the decision-making process should receive a share. The affiliate marketing ecosystem has achieved similar attribution tracking and revenue distribution, and more advanced versions can automatically track all contributors to the information chain and give rewards - blockchain can obviously play a role in the traceability chain.
However, such systems require new infrastructure with special features: a micropayment system that can handle multi-source microtransactions, an attribution protocol that fairly evaluates various contributions, and a governance model that ensures transparency and fairness. Existing blockchain tools have shown potential, such as Rollup and Layer2 solutions, AI-native financial institutions Catena Labs, and financial infrastructure protocols 0xSplits, which can achieve nearly zero-cost transactions and more sophisticated payment splits.
Blockchain will enable smart payment systems through the following mechanisms:
• Nanopayments can be split across multiple data providers, with a single user interaction automatically distributing micropayments to all contributing sources via smart contracts.
• Smart contracts support executable traceable payments based on completed transactions, compensating sources of information that were identified as influencing purchase decisions after the transaction occurred in a fully transparent and traceable manner
• Support complex programmable payment allocation schemes, achieve fair distribution of benefits through code-enforced rules rather than centralized decision-making, and establish trustless financial relationships between autonomous agents
As these emerging technologies mature, they will create new economic models for the media industry that capture the entire value chain – from creators to platforms to users.
8. Blockchain as an intellectual property and traceability register
By: Scott Duke Kominers, a16z crypto research partner
The rise of generative AI urgently requires efficient and programmable intellectual property registration and tracking mechanisms - to ensure that the source of content can be traced, and to support business models around IP access, sharing and remixing. The current IP framework relies on high-cost intermediaries and post-event accountability, and is no longer able to adapt to the new era of AI consuming content instantly and generating variants with one click.
We need an open, public registration system that provides clear proof of ownership, allows IP creators to interact conveniently and efficiently, and allows AI and other network applications to connect directly. Blockchain is the perfect solution: it can complete IP registration without an intermediary, provide tamper-proof traceability, and allow third-party applications to easily identify, authorize, and call these IPs.
The idea that technology can protect IP has naturally raised many questions, given that the first two eras of the internet (and the ongoing AI revolution) are often associated with weakened intellectual property protections. The problem is that most current IP business models focus on excluding derivative works rather than incentivizing and monetizing those creations. But programmable IP infrastructure not only allows creators, franchisees, and brands to clearly define IP ownership in the digital space, it also opens the door to IP-sharing business models centered on digital applications such as generative AI—effectively turning generative AI’s primary threat to creative work into an opportunity.
We have seen creators test new models in the NFT space early on, with companies using NFT assets on Ethereum to achieve network effects and value accumulation under the CC0 brand. Recently, protocols and even dedicated blockchains (such as Story Protocol) have emerged specifically for standardized, composable IP registration and authorization. Some artists have begun to license their art styles and works for creative remixing through protocols such as Alias, Neura, and Titles. Incention's Emergence series allows fans to participate in the co-creation of science fiction universes and characters, tracking the creative content of each contributor through a blockchain registry built on Story.
9. Web crawlers to help content creators monetize
By Carra Wu, Partner at a16z Crypto Investment Team
The AI agents with the best product-market fit today are not programming or entertainment assistants, but web crawlers—digital agents that autonomously traverse the internet, collect data, and decide on link-tracking paths.
It is estimated that nearly half of all web traffic originates from non-human entities. Bots routinely ignore robots.txt (which is used to tell automated crawlers what access rights to a website, but is in fact a weak constraint), and the data they collect eventually becomes a competitive barrier for some tech giants. Worse, websites have to bear the bandwidth and CPU resource costs for these uninvited guests, as if they are serving an endless stream of anonymous data harvesters. The interception solutions provided by CDN (content distribution network) service providers such as Cloudflare are actually remedies that should not exist.
We have pointed out that the original contract of the Internet - the economic covenant between content creators and distribution platforms - is facing collapse. The data confirms this trend: in the past 12 months, website owners have begun to block AI crawlers on a large scale. In July 2024, only about 9% of the world's top 10,000 websites blocked AI crawlers. Now the proportion has reached 37%. As website owners upgrade their defenses and user dissatisfaction accumulates, this number will continue to rise.
If we don't rely on CDNs to completely block visitors suspected of crawling, can we find a compromise? Instead of abusing a system designed for human traffic, AI crawlers may be able to pay for the right to collect data. This is where blockchain comes in: in this scenario, each crawler agent will hold cryptocurrency and negotiate on-chain with the website's "bouncer" agent or paywall protocol through the x402 protocol (of course, the challenge is that the Robots Exclusion standard that has been used since the 1990s is deeply rooted, and it requires large-scale group collaboration involving CDN giants such as Cloudflare to break through).
At the same time, human users can continue to obtain content for free by verifying their real identity through World ID (see above). In this way, content creators and website owners can obtain reasonable compensation for AI training sets during the data collection stage, while humans can still enjoy the Internet with free information.
10. A new advertising paradigm that is both accurate and private
By Matt Gleason, a16z crypto security engineer
AI is already changing the way we shop online, but what if the ads we see every day are actually useful? People hate ads for obvious reasons: irrelevant ads are just noise, while overly precise AI ads (based on massive amounts of consumer data) are creepy. Other apps monetize through unskippable ad walls (like streaming services or game levels).
Encryption technology can reconstruct the advertising mechanism and solve these pain points. Combined with the personalized AI agent of blockchain, it can find a balance between "irrelevant advertising" and "terrifying precision" - placing advertisements based on user-defined preferences. The key is that all this does not require global exposure of user data, and can directly compensate data sharers or advertising interactors.
The required technical elements include:
• Low-fee digital payments : To compensate users for ad interactions (views/clicks/conversions), companies need to send small payments at high frequencies. This requires a high-throughput, near-zero-fee payment system.
• Privacy-preserving data verification : AI needs to be able to prove that consumers meet certain demographic characteristics. Zero-knowledge proof can complete the verification while protecting privacy.
• Incentive mechanism : If the Internet adopts a monetization model based on micropayments (such as <$0.05 per interaction), users can choose to watch ads in exchange for rewards, transforming the current "extraction model" into a "participation model."
Humans have been pursuing advertising relevance for hundreds of years (offline) and decades (online). Reconstructing advertising through encryption and AI perspectives will eventually make it truly useful: accurate but not shocking, achieving a win-win situation for all parties - for builders and advertisers, unlocking a more sustainable and interest-aligned new incentive structure; for users, providing more ways to explore the digital world.
This will not only not devalue the ad space, but will actually increase its value. It will also hopefully overturn the current entrenched extractive advertising economy and replace it with a more humane system where users are viewed as participants, not products.
11. AI partners owned and controlled by humans
By Guy Wuollet, Partner at a16z Crypto Investment Team
Modern people spend more time on electronic devices than face-to-face communication, and more and more time is spent interacting with AI models and AI-selected content. These models essentially provide some kind of companionship - whether it is entertainment, information acquisition, interest satisfaction or children's education. It is not difficult to imagine that in the near future, AI-based educational assistants, health consultants, legal assistants and emotional partners will become the mainstream way of human interaction.
The AI companions of the future will be infinitely patient and deeply attuned to the specific needs of individual users. They will not only be assistants or robot servants, but will also likely develop into cherished "relationships". Therefore, it becomes critical to determine who owns and controls these relationships (users or companies and other intermediaries). If you have worried about content moderation and censorship on social media over the past decade, this issue will become exponentially more complex and more personal in the future.
The argument that censorship-resistant hosting platforms (such as blockchains) provide the most powerful path to user-controllable, uncensorable AI is not new (it has been discussed above). While individuals can run local models or purchase their own GPUs, most people either cannot afford it or lack the technical ability to do so.
Although it will take some time for AI companions to become ubiquitous, the technology is evolving rapidly: text-based anthropomorphic companions are already quite mature, avatars have significantly improved, and blockchain performance continues to improve. Ensuring the ease of use of censorship-resistant assistants requires a better user experience for crypto applications. Fortunately, wallets such as Phantom have greatly simplified blockchain interactions, with embedded wallets, pass keys, and account abstraction technologies that allow users to self-host wallets without memorizing seed phrase. With the help of high-throughput trustless computers (using technologies such as optimistic proofs and ZK coprocessors), it will become possible to establish meaningful and lasting relationships with digital companions.
In the near future, the discussion will shift from "when can we see lifelike digital companions" to "who can control them and how."
