a16z: 11 major use cases to understand the cross-fusion of AI and encryption

Source: a16z crypto; Translation: Golden Finance xiaozou

The economic model of the Internet is undergoing changes. When the open Internet collapses into a simple prompt box, we can’t help but ask: Will AI lead us to the open Internet, or will it build a new paywall maze? And will the highly centralized giant companies or the vast user community control all this?

This is the area where encryption technology shows its strengths. We have discussed the intersection of AI and encryption technology many times. In short, blockchain provides a new paradigm for building Internet services and networks - decentralization, trustworthy and neutrality, and truly owned by users. By reconstructing the economic model of the current system, blockchain can effectively check and balance the increasingly obvious centralized trends in AI systems, helping to build a more open and reliable Internet.

Encryption technology can optimize AI systems and vice versa - this concept is not new, but is often vaguely defined. Certain cross-cutting fields have attracted builders and users, such as verifying "human proof" at a time when low-cost AI is rampant. But other use cases still seem to take years or even decades to achieve. In this article, we will share 11 practical use cases for the integration of encryption and AI, aiming to stimulate discussions on issues such as feasibility and challenges to be solved. These cases are based on the technology being developed at the moment, from handling massive micropayments to ensuring human control and future AI, all involved.

1. Persistent data and context in AI interaction

Author: Scott Duke Kominers, Partner in Research at a16z crypto

Generative AI is thriving on data, but for many applications, context (i.e., state and background information related to interaction) is no less important than data or even more critical.

Ideally, an AI system (whether it is an agent, a large language model interface or other applications) should remember a lot of details like the type of project you are working on, communication style, preferred programming languages, and other projects. But in reality, users often need to re- establish these contexts in different interactions within a single application (such as every time a new ChatGPT or Claude session is started), let alone switch between systems.

Currently, the context of generative AI applications is almost impossible to migrate between different systems.

With blockchain technology, AI systems can transform key context elements into persistent digital assets. These assets can be loaded at the beginning of the session and transported seamlessly across AI platforms. More importantly, blockchain may be the only solution that has both forward compatibility and interoperability commitments—because these features are based on the core attributes of the blockchain protocol.

The gaming and media are natural applications: user preferences can persist across different games and environments. But the real value lies in the field of knowledge application (AI needs to understand the user's knowledge structure and learning methods) and professional AI use cases such as programming. Of course, enterprises have long begun to develop customized robots with business-specific global contexts—but such contexts are usually not able to be migrated, and even among different AI systems used within the same organization.

Organizations are just beginning to realize this problem. The closest universal solution to the present is a custom robot with a fixed persistence context. But the context portability between users within the platform has begun to sprout off the chain: for example, the Poe platform allows users to rent out their own customized robots.

Introducing such activities onto the chain will enable our interactive AI systems to share a context layer that contains all the key elements of digital activities. They can immediately understand our preferences and optimize the user experience more accurately. On the contrary, just like the on-chain intellectual property registration mechanism, allowing AI to reference persistent chain contexts can also create new market interactions around prompt words and information modules—for example, users can directly authorize or monetize their own expertise while maintaining control over the data. Of course, the shared context will also achieve many possibilities that we have not yet envisioned.

2. The general identity system of the agent

Author: Sam Broner, Partner of a16z crypto investment team

Identity—the authoritative credentials that record the nature of things—is the invisible infrastructure that underpins today’s digital discovery, aggregation and payment systems. Since the platform encloses this infrastructure within the wall, the identity we perceive is only part of the finished product: Amazon assigns identifiers (ASIN or FNSKUs) to the product, centrally displays the product and assists users in discovering and paying. The same is true for Facebook: user identities form the basis of their information flow, running through all scenarios within the application, including Marketplace product lists, natural posts and paid ads.

With the development of AI agents, all this is about to change. When more companies use their agents for customer service, logistics, payment and other scenarios, their platforms will become less like applications of a single interface, but will run across multiple carriers and platforms, accumulate deep contexts, and perform more tasks for users. However, if the agent identity is bound to a single market, it will not be available in other important scenarios (mail threads, Slack channels, and other products).

Therefore, agents need a unified "digital passport". Without it, we will not be able to pay for an agent, verify its version, query its functionality, confirm its proxy objects, or track its cross-app reputation. Agent identities need to have wallets, API registries, update logs and social credentials functions - allowing any interface (email, Slack, or other agents) to be identified and interacted with in a unified way. The lack of the shared basic element of "identity", the infrastructure needs to be rebuilt from scratch every time the integration is integrated, and the discovery mechanism is always in a temporary state, and the user will lose context when switching channels.

We have the opportunity to design agile infrastructure from first principles. So how to build a trusted neutral identity layer that is richer than DNS records? Instead of recreating an integrated platform that bundles identity with discovery, aggregation, and payment, it is better to allow agents to receive payments, list functions, and exist in multiple ecosystems without being subject to specific platforms. This is the value of the intersection of encryption technology and AI - blockchain networks provide permissionless composability, allowing developers to create more useful agents and better user experiences.

Currently, vertical integration solutions such as Facebook or Amazon do provide a better user experience—the inherent complexity of creating excellent products includes ensuring top-down coordination and unity of all links. However, this convenience is expensive, especially when the cost of building agent aggregation, marketing, monetization and distribution of software decreases, while the application scenarios of agents continue to expand. While it still takes effort to match the user experience of vertically integrated providers, the trusted and neutral agent identity layer will allow entrepreneurs to truly have their own digital passports and inspire them to boldly innovate in the distribution and design field.

3. Forward-compatible human proof mechanism

Author: Jay Drain Jr., investment partner at a16z crypto; Scott Duke Kominers, research partner at a16z crypto

As AI increasingly penetrates into various types of network interactions (from deep fakes to social media manipulation, providing support for various robots and agents), it becomes increasingly difficult to identify whether online interaction objects are real humans. This crisis of trust is not a hidden danger in the future, but the current reality - from the water army comment section of the X platform to the robot of dating software, the boundary between reality and virtual is blurring. In this environment, humanity has proven to be a critical infrastructure.

Digital ID cards (including centralized IDs used by the U.S. Transportation Security Administration) are a way to verify human identities. This type of ID includes all credentials that can prove personal identity, such as username, PIN code, password, third-party authentication (such as citizenship or credit rating). The value of decentralization is obvious here: when this data is stored in a centralized system, the issuer can revoke access, charge fees, or assist in monitoring at any time. Decentralization completely reverses this power structure: users, rather than platform gatekeepers, control their identities, making them safer and censorship-resistant.

Unlike traditional identity systems, decentralized human proof mechanisms (such as Worldcoin's Proof of Human) allow users to independently store and verify their human identities while protecting privacy and trustworthiness. Just as driver's licenses are available everywhere without being restricted by the time and location of issuance, decentralized PoP (Proof of Personhood) can serve any platform as a reusable base layer, including platforms that have not yet been born. In other words, blockchain-based PoP has forward compatibility because it has:

Portability: The protocol is available for integration as a public standard for any platform. Decentralized PoP is managed through public infrastructure and is completely controlled by users. This makes it thoroughly portable and compatible with any platform now or in the future.

No permission accessibility: The platform can choose to recognize PoP IDs independently without going through a gatekeeper API that may discriminate against different use cases.

The challenge facing this field is adoption rate. Although there are no large- scale human-proven applications, we expect critical user count, early partner and killer applications to accelerate their popularity. Each application that adopts a specific digital ID standard will increase the value of the ID to the user, thereby attracting more users to obtain the ID, thereby prompting more applications to integrate the ID as a human authentication method (because the on-chain ID is inherently interoperable, this network effect can be formed quickly).

We have seen mainstream consumer applications in the gaming, dating and social media fields announce partnerships with World ID to help users confirm that they are playing, chatting and trading with real humans (and expected specific objects). This year, new identity protocols such as Solana Attestation Service (SAS) emerged—although human proofs are not issued directly, SAS allows users to associate off-chain data (such as KYC inspections required for compliance or investment qualification certification) with Solana wallet privacy to build a decentralized identity. These signs suggest that the turning point of decentralized PoP may not be far away.

Human proof is not only about banning robots, but also about demarcating the clear boundary between AI agents and human networks. It enables users and applications to distinguish between human-computer interaction and creates space for a better, safer and more realistic digital experience.

4. Decentralized Physical Infrastructure Network (DePIN)

for AI

Author: Guy Wuollet, Partner of a16z crypto investment team

Although AI is a digital service, its development is increasingly subject to bottlenecks in physical infrastructure. Decentralized Physical Infrastructure Network (DePIN)—a new model for building and operating entity systems—can help popularize the computing infrastructure required for AI innovation, making it less costly, more resilient and more censor-resistant.

How to achieve it? Energy and chip acquisition are the 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. Together, these computers can form a market for unlicensed computing resources, creating a level playing field for the development of new AI products.

Other application scenarios include distributed training and fine-tuning of large language models, and distributed networks of model inference. Decentralized training and reasoning can significantly reduce costs because it uses originally idle computing resources. It also provides censorship resistance to ensure that developers will not be deprived of the platform's right to use by hyperscale cloud service providers (centralized cloud service giants that provide flexible scalable computing infrastructure).

The AI ​​model focuses on a few companies for a long time; decentralized networks help create more cost-effective, censor-resistant and scalable AI systems.

5. **Infrastructure and protection mechanism for interaction between

AI intelligence, terminal service providers** and users

Author: Scott Duke Kominers, Partner in Research at a16z crypto

As AI tools' ability to perform complex tasks and multi-level interaction chains improves, the demand for autonomous interaction between agents will increase significantly.

For example, an AI agent may need to obtain specific computing data, or call a professional agent to perform special tasks - such as assigning statistical robots to develop and run model simulations, or enabling image generation robots in marketing material production. AI agents can also create huge value by completing the entire transaction process on behalf of users—such as searching for air tickets based on preferences, or discovering and ordering a new book of a certain type.

There is currently no common inter-agent market. Such cross-system queries are mainly implemented through explicit API connections, or are limited to a closed ecosystem that supports internal agent calls.

Broadly speaking, most AI agents currently operate in an island-like ecosystem, with relatively closed APIs and lacking architectural standardization. Blockchain technology can help protocols establish open standards, which is crucial for short-term adoption. In the long run, this also supports forward compatibility: new AI agents can seamlessly access the existing underlying network when they appear. Thanks to the architectural features of interoperability, open source, decentralization and easy-to- upgrade, blockchain can better adapt to AI innovation iteration.

With the development of the market, many companies have begun to build blockchain infrastructure for inter-agent interaction: for example, Halliday recently launched a standardized cross-chain architecture protocol that supports AI workflow interaction, ensuring that AI behavior does not deviate from user intentions through protocol-level protection. Catena, Skyfire and Nevermind use blockchain to achieve independent payment between agents without manual intervention. More similar systems are under development, and Coinbase has even begun providing infrastructure support for these attempts.

6. Maintain synchronization of AI/ atmosphere coding

applications

Author: Sam Broner, Partner of the a16z crypto investment team; Scott Duke Kominers, Partner of the a16z crypto research

The revolutionary advances in generative AI have made software development easier than ever. The coding efficiency is on an order of magnitude, and more importantly, it is now possible to program in natural language, and even inexperienced developers can fork existing programs or build new applications from scratch.

However, while AI-assisted coding creates new opportunities, it also introduces a large amount of entropy increase inside and outside the program. Although "Vibe coding" abstracts the complex dependency network at the bottom of the software, this programming method may cause hidden dangers in terms of functionality and security as the source library and other inputs change. In addition, when people use AI to create personalized applications and workflows, the difficulty of these systems connecting with other systems also increases. In fact, the operation logic and output structure of two atmosphere encoding programs that perform the same task may be completely different.

For a long time, standardization of ensuring consistency and compatibility was first undertaken by the file format and operating system, and later by the integration of shared software and APIs. But in a world where software evolves, deforms and forks in real time, the standardization layer needs to be broad accessibility and continuous upgradeability, while maintaining user trust. More importantly, AI alone cannot solve the problem of inspiring people to build and maintain these links.

Blockchain technology can solve these two problems at the same time: embedded in user-customized software builds through synchrony layers, dynamic updates to ensure changing cross-platform compatibility. In the past, large companies could spend millions of dollars to hire "system integrators" such as Deloitte to customize Salesforce instances, but now engineers can create custom interfaces to view sales information in a weekend. But as the number of customized software surges, developers need assistance to keep these applications running in sync.

This is similar to the current development model of open source software libraries, but the difference is that it is continuous updates rather than regular releases – and an added incentive layer. These two points are easier to achieve with the support of encryption technology. Just like other blockchain-based protocols, co-ownership of the synchronization layer inspires parties to continue investing in improvement. Developers, users (and their AI agents) and other participants are rewarded for introducing, using and developing new features and integrations.

Instead, sharing ownership makes all users relevant to the overall successful stake in the protocol, which becomes a buffer against malicious behavior. Just as Microsoft does not break the .docx file standard to avoid affecting user and brand reputation, co-owners of the synchronization layer are equally no motive to introduce botched or malicious code into the protocol.

Like all previous software standardized architectures, there is huge potential for network effects in this field. As the "Cambrian explosion" of AI encoding software continues, heterogeneous system networks that need to be connected will expand dramatically. In short: the atmosphere coding should be kept in sync, and it should not be based on Vibe alone. Encryption technology is exactly the answer.

7. Micro payment system that supports revenue sharing

Author: Liz Harkavy, Partner of a16z crypto investment team

AI tools and agents represented by ChatGPT, Claude and Copilot provide a new and convenient way to enjoy the digital world. But regardless of pros and cons, they are shaking the economic foundation of the open Internet. The impact of reality has already emerged - education platforms are facing a sharp decline in traffic as students turn to AI tools, and many American newspapers are suing OpenAI for copyright infringement. If the incentive mechanism cannot be reconstructed, we will witness the increasingly closed Internet, the number of paywalls and the number of content creators is decreasing.

Although policy measures exist, while legal procedures are advancing, a number of technical solutions are emerging. Perhaps the most promising (and most complex) solution is to implant a revenue sharing system in a network architecture: when AI-driven behavior facilitates transactions, the content sources involved in the decision-making process should be shared. The affiliate marketing ecosystem has achieved similar attribution tracking and profit distribution, while more advanced versions can automatically track all contributors of the information chain and give rewards - blockchain can obviously play a role in the traceability chain.

But such systems require new infrastructure with special functions: micropayment systems that can handle multi-source microtransactions, attribution protocols that fairly evaluate various contributions, and governance models that ensure transparency and impartiality. Existing blockchain tools have shown potential, such as Rollup and Layer2 solutions, AI-native financial institution Catena Labs, financial infrastructure protocol 0xSplits, etc., which can achieve nearly zero-cost transactions and more refined payment splits.

Blockchain will empower intelligent payment systems through the following mechanisms:

• Nanopayment can be split into multiple data providers, and a single user interaction can automatically distribute extremely small payments to all contribution sources through smart contracts.

• Smart contracts support executable traceability payments based on completed transactions, and compensate for information sources that are confirmed to affect purchase decisions after the transaction occurs in a completely transparent and traceable manner.

• Support complex and programmable payment distribution schemes, achieve fair distribution of returns through code mandatory rules rather than centralized decisions, and establish a trust-free financial relationship between autonomous agents

As these emerging technologies mature, they will create new economic models for the media industry that capture the complete value chain—from creators to platforms to users.

8. **Blockchain as intellectual property and traceability

registration book**

Author: Scott Duke Kominers, Partner in Research at a16z crypto

The rise of generative AI urgently requires an efficient and programmable intellectual property registration and tracking mechanism - to ensure that content sources are traceable, and to support business models around IP access, sharing and remixing. The current IP framework relies on high-cost intermediaries and post-accountability, and can no longer adapt to the new era of instant AI consumption of content and one-click generation of variants.

We need an open and public registration system that can provide clear proof of ownership, allowing IP creators to interact conveniently and efficiently, and allowing direct connection between AI and other network applications. Blockchain is a perfect solution: IP registration can be completed without an intermediary, providing tamper-free proof of origin, and allowing third-party applications to easily identify, authorize and call these IPs.

When the first two eras of the Internet (and the ongoing AI revolution) are often associated with weakening intellectual property protection, the idea that technology can protect IP will naturally arouse many doubts. The problem is that most current IP business models focus on excluding derivative works rather than incentivizing and monetizing creations. But programmable IP infrastructure not only allows creators, franchisees and brands to clarify IP ownership in the digital space, but also opens the door to IP sharing business models centered on digital applications such as generative AI - which actually turns the main threat of generative AI to creative work into opportunities.

We have seen creators test new models in the early stages of the NFT field, and some companies use NFT assets on Ethereum to achieve network effects and value accumulation under the construction of CC0 brands. Recently, there have been protocols specially built for standardized, combinable IP registration and authorization, and even dedicated blockchains (such as Story Protocol). Some artists have begun to authorize their artistic styles and works to be used for creative remixing through agreements such as Alias, Neura and Titles. Incention's Emergence series allows fans to participate in the co-creation of the science fiction universe and characters, and tracks the creation content of each contributor through a blockchain registration book built on Story.

9. Internet crawlers that help content creators monetize

Author: Carra Wu, Partner of a16z crypto investment team

The AI ​​agent with the most product market compatibility at the moment is not programming or entertainment assistants, but web crawlers—digital agents who independently shuttle the Internet, collect data and decide link tracking paths.

It is estimated that nearly half of the network traffic has originated from non-human subjects. Crawlers often ignore the robots.txt protocol (this application is used to inform automated crawler websites to access, but in fact they are weak in binding), and the data they collect eventually becomes a competitive barrier for some technology giants. To make matters worse, the website has to bear bandwidth and CPU resource costs for these uninvited guests, as if serving an endless anonymous data harvester. The interception scheme provided by CDN (content distribution network) service providers such as Cloudflare are actually remedial measures 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 top 10,000 websites in the world blocked AI crawlers, and now this proportion has reached 37%. As the website's main defense measures are upgraded and user dissatisfaction accumulates, this number will surely continue to rise.

If you do not rely on CDN to completely ban visitors who are suspected of crawlers, can you find a compromise? Rather than abuse systems designed for human traffic, AI crawlers may be able to obtain data collection rights for a fee. This is where blockchain comes in: in this case, 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, which has been used since the 1990s, has been deeply rooted and 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 previous article). In this way, content creators and website owners can obtain reasonable compensation from the AI ​​training set during the data collection process, while humans can still enjoy the Internet with free information.

10. A new paradigm of both accurate and privacy

Author: Matt Gleason, a16z crypto security engineer

AI has begun to change the way we shop online, but what if the ads we see on a daily basis can be really useful? The reason people hate advertising is obvious: irrelevant advertising is purely noise, while overly accurate AI advertising (based on massive consumption data) is creepy. Other applications are monetized through non-skippable ad walls such as streaming services or game levels.

Encryption technology can reconstruct advertising mechanisms and solve these pain points. Combined with blockchain’s personalized AI agents, we can find a balance between “unrelated advertising” and “horror precision”—select ads according to user custom preferences. The key is that all this does not require global exposure of user data, and can directly compensate data sharers or advertising entrants.

Required technical elements include:

• Low-rate digital payment : In order to compensate users' advertising interactions (watch/click/conversion), enterprises need to send small payments at high frequency. This requires a high throughput, near-zero fee payment system.

• Privacy protection data verification : AI needs to prove that consumers meet certain demographic characteristics. Zero-knowledge proof can be verified while protecting privacy.

• Incentive mechanism : If the Internet adopts a monetization model based on micropayment (such as <0.05 US dollars per interaction), users can choose to watch advertisements for compensation, and transform the current "extraction model" into "participation model".

Humans have been pursuing advertising relevance for hundreds of years (offline) and decades (online). Reconstructing advertisements through encryption and AI perspectives will eventually make them truly useful: accurate but not terrifying, achieving win-win results for all parties - for builders and advertisers, unlocking a more sustainable and consistent new incentive structure; for users, providing more ways to explore the digital world.

This will not only not detract from the value of the advertising space, but will instead increase its value. It is more likely to subvert the current deeply rooted extractive advertising economy and replace it with a more humanized system: users are seen as participants rather than products.

11. AI partners owned and controlled by humans

Author: Guy Wuollet, Partner of a16z crypto investment team

Modern people spend more time on electronic devices than face-to-face communication, and more and more of the time is spent interacting with AI models and content screening. These models have essentially provided 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 interaction methods for human beings.

Future AI partners will have unlimited patience and can deeply adapt to the specific needs of individual users. They are not only assistants or robot servants, but are more likely to develop into highly valued "relationships". Therefore, ownership and control of these relationships (users or businesses and other intermediaries) become crucial. If you have been concerned about the review and censorship of social media content over the past decade, this issue will become exponentially complex and more personal in the future.

Censorship-resistant hosting platforms (such as blockchain) provide the most powerful path to achieving user-controllable and non-censored AI - this is not a new argument (already discussed in the previous article). Although individuals can run local models or purchase their own GPUs, most people either cannot afford it or lack technical capabilities.

Although it will take some time for AI partners to become popular, related technologies are rapidly evolving: text-based anthropomorphic partners are quite mature, virtual image is significantly improved, and blockchain performance continues to improve. To ensure the ease of use of censorship- resistant assistants, it is necessary to rely on encrypted applications to have a better user experience. Thankfully, wallets such as Phantom have greatly simplified blockchain interactions, with embedded wallets, pass keys and account abstraction technologies allowing users to self-host their wallets without memorizing mnemonics. With high-throughput trust-free computers (using technologies such as optimistic proofs and ZK coprocessors), it will be possible to establish meaningful and lasting relationships with digital partners.

In the near future, the focus of discussion will shift from "when can you see realistic digital partners" to "who can control them and how to control them".