a16z: 11 application scenarios that integrate Crypto and AI

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

Compiled by: Saoirse, Foresight News

The economic logic of the Internet has quietly changed. When the open network gradually shrinks into a "command input bar", we have to think: Will artificial intelligence lead us to an open Internet, or will it fall into a maze composed of new payment barriers? Will control be in the hands of large centralized enterprises or in the hands of a wide range of users?

This is where Crypto comes in. We have discussed the intersection of AI and Crypto many times - in short, blockchain is a new paradigm for reconstructing Internet service architecture, able to build a decentralized, trustworthy and neutral network system owned by users. By redefining the economic rules that support existing systems, blockchain provides an effective way to check and balance the centralized trends in the AI ​​field to achieve a more open and resilient Internet ecosystem.

The concept of two-way empowerment of Crypto and AI systems has long existed, but the combination of the two has always lacked a clear definition. Some cross-cutting fields (such as "human identity" verification in the context of the flood of low-cost AI tools) have attracted the attention of developers and users, while other application scenarios may take years or even decades to implement. Therefore, this article will share 11 cross-border application scenarios of AI and Crypto, aiming to promote related discussions: Exploring the potential possibilities and challenges of AI and Crypto, and looking forward to more innovative directions. These scenarios are based on the current technology level, covering diverse fields, from massive micropayment processing to ensuring humans have dominance in future AI interactions.

Identity and data management

1. Persistent data and context environment in AI interaction

Written by: Scott Duke Kominers

The development of generative AI is highly dependent on data, but in many application scenarios, the importance of context (i.e., interactively related state and background information) is no less than that of data, or even more critical.

Ideally, whether it is an agent, a large language model interface or other applications, the AI ​​system can remember many details such as the user's work type, communication style, preferred programming language, etc. However, in reality, users often need to reset these contexts in different sessions of the same application (such as turning on a new ChatGPT or Claude session), let alone switch between different systems. Currently, the context of a generative AI application is almost impossible to migrate to other applications.

With blockchain technology, AI systems can transform key context elements into persistent digital assets, allowing them to load at session startup and seamlessly transmit across different AI platforms. Furthermore, based on its characteristics, blockchain may be the only solution that meets both "forward compatibility" and "interoperability" requirements.

This application is especially applicable in AI-mediated gaming and media fields—user preferences (from difficulty settings to key bindings) can be consistent across games and scenarios. But the real value is reflected in knowledge application scenarios (such as AI needs to understand the user's knowledge reserves and learning models) and professional AI applications (such as programming assistance). Of course, some enterprises have developed custom robots for specific business backgrounds, but in such scenarios, the context is usually not portable across systems, and it is difficult to share even among different AI tools within the enterprise.

Organizations are just beginning to realize this problem, and the current universal solution is custom robots with fixed backgrounds. However, the contextual migration between users within the platform has begun to emerge off-chain: for example, on the Poe platform, users can rent custom robots to others for use.

After such scenarios are put on the chain, the AI ​​system we interact with will be able to share a context layer that contains all the key elements of digital activity. They will immediately understand user preferences and optimize user experience more accurately. On the contrary, for example, the on-chain intellectual property registration system allows AI to reference persistent on-chain contexts, and also creates possibilities for new market-oriented interactions around prompt words and information modules - users can directly authorize or commercialize their own expertise while retaining data control. Of course, the shared context will also spawn many possibilities that we have not yet foresee.

2. General Identity System of Agent

Written by: Sam Broner

Identity (i.e. "the authoritative record of the essential attributes of something") is the underlying architecture that supports today's digital discovery, aggregation and payment systems. Since the platform encloses this architecture within the ecological wall, the identity in the eyes of users has become part of the product's functions: Amazon assigns unique identifiers (ASIN or FNSKUs) to the product, centrally displays the product and assists users in completing discovery and payment; the same is true for Facebook, which is the core basis of its information flow and the entire in-app discovery function, including product lists, native posts and paid advertisements.

With the evolution of AI agents, this pattern is about to change. When more enterprises adopt agents (used in customer service, logistics management, payment processing and other scenarios), their platforms will no longer be limited to a single interface application, but will span multiple platform ecology, accumulate deep context, and perform more diversified tasks for users. But if the proxy identity is bound to a single market, it will lose availability in other key scenarios such as email threads, Slack channels, and other products.

Therefore, the agent needs a single, portable "digital passport". Without this passport, it will be impossible to determine how to pay the agent, verify its version information, query its functional attributes, know its service objects, or trace its reputation records between different applications and platforms. Agent identities need to be equipped with multiple functions such as wallet, API registry, update log and social proof to ensure that any interface (email, Slack, or other agent) can be parsed and interacted with unified standards. The lack of shared information of "identity", every time the system integrates, the underlying architecture needs to be built from scratch, the discovery mechanism will always be in a temporary state, and the user will lose context information every time the channel or platform is switched.

We have the opportunity to design agency infrastructure based on basic principles. So, how to build a trusted neutral identity layer that is more complete than DNS records? Agents should not repeat the mistakes of "identity and discovery, aggregation, and payment functions" in a single platform, but should be able to accept payment and display functions in multiple ecosystems without worrying about being locked on a specific platform. This is the value of Crypto's cross-border integration with AI - the permissionless combination provided by blockchain networks can help developers build more practical agents and better user experiences.

Overall, vertical integration solutions (such as Facebook or Amazon) currently have a better user experience. One of the inherent challenges of creating a great product is ensuring that components operate in concert from top to bottom. But this convenience is expensive, especially when the cost of building agent aggregation, marketing, commercialization and distribution software continues to decline, and the coverage of agent applications continues to expand. Although it still needs to work hard to achieve the user experience level of vertically integrated platforms, building a trusted and neutral identity layer for agents will enable entrepreneurs to independently control the "digital passport" and promote innovative exploration at the distribution and design levels.

3. Forward-compatible "human identity" proof mechanism

Written by: Jay Drain Jr., Scott Duke Kominers

As AI technology penetrates into various network interaction scenarios (including deep fakes and social media manipulation), it becomes increasingly difficult to determine whether to interact with real humans online. This collapse of trust system has already happened - from comments from platform X (formerly Twitter) to robot accounts in dating applications, the boundaries between virtual and reality are gradually blurring. In this context, "human identity" proves to become the core infrastructure of the digital ecosystem.

One way to prove that one is human is by using digital IDs (including the centralized IDs used by TSA). Digital IDs cover all elements that can be used for identity verification, such as username, PIN, password, third-party proof (such as citizenship or credit rating). Decentralization shows significant value here: when data is stored in a centralized system, the issuer may revoke access, charge additional fees, or implement monitoring; while the decentralized model reverses this logic - users (rather than platform managers) control their own identity, making them safer and censorship resistant.

Unlike traditional identity systems, decentralized "human identity" proof mechanisms (such as World's Proof of Human) allow users to independently manage identity information and verify "human attributes" in a privacy-protected and trustworthy and neutral way. Just as a driver's license can be used in any region (whenever and where it is issued), the decentralized "human identity" proof can be used as a common underlying protocol across platforms, even for emerging platforms that have not yet been born. In other words, the blockchain-based "human identity" proves to be forward-looking compatibility because it has the following advantages:

• Portability: The relevant protocols are public standards and can be integrated on any platform. The decentralized "human identity" proof can be managed through public infrastructure, controlled by users independently, and is fully portable, and any platform can be compatible now or in the future;
• No permission access: The platform can choose to recognize the "human identity" ID independently without passing an API that may discriminate against different use cases.

The challenge facing this field is to implement applications: although there has not yet been a "human identity" proof application scenario with actual scale, we expect that when the number of users reaches a critical scale, early partnerships form, and killer applications appear, the adoption rate will accelerate. 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, forming a positive cycle (and because the on-chain ID is designed to be interoperable, network effects can accumulate quickly).

We have seen mainstream consumer applications such as gaming, dating and social media announce partnerships with World ID to help users confirm that they are playing games, chatting and trading with real humans (rather than programs). This year, new identity protocols such as Solana Attestation Service (SAS) emerged. Although SAS is not a proof of human identity issuer, it allows users to privately associate off-chain data (such as compliance KYC or investment qualifications) with Solana wallets to help build a decentralized identity system. All these signs suggest that the turning point of decentralized "human identity" proof may not be far away.

The significance of "human identity" proof is not only to block robots, but also to define a clear boundary between AI agents and human networks. It enables users and applications to distinguish human interaction behavior from machines, thus creating a better, safer and authentic digital experience.

Decentralized infrastructure

4. Decentralized infrastructure (DePIN) in the AI ​​field

Written by: Guy Wuollet

Although AI is a digital service, its development is increasingly subject to physical infrastructure. Decentralized Infrastructure Network (DePIN) provides a new model for building and operating real-world systems, helping to democratize the computing infrastructure behind AI innovations, making it more economical, resilient and censor-resistant.

How to achieve this? The two core challenges facing the development of AI are computing power supply and chip acquisition. Decentralized computing power networks can provide more computing power, and developers are also using DePIN to aggregate idle chip resources from gaming PCs, data centers, etc. These computing devices can form a licenseless computing market, 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 (due to the use of idle computing resources) can significantly reduce costs while providing censorship resistance to ensure that developers are not terminated by hyperscale cloud service providers (such as centralized cloud service giants).

The problem of a few companies monopolizing AI models has long existed, and decentralized networks help build a more economical, censor-resistant and scalable AI ecosystem.

5. Infrastructure and rules framework for interaction between AI agents,

terminal service providers and users

Written by: Scott Duke Kominers

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

For example, an AI agent may need to request specific data related to a computing task, or recruit professional AI agents to complete specific tasks (such as assigning statistical robots to develop and run model simulations, or calling an image generation robot when creating marketing materials). AI agents will also create important value in completing the entire transaction process or other activities on behalf of users, such as finding and booking air tickets based on user preferences, or discovering and ordering new books of their favorite types.

Today, there is no mature general-purpose inter-agent interaction market, and such cross-system queries can be mostly implemented through explicit API connections or in an AI ecosystem that uses inter-agent calls as internal functions.

In general, most AI agents today operate in siloed ecosystems, with relatively closed API interfaces and lacking architectural standardization. But blockchain technology can help protocols establish open standards, which is crucial for short-term applications; in the long run, it also supports forward-looking compatibility - with the evolution and birth of new AI agents, they can access the same underlying network. Blockchain can adapt more flexibly to the innovative needs of the AI ​​field due to its interoperable, open source, decentralized and often more upgradeable architectural features.

With the development of the market, many companies have begun to build blockchain infrastructure for inter-agent interaction: for example, Halliday recently launched relevant protocols to provide standardized cross-chain architectures for AI workflows and interactions, and provide protection mechanisms at the protocol layer to ensure that AI behavior does not exceed user intentions; Catena, Skyfire and Nevermind use blockchain technology to support automatic payments between AI agents without human intervention. More such systems are under development, and Coinbase has even begun providing infrastructure support for these explorations.

6. Ensure synchronization of AI/custom programming applications

Written by: Sam Broner, Scott Duke Kominers

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

However, while AI-assisted coding creates new opportunities, it also introduces a lot of uncertainty inside and outside the program. "Vibe coding" abstracts the complex dependency network of software's underlying layer, but it also makes programs susceptible to functional and security vulnerabilities when source libraries and other inputs change. Additionally, interacting with other systems becomes more difficult when people use AI to create personalized applications and workflows—in fact, even two "custom programming" programs with the same functions can have significant differences in their operational logic and output structure.

Historically, standardization to ensure software consistency and compatibility was first undertaken by file formats and operating systems, and in recent years, it has relied on sharing software and API integration. But in a new era of real-time evolution, iteration and branching of software, the standardization layer needs to have broad accessibility and continuous upgradeability while maintaining user trust. Additionally, AI alone cannot solve the problem of “inspiring people to build and maintain these connections”.

Blockchain technology solves these two problems at the same time: the protocolized synchronization layer can be embedded in the user's customized software architecture and updated dynamically to ensure cross-system compatibility as the environment changes. Historically, large companies may have paid millions of dollars to “system integrators” such as Deloitte to customize Salesforce instances. Today, engineers can create custom interfaces to view sales information over the weekends, but as the number of customized software grows, developers need professional support to keep these applications running in sync. (Note: Salesforce is a customer relationship management CRM software service provider founded in the United States in March 1999)

This is similar to today's open source software library development model, but has continuous updates (rather than regular releases) and incentives - both of which are easier to implement due to Crypto technology. Like other blockchain-based protocols, the shared ownership mechanism at the synchronization layer encourages parties to actively invest in improvements: developers, users (and their AI agents) and other consumers can be rewarded for introducing, using and optimizing new features and integrations.

On the contrary, shared ownership enables all users to be closely bound to the overall protocol, forming a buffer mechanism to resist malicious behavior -just as Microsoft will not easily destroy the .docx file standard (because it will have a chain impact on users and brands), co-owners of the synchronization layer will not tend to introduce inefficient or malicious code into the protocol.

Like all software standardized architectures we have seen, there is a huge potential for network effects here. With the continuous evolution of the "Cambrian explosion" of AI encoding software, heterogeneous system networks that need to maintain communication will expand exponentially. In short: "custom programming" requires not only "coding style", but also Crypto technology to maintain system synchronization.

New economy and incentive model

7. Micro payment system that supports income sharing

Written by: Liz Harkavy

AI agents and tools such as ChatGPT, Claude, and Copilot provide new ways to navigate the digital world, but for better or worse, they are shaking the economic foundations of the open Internet. We have seen a concrete manifestation of this trend – for example, education platforms have seen a sharp drop in traffic due to students’ extensive use of AI tools, and several US newspapers are suing OpenAI for alleged copyright infringement. Without re-adjusting the incentive mechanism, we may face an increasingly closed Internet: the number of payment barriers increases and the number of content creators decreases.

Of course, policy solutions always exist, but as they advance through judicial procedures, a range of technical solutions are emerging. Perhaps the most promising (and technically challenging) solution is to embed revenue sharing mechanisms in network architectures: when AI-driven behavior facilitates transactions, the source of content that provides information support for the decision should receive the corresponding share. The affiliate marketing ecosystem is already conducting similar attribution tracking and revenue sharing, while more advanced versions can automatically track and reward all contributors in the information chain—blockchain technology can obviously play a key role in tracing this traceability chain.

But such systems also require new infrastructure with other capabilities—especially micropayment systems that can handle cross-multi-source microtransactions, attribution protocols that fairly evaluate the value of different contributions, and governance models that ensure transparency and equity. Many existing blockchain tools (such as Rollups and Layer2, AI native financial institution Catena Labs, and financial infrastructure protocol 0xSplits) have shown application potential, supporting nearly zero-cost transactions and finer-grained payment splits.

Blockchain can implement complex proxy payment systems through a variety of mechanisms:

• Nanopayments can be split to multiple data providers, allowing a single user interaction to trigger micropayments to all sources of contribution through automatic smart contracts.
• Smart contracts support triggering executable retroactive payments upon completion of the transaction, reimbursing the source of information for purchasing decisions in a fully transparent and traceable manner.
• In addition, blockchain supports complex and programmable payment split distribution, ensuring that revenue is distributed fairly through code mandatory rules rather than relying on centralized decisions, creating trustless financial relationships among autonomous agents.

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

8. Blockchain as intellectual property and traceability registration

system

Written by: Scott Duke Kominers

The development of generative AI has created an urgent need for efficient programmable intellectual property registration and tracking mechanisms—both to clarify ownership and support business models around intellectual property access, sharing and re-creation. The existing intellectual property protection framework relies on expensive intermediaries and post-rights protection measures, and cannot adapt to the era needs of AI instantly consuming content and generating new variants with one click.

What we need is an open public registration system that provides clear proof of ownership so that intellectual property creators can participate efficiently in interactions, and that AI and other network applications can be directly connected. Blockchain technology is ideal: intellectual property registration can be completed without intermediaries, providing tamper-proof traceability, and allowing third-party applications to easily identify, authorize and use the intellectual property.

Some people are skeptical about the idea that technology can protect intellectual property rights—after all, the first two eras of the Internet (and the continuous evolution of AI revolution) are often associated with weakening intellectual property protection. Part of this is that many intellectual property-based business models today focus on “prohibited derivative works” rather than incentivizing and commercializing derivative creations. But programmable intellectual property infrastructure not only enables creators, brands and IP owners to explicitly establish ownership in the digital space, but also opens the door to a “business model around IP sharing (for generative AI and other digital applications)”—which actually translates the main threat of generative AI to creation into opportunities.

We have seen creators try new models in the early stages of the NFT field: enterprises use NFT assets on Ethereum to support network effects and value accumulation under the CC0 brand building; recently, infrastructure providers are also building protocols for standardized and composable IP registration and authorization (such as Story Protocol) and even dedicated blockchains. Some artists have begun to use these tools, authorizing their styles and works for creative recreation through agreements such as Alias, Neura, Titles, etc. Incention's Emergence series invites fans to create the science fiction universe and its characters, and the blockchain registry built by Story Protocol can track the creators of each element.

9. The web crawler mechanism to compensate content creators

Written by: Carra Wu

Today, the AI ​​agents that best meet market demand are not programming or entertainment tools, but web crawlers - they browse the web, collect data and decide to crawl the source.

It is estimated that nearly half of the current network traffic originates from non-human subjects. Crawlers often ignore the robots.txt protocol (the file should have informed the automatic crawler whether to allow access to the website, but the actual binding force is weak) and use the crawled data to strengthen the market barriers of tech giants. What's worse is that the website needs to pay for these uninvited customers and bear the cost of providing bandwidth and CPU resources to a massive number of unidentified crawlers. In response, CDNs such as Cloudflare (content distribution networks) provide blocking services, but this is just a pieced together solution that shouldn't have existed.

We have pointed out that the Internet's native protocol (the economic agreement between content creators and distribution platforms) may collapse, and data is confirming this trend. In the past 12 months, website owners blocked AI crawlers on a large scale: In July 2024, only 9% of the world's top 10,000 websites were banned. Now this proportion has reached 37%, and this number will continue to rise as more website operators upgrade their technology and user dissatisfaction intensifies.

If you do not rely on CDN to completely block accesses that are suspected to be crawlers, can you find a compromise? AI crawlers should not use the system designed for human traffic for free, but should pay for the right to capture data. This is where blockchain comes in: each web crawler agent can hold Crypto and negotiate on-chain with the "access agent" or paywall protocol of each website through the x402 protocol (of course, the challenge is that the robots.txt protocol has deeply embedded the Internet business logic since the 1990s, and requires large-scale collaboration or CDN participation such as Cloudflare to break through).

Meanwhile, humans can access content for free through World ID proving identity (see Chapter 3). In this mode, content creators and website owners can receive compensation when collecting AI data sets, while human users can still enjoy the "freedom of information" Internet.

Future ownership of AI

10. Personalized advertising for privacy protection

Written by: Matt Gleason

AI has begun to influence online shopping experience, but what if the ads you see everyday are "really useful"? The reason why people dislike advertising is obvious: invalid advertising is purely noise, and excessively accurate AI advertising based on massive consumption data seems to invade privacy. Other applications make profits by setting limits on content (such as streaming services or game levels) by "no skip ads".

Crypto provides the possibility to refactor the advertising model. Personalized AI agents combined with blockchain can find a balance between "unrelated advertising" and "over-precision advertising" and serve ads based on user-defined preferences. More importantly, this model does not need to expose the user's global data, but can directly compensate users who actively share data or interact with advertisements.

To achieve this goal, the following technical requirements must be met:

• Low-rate digital payment: In order to compensate users' advertising interactions (watch, click, and conversion), enterprises need to send small payments at high frequency, which requires the system to have high-speed processing capabilities and nearly zero handling fees;
• Privacy protection data verification: AI agents need to prove that the user meets specific population attributes, and zero-knowledge proof can complete attribute verification under the premise of protecting privacy;
• Incentive model: If the Internet adopts a profit model based on micropayment (such as a single interaction is less than US$0.05, see Chapter 7), users can choose to "watch ads for small compensation" independently, and transform the existing "exploitation" model into a "participation" model.

For decades, online (and even offline for hundreds of years) advertising has been pursuing "relevance". Reconstructing advertisements from the perspective of Crypto and AI will eventually make them more practical - customization without intrusion, benefiting all parties: for developers and advertisers, unlock more sustainable and incentivized business models; for users, gain more paths to explore the digital world.

This will not only increase the value of advertising space, but will also subvert the deeply rooted "exploitative" advertising economy today and build a more people-oriented system - users are no longer traded goods, but are the subjects involved.

11. AI companion owned and controlled by humans

Written by: Guy Wuollet

Today, people have more time on devices than offline interactions and are increasingly used to interact with AI models and AI-generated content. These models have begun to provide companionship value, whether it is entertainment, information acquisition, satisfying niche interests or educating children. It is not difficult to imagine that in the near future, AI partners for education, medical care, legal consultation and emotional companionship will become the mainstream interaction method.

Future AI partners will have unlimited patience and be customized to individual needs – they are no longer just tools or robot servants, but may become highly valued relationships. Therefore, the question of "who owns and controls these relationships" is crucial (whether it is the user, or the intermediary such as the enterprise). If you have been worried about social media content screening and censorship over the past decade, the issue will become more complex and private in the future.

The view that “a censorship-resistant blockchain hosting platform is the most feasible path to implement user control of AI” has been discussed many times (as mentioned above). In theory, individuals can run device-side models or purchase GPUs themselves, but most people are either unable to bear it or lack technical capabilities.

Although it still takes time to popularize AI partners, related technologies are being iterated rapidly: anthropomorphic text interaction partners are quite mature, visual avatar technology has made significant progress, and blockchain performance has been continuously improved. To ensure that censorship-resistant partners are easy to use, we need to rely on a better user experience to enable encryption-driven applications. Thankfully, wallets like Phantom have greatly simplified blockchain interactions, with embedded wallets, password keys and account abstraction technologies allowing users to hold self-hosted wallets without memorizing mnemonics. High-throughput trusted computing technologies based on Optimistic and ZK coprocessors will also help build deep and lasting relationships with digital companions.

In the near future, the focus of discussion will shift from “when will a realistic digital companion appear” to “who can control them”.