a16z: 5 ways the block chain helps AI proxy infrastructure

Artificial intelligence agents move rapidly from “co-pilot” to economic players, even faster than the surrounding infrastructure。

While agents can now perform tasks and transactions, they lack standardized methods to prove their identity, authority and cross-environment compensation. Identity information could not be shared across the platform, payment modalities had not achieved the default programmability and coordination was being done independently。

Block chains solve this problem at the infrastructure level. The public ledgers provide receipts for each transaction, which can be audited by anyone. The wallet provides the user with portable identity information. The stabilization currency provides an alternative settlement. These are not remote future technologies. They are now operational and can help users to operate as real economic agents without permission。

 

1. Non-human identity

The current bottleneck in the proxy economy is no longer intelligence, but identity。

IN THE FINANCIAL SERVICES SECTOR ALONE, THE NUMBER OF NON-HUMAN IDENTITIES (AUTOMATED TRADING SYSTEMS, RISK ENGINES, FRAUD MODELS) IS ALREADY ABOUT 100 TIMES GREATER THAN THAT OF HUMAN EMPLOYEES. WITH THE LARGE-SCALE DEPLOYMENT OF MODERN PROXY FRAMEWORKS (LLM USING TOOLS, AUTONOMOUS WORKFLOWS, MULTI-AGENT CONFIGURATIONS), THIS PROPORTION WILL CONTINUE TO RISE IN ALL SECTORS。

However, the agents still do not have bank accounts. They can interact with the financial system, but the mode of interaction lacks portability, probabilities and is not tacitly credible. They lack standardized means of proof of authority, cannot operate independently across platforms and cannot be held accountable for their actions。

CurrentWhat is missing is a generic layer of identity — a proxy-equivalent SSL protocol that allows for standardized coordination between platforms. Despite significant attempts, the methodology is fragmented: on the one hand, vertically integrated, first in French currency; on the other hand, encrypted originals, open standards (e.g., x 402 and the emerging proxy identity proposal); and on the other hand, the development of frameworks such as the MCP (model context protocol), which attempts to connect with the application bridge。

There is still no widely used, interoperable way for one agent to prove to another who it represents, what it is allowed to do and how it is remunerated。This is itKYAThe core concept。

JUST AS HUMANS RELY ON CREDIT HISTORY AND KYC (KNOW YOUR CUSTOMER), AGENTS ALSO NEED AN ENCRYPTED SIGNED CERTIFICATE THAT BINDS AGENTS TO THEIR CLIENTS, POWERS, CONSTRAINTS AND REPUTATIONS. THE BLOCK CHAIN PROVIDES A NEUTRAL COORDINATION LAYER FOR ALL OF THIS: PORTABLE IDENTITY, PROGRAMMABLE WALLETS, AND VERIFIABLE CERTIFICATES THAT CAN BE ANALYSED IN CHAT APPLICATIONS, API AND MARKETS。

WE HAVE SEEN THE EARLY EMERGENCE OF A CHAIN-BASED PROXY REGISTRATION FORM, AN ORIGINAL AGENT WHO USES USDC'S WALLET, AN ERC STANDARD FOR "TRUST MINIMIZATION AGENTS" AND A DEVELOPER'S TOOLKIT THAT COMBINES IDENTITY WITH EMBEDDED PAYMENT AND FRAUD CONTROL。

However, pending the introduction of a common identification standard, vendors would still block agents at the firewall。

 

2. Governance of artificially intelligent operating systems

The agent began to operate the real system, which raised some new problems。

The question is who really controls everything. Consider that, in a community or company, artificial intelligence systems are responsible for coordinating critical resources, whether in the allocation of funds or supply chain management. Even if people vote to decide on policy changes, this power is weak if the bottom artificial intelligence layer is controlled by a single supplier who can send models to update, adjust constraints or override decisions. Formal layers of governance may be decentralised, but operational layers are still centralized; those who control models ultimately control results。

When intelligent bodies assume governance roles, they introduce a new layer of dependence. In theory, this can make direct democracy easier to implement: everyone can have a representative of artificial intelligence, who is responsible for understanding complex proposals, weighing their merits and voting according to their stated preferences。

But such a vision can only be achieved if these intelligent bodies are truly accountable to the people they represent, can be universal among different service providers and are technically limited and can only follow human instructions. Otherwise, the system that eventually gets is seemingly democratic, but it is actually driven by non-transparent model behaviour, which is virtually uncontrollable。

If the current reality is that intelligent bodies are constructed from a small number of base models, then we need ways to prove that the behaviour of intelligent bodies is in the interest of users rather than model companies. This may require multiple layers of encryption assurance: (1) which training data, fine-tuning or enhanced learning processes are derived from model examples; (2) precise tips and instructions to control a particular intelligence body; (3) a record of the actual behaviour of the intelligence body in the real world; and (4) reliable assurance that, once deployed, the provider cannot change the command or retrain the intelligence body so that it cannot operate without the knowledge of the user. In the absence of these guarantees, the governance of the intelligent eventually leads to the governance of the party controlling the weight of the model。

This is where encrypted money works. If collective decision-making is documented and automatically implemented, artificial intelligence systems can be required to implement validated results. If the agent has an encrypted identity and a transparent implementation log, one can check whether his agent has complied with the rules. Moreover, if the artificial intelligence layer is user-owned and portable rather than locked on a single platform, no company can change the rules through model updating。

Ultimately, the governance of artificial intelligence systems is actually an infrastructure challenge, not a policy challenge. True authority depends on building enforceable safeguards in the system itself。

 

3. Filling gaps in traditional payment systems in artificially intelligent primary enterprises

AI agents started buying things — web-page grabs, browser sessions, image generation — and the stabilization currency is becoming an alternative settlement for these transactions. At the same time, a new, agency-oriented market is taking shape. For example, the Tripe and Tempo MPP markets brought together over 60 services designed for AI agents. In the first week on the line, it processed more than 34,000 transactions at a lower fee of as much as US$ 0.003, one of the default payment methods。

the difference is in the way these services are visited. no closing page. proxy reads schema, sends requests, pays and receives output in an exchange. they represent a new “headless” business: there is only one server, one set of end points and the price of each call. there's no front end - either shop or sales team。

The payment track to achieve this is on line. The Coinbase x 402 and MPPs use different methods, but will be paid for directly embedded in HTTP requests. Visa is also expanding the card track in a similar direction by providing a CLI tool that allows developers to spend from the terminal and businesses to receive stable coins at the back end。

Data are still at an early stage. After filtering off non-organic activities such as washing trades, x 402 processed approximately $1.6 million per month in proxy-driven payments, well below the $24 million recently reported by Bloomberg (citing 402.org data). But the surrounding infrastructure is rapidly expanding: Stripe, Cloudflare, Vercel and Google have integrated x402 into their platforms。

There are enormous opportunities in the area of developers ' tools, and the rise of Vibe Codeing has expanded the group of software developers and the potential market for developers ' tools. Companies like Merit Systems are working to build future-oriented solutions, and they have launched AgentCash, a CLI wallet and market platform that connects MPP and x402 protocols. These products allow agents to purchase the required data, tools and functions using stable currency in a single account. For example, sales team agents can use data from Apollo, Google Maps and Whitepages to enrich potential customer information without leaving the command line interface。

There are a number of reasons why such agents favour encrypted payments (and the emerging card-based solution). One is insurance. When it paid the processor to access the business, it took the risk for the business. A headless businessman without a website or legal entity has difficulty securing traditional processors. The second is that a stable currency can be programmed without permission on an open network: any developer can have endpoints to support payment without having to integrate the payment processor or sign a commercial agreement。

We've seen this pattern before. Each shift in the business model has led to new businesses, and the existing system has been difficult to provide for them at first. The company that built the infrastructure was not betting on the income of $1.6 million per month, but on the level of revenue that would have reached when the agent became the default buyer。

 

4. Repricing trust in the proxy economy

For 300,000 years, human awareness has been a bottleneck to progress. Today, artificial intelligence is pushing the marginal cost of implementation to zero. When scarce resources become abundant, constraints shift. When intelligence becomes cheap, what becomes expensive? Authentication。

In the smart economy, the real limit to expansion is our biological instincts, oursAuditandEvaluationMachine decision-making. The amount of ingestion by intelligent bodies has far exceeded human monitoring capabilities. As surveillance costs are high and failure takes time to manifest, the market tends to reduce investment in oversight. Human synergy is rapidly becoming a reality。

However, the deployment of unverified agents poses an additional risk. The system relentlessly optimizes “agents” indicators, while moving away from human intentions, creating a false productivity illusion that masks the huge accumulation of artificial intelligence debt. In order to safely entrust the economy to the machine, trust can no longer depend on manual auditing -Trust mustHard Encoding to StructureBy itself。

When anyone can generate content free of charge, the most important is the verifiable source — knowing where it comes from and whether it is credible. The chain of blocks, as well as the chain-based certification and decentrization of digital identification systems, have changed the economic boundaries of secure deployment. Artificial intelligence is no longer considered a black box but has clear and auditable historical records。

AS MORE AND MORE AI AGENTS START TRADING WITH EACH OTHER, CLEARING MECHANISMS AND TRACEABILITY SYSTEMS BECOME INEXTRICABLY LINKED. MONEY TRANSFER SYSTEMS — SUCH AS STABILIZATION CURRENCY AND SMART CONTRACTS — CAN ALSO CARRY ENCRYPTED RECEIPTS, RECORD WHO HAS DONE WHAT AND WHO IS RESPONSIBLE FOR PROBLEMS。

The comparative advantage of humankind is growing: from the discovery of small errors to the development of strategic directions to the assumption of responsibility in the event of problems. The enduring advantage belongs to those who are able to encrypt, insure and take responsibility for outputs when they fail。

The size of the lack of certification is a risk that will accrue over time。

 

5. Retention of user control

FOR DECADES, LAYERS OF ABSTRACTION HAVE CHANGED THE WAY USERS INTERACT WITH TECHNOLOGY. THE PROGRAMMING LANGUAGE ABSTRACTED THE MACHINE CODE. THE COMMAND LINE WAS REPLACED BY A GRAPHICAL USER INTERFACE, WHICH THEN EVOLVED TO A MOBILE APPLICATION AND APPLICATION INTERFACE (API). EACH CHANGE HIDES MORE BOTTOM-UP COMPLEXITY, WHILE KEEPING USERS IN CONTROL。

In the intelligent world, users specify results rather than actions, and the system decides how to achieve them. Smarts not only abstract the way the mission is done, but abstractTask implementerI don't know. When the user sets the initial parameters, the system goes back behind the screen and the system runs itself. The role of the user is changed from interactive to oversight; unless the user intervenes, the system is defaulted to be “opened”。

As users entrust more assignments to agents, new risks arise: vague input may lead agents to act on erroneous assumptions without the knowledge of the users; failure may not be reported, leading to a lack of clear diagnostic pathways; and a single approval may trigger a multi-step workflow that no one has anticipated。

This is where encryption works。The core of encryption technology has always been to minimize blind trust。As users give more and more decision-making power to software, smart-body systems make the issue more prominent and increase our demanding rigour in the design of systems — we need clearer boundaries, greater transparency and stronger assurance of the functioning of these systems。

To meet this challenge, a new generation of encrypted raw tools emerged. For example, Delegation Toolkit in MetaMask, AgentKit and proxy wallets in Coinbase and AgentCash in Merit Systems allow users to define the actions that agents can perform and those that cannot be performed at the level of smart contracts. Instead, an intent-based architecture such as NEAR Intens (which, since the fourth quarter of 2024, has over $15 billion in cumulative transactions at the Decentro-Exchange (DEX)) allows users to set the desired results — for example, “bridge-to-coin and pledge” — without specifying the specific means of realization。

***

Artificial intelligence makes scale cheap, but it is difficult to build trust. Encrypted money can rebuild trust on a large scale。