On-chain proxy infrastructure stack: The track for building autonomous Web3

Disclosure: This article does not constitute investment advice. Page content and materials are for educational purposes only.

The Graph points out that on-chain AI agents need identity, structured data and payment capabilities to act autonomously.

Summary

On-chain AI agents need identity, reliable blockchain data and payment tools to achieve online autonomous operations. ERC-8004 and account abstraction can grant explicit permissions to autonomous agents while supporting verifiable on-chain reputation. Subgraphs, MCPs, and x402 can help agents access data, pay services, and perform blockchain operations.

Artificial intelligence and blockchain have basically developed along parallel tracks for many years. AI is becoming increasingly powerful in reasoning, understanding instructions, and making decisions, while blockchain networks create programmable financial systems that operate without centralized intermediaries. The next stage is to combine the two.

The result could be an Internet with not only human users, but also autonomous AI agents-who can discover information on their own, make decisions, interact with protocols, and pay for services.

To achieve this future, it is not enough to rely on increasingly powerful big language models. As The Graph Foundation explained in its August 18 blog post "Analytical Chain Agents Infrastructure Stack", autonomous agents need infrastructure that can transform AI reasoning into reliable blockchain operations. The Graph summarizes the emerging infrastructure stack into three basic requirements: identity, environmental awareness, and economic proxy capabilities.

Identity: Provide agents with on-chain passports

Before an autonomous agent can conduct a transaction, blockchain needs a reliable way to identify it and determine the operations it is authorized to perform.

Simply handing over the private key of the user's wallet to the AI agent can pose obvious problems. One wrong decision or illusion can expose all assets in that wallet. At the same time, it is difficult to distinguish whether operations are performed by humans or autonomous systems.

ERC-8004, which The Graph calls the "Untrusted Proxy Standard," solves this problem through three on-chain registries covering identity, reputation, and verification. These registries allow agents to establish recognizable identities and interact between participants without prior trust.

Combined with account abstraction, this model can also create strictly defined permissions. For example, an agent may be authorized to trade only a certain amount per day without having unlimited control over user funds.

As agents accumulate historical records, the value of identity becomes higher and higher. Because actions and feedback can be recorded on the chain, agents can accumulate a verifiable reputation that can be evaluated by other agents and smart contracts.

The Graph supports this layer through the Agent0 subgraph, which indexes the agent's registration information, metadata, reputation information, and verification activities across multiple networks. According to The Graph, this allows agents to search for other agents based on characteristics such as capabilities or reputation without having to independently scan blockchain history.

Data: Helping AI understand the world on the chain

Knowing who the agent is only solves part of the problem. Autonomous agents also need accurate information about the environment in which they operate.

Blockchain contains a large amount of transparent data, but transparency does not mean easy access. Information is distributed across blocks, transactions, events, and smart contracts. Letting large-language models process this raw data directly is inefficient and may increase the risk of wrong conclusions.

This is why blockchain indexing has become a critical part of the agency infrastructure.

The Graph's subgraph organizes blockchain information into a structured, searchable data set. When combined with the Model Context Protocol (MCP), this index information can be used directly by AI systems. The Graph describes the subgraph MCP as a "translator" between agents and complex blockchain data.

Suppose that the task of an autonomous trading agent is to find favorable opportunities involving ETH trading pairs. Before executing a transaction, agents may need to compare liquidity across agreements, assess current conditions and confirm that opportunities still exist. Instead of trying to parse millions of blockchain logs, agents can query relevant subgraphs through MCP, obtain structured information and reason.

This illustrates an important difference in proxy infrastructure: the intelligence layer and the data layer solve different problems. The large-language model can decide what information is needed and reason about answers, while the indexing infrastructure is responsible for providing reliable blockchain information in available formats.

Payment: Giving agents economic agency capabilities

The last piece of the puzzle is payment capabilities.

Today's Internet payment infrastructure is designed mainly around individuals and businesses. Users create accounts, manage subscriptions, enter payment information, or manually authorize transactions. Continuously running autonomous software cannot rely on these workflows.

The emerging x402 standard provides another model. It restores the HTTP "402 Payment Required" status code, allowing services to request payments directly in Internet requests.

The Graph has integrated x402 into its subnet gateway, allowing agents to pay for individual queries in USDC without having to maintain traditional API accounts or keys. The agent requests data, receives the payment request, signs the payment, resubmits the request, and then obtains the required information. The Graph's GraphTally infrastructure handles settlement with the indexer in the background.

Its significance is not limited to payment blockchain queries. Machine-to-machine commerce requires payment systems suitable for massive small automated transactions. If agents are constantly purchasing data, calculations, or services between themselves, on-demand micropayments can provide an economic model that is closer to the way autonomous software actually works.

From Web3 users to Web3 proxies

Combining these layers provides a clearer picture of the on-chain proxy stack.

The agent establishes identity and operates within preset permissions. It accesses structured blockchain data to understand the current situation. It can then purchase the information or services it needs and perform the authorization action. The entire process can be repeated without human approval of each intermediate step.

The Graph calls this process a "proxy loop." ERC-8004 provides identity and accountability, subgraphs and MCPs provide context awareness, and x402 and GraphTally support autonomous payments and settlements.

This architecture also points to a broader shift in the way blockchain infrastructure is designed. Most of today's Web3 assumes that users sit behind a screen, operate the interface, connect to their wallets, and approve transactions. Agent-centric environments require an infrastructure that is machine-readable, programmable and economically autonomous by default.

The Graph has provided blockchain data infrastructure for more than 60 networks and reports that as of early 2026, it had provided more than 1.27 trillion queries to more than 75,000 projects. The rise of autonomous agents may bring a new class of users to such infrastructure: the software itself.

AI may provide an inference engine for the emerging proxy Internet, but intelligence alone cannot create an autonomous economy. Agents also require identity, trustworthy data, and native transaction methods. The development of the underlying stack will determine whether on-chain AI stays in the experimental stage or becomes a machine-to-machine functional economy running on the scale of the Internet.