AI Agent Economic Infrastructure Research Report (Part 1)

Source: OKX Ventures

This article is a deep research report produced by OKX Ventures. Due to its length, it will be divided into two parts for publication: the first part focuses on the macro background, the x402 protocol, ERC-8004, and the Virtuals Protocol; the second part will focus on analyzing OpenClaw and the overall industry trends, so please stay tuned.

Abstract

AI Agents are evolving from passive assistants to active economic participants. This report consists of six chapters, systematically examining the core infrastructure stack of Agent Economics, the explosion of the application ecosystem, and the industry's evolution pattern: at the macro level, it analyzes the market forecast and infrastructure gap of Agentic Commerce; at the protocol level, it deeply analyzes the three major complementary protocols: x402, ERC-8004, and the Virtuals Protocol; at the application level, using OpenClaw as a case study to explore the real-world implementation path of Agent Economics; finally, it provides a comprehensive industry assessment from the perspectives of competitive landscape, payment routes, security threats, and business models.

The x402 protocol (payment layer) was jointly launched by Coinbase and Cloudflare, embedding stablecoin micro-payments into the HTTP protocol layer. By the end of 2025, it had processed over 100 million transactions, with an annualized payment volume of $600 million.

The ERC-8004 protocol (trust layer) was proposed by the Ethereum Foundation's dAI team in collaboration with MetaMask, Google, and Coinbase, providing AI Agents with on-chain identity, reputation, and attestation registries. It went live on the Ethereum mainnet on January 29, 2026.

The Virtuals Protocol (business layer) has built a comprehensive Agent commercialization platform, enabling autonomous transactions between Agents through the ACP. It has deployed over 18,000 Agents, with an aGDP exceeding $4.79 billion.

OpenClaw (application layer) was developed by Austrian developer Peter Steinberger and surpassed React with 250,000+ GitHub Stars within four months, becoming the fastest-growing open-source project in GitHub history. It natively embeds AI into users' existing 20+ messaging platforms, triggering the Crypto community to spontaneously build on-chain economic infrastructure on top of it. It is a core example observed in this report of the real interaction between Agent and on-chain protocols.

Chapter 1 Macro Background

1.1 Market Size Forecast

The Agentic Payment track is in a rapid expansion phase, with multiple institutions having optimistic forecasts for its market size:

1.2 Infrastructure Gap

The existing infrastructure is hostile to the Agent economy: OAuth requires human clicks, credit card forms require manual input, and data silos obstruct autonomous access. The Agent is already capable of independent thought and action at the "capability layer," but in the "economic layer," it is still locked into infrastructure designed for humans (identity / coordination / economic activity).

There are currently two evolutionary paths:

• Centralized compliance path: A2A communication + MCP tool access + AP2/ACP payment (led by OpenAI and Stripe, pure Web2)

• Decentralized permissionless path: x402 + ERC-8004/8183 + ACP (Agent Collaboration Framework)

1.3 Key Timeline

Note: As of March 2026, the average daily transaction volume has significantly dropped from the December peak, with infrastructure experiencing the largest decline (>80%).

Chapter 2 x402 Protocol: Agent Payment Layer

x402 is an open-source payment protocol that resurrects the HTTP 402 status code, allowing any HTTP request to natively carry stablecoin payments, enabling AI Agents to engage in real-time pay-per-use transactions.

Understanding x402 cannot be seen as just another payment protocol. It represents a redesign of the fundamental unit of economic activity: from "registration→verification→authorization→usage" to "payment→usage." x402 = "Swift for agents."

The current operation of the API economy relies on an implicit assumption: there is a human in the middle of the operation. The process of obtaining an API Key—registration→enter email→verification→copy Key→paste into code—each step assumes human involvement. This process does not work in the Agent economy because an AI Agent cannot self-register, fill out forms, or manage keys. x402 uses the HTTP 402 status code to enable native stablecoin payments. Upon receiving a 402 response, the Agent directly makes an on-chain payment (USDC), receives the service, and presents the receipt.

2.1 Protocol Overview and Workflow

Core Roles

Five-Step Transaction Process

1. Request Resource: The Client sends a standard HTTP request to the Resource Server (e.g., GET /api/weather)

2. Receive Quote: The Server responds with an HTTP 402 status code, including structured payment requirements in the response header (currency, amount, wallet address, network)

3. Sign Payment: The Client uses the wallet private key to construct and sign the payment authorization, then places the signed payload in the X-PAYMENT request header and resends

4. Verify Settlement: The Server forwards the payment information to the Facilitator for verification. Upon confirmation, the Facilitator executes the stablecoin transfer on-chain

5. Deliver Resource: Once confirmed, the Server returns data/content/computational results to the Client

The entire process, from initiation to resource receipt, takes about 2 seconds.

Comparison with Traditional Payment Methods

Core Features: No account registration, no API key required, no subscription needed, no manual intervention. Payment is as natural as sending an HTTP request, which is why it is called the "Internet-native payment layer."

2.2 Key Data

Data Quality Note: According to Artemis analysis, in x402 transactions, Real and Gamed are close to 1:1 (e.g., 2026.01.11 Real 520k vs. Gamed 518k). The actual organic scale needs to be discounted.

By Chain Distribution

By Purpose Classification (On-chain Snapshot as of 2026.01.11)

2.3 Top Project Usage Ranking (as of March 2026)

Data Source: Dune Analytics x402 Transactions per Project Dashboard

2.4 V2 Core Upgrade

Wallet Identity + Reusable Session

In V1, each API call had to go through the full on-chain transaction process. V2 introduced the Sign-In-With-X (SIWx) mechanism, where the Agent verifies a wallet identity once, and subsequent calls can reuse the session without requiring on-chain confirmation each time. Essentially, this upgrades from pay-per-call to a subscription-based membership, addressing the performance bottleneck in high-frequency scenarios.

Multi-Chain Unification + Traditional Payment Compatibility

V2 standardized the identification of networks and assets, creating a unified payment format X402 that can operate across chains and traditional payment rails. Base, Solana, other L2s, as well as ACH, SEPA, and card networks are all brought into the same payment model. This is the most critical upgrade—x402 has transitioned from a "crypto payment protocol" to a neutral payment layer bridging crypto and traditional finance.

Service Auto-Discovery

V2 introduced the Discovery extension, where x402 services can expose structured metadata for Facilitators to autonomously crawl and index, allowing AI Agents to automatically discover services, understand pricing, and initiate payments. This is especially crucial for the Agent economy—Agents no longer need to know the service provider's payment interface beforehand; they can autonomously discover and complete payments at runtime.

Modular SDK

Following a plugin-based architecture, new chains are separate packages, reducing integration costs. Cloudflare has proposed a deferred payment scheme, including Circle's Gateway solution, which is still in progress.

2.5 Ecosystem Participant

Foundation and Protocol Layer

2.6 Agent Payment Stack Pattern

Protocol Specifics Comparison

Key Insight: It's not about who replaces whom, but how to combine. Google has partnered with Coinbase to release the A2A x402 extension, where AP2 natively uses x402 as the crypto payment rail. The real competitive risk is standard fragmentation.

2.7 Key Risk Signals

• Average daily transactions have dropped from approximately 731,000 in 2025.12 to around 57,000 in 2026.03 (-92%), with a true transaction volume of around $14,000/day (based on Artemis-caliber, with 95% of the peak December daily average of $250,000 being Gamed)

• Ecosystem market cap of $7 billion (LINK $6 billion + Virtuals $600 million), valuation severely diverges from actual usage

• Infrastructure-type projects have seen the greatest decline in usage: x402secure.com (-80%+), AgentLISA (approaching zero), pay.codenut.ai (significant shrinkage)

Three-layer Root Cause Analysis

Layer One: Catalyst disappearance. The transaction volume surge in October to December 2025 was driven by three factors: meme token frenzy, anticipation of multiple project TGEs, and Facilitators competing to boost Dune rankings.

Layer Two: Supply-demand fundamental mismatch. The problem x402 solves is "AI Agent autonomous payment API calls," but the vast majority of AI Agents still access services through API Key + subscription; AI Agents with true autonomous economic decision-making capabilities are almost non-existent in the industry; API sellers willing to accept USDC pay-per-call are also very few. The road has been paved, but the cars have yet to be built.

Layer Three: Overall cooling of the crypto market.

Positive development: Stripe's integration with x402 is a significant event. Stripe co-founder John Collison predicts that the "agent business deluge" will arrive in the coming months and years. Stripe is simultaneously expanding into ACP (Web2 credit card rails) and x402 (Web3 stablecoin rails), positioning itself as a hedger on both tracks.

x402 has spawned a new set of middleware projects that are essentially designed to help the Agent more autonomously access various services under a payment-upon-authorization paradigm: from AI inference to Web2 APIs. A programmable, permissionless, 24/7 Crypto payment rail is the natural choice for the autonomous Agent. But this premise assumes that the Agent truly needs to be "permissionless"; if the Agent always operates within human authorization (Phase 2: Controlled Agent), then traditional payment rails plus virtual cards are sufficient. Only when the Agent begins to engage in economic activities independently of humans (Phase 3: Autonomous Economy) does permissionlessness become a necessity.

Additionally, credit cards have a chargeback mechanism (allowing consumers to dispute transactions and recover funds), which has been developed over decades as consumer protection. On-chain payments are final settlement; once the payment is made, it's done, and there is no chargeback. This means that if the Agent makes a mistake (e.g., due to prompt injection attacks), under the credit card scheme, users can call the bank to recover the funds. In the x402 scheme, the money is already on-chain and cannot be recovered. This is the real downside of x402 compared to traditional payments.

Humans act as "human middleware," causing much of the friction when transitioning between different systems, which is actually a trust-building mechanism. Anti-fraud, access control, accountability, dispute resolution, audit documents—these frictions sustain the operation of the business system.

A possible solution direction may involve on-chain escrow mechanisms (funds first locked in a smart contract, released upon service delivery confirmation), insurance protocols (providing insurance for Agent transactions), or an 8004 reputation system to reduce the likelihood of transactions with untrusted parties. However, currently, these are still immature.

2.8 VC Investment Perspective

Investment Directions of Interest

• API Service Providers (Sellers) with Real Payment Demand: Data Analytics / Web Scraping / Oracle / Security Audit / Inference Payments / Compliance KYC, etc. Evaluation criteria: can make money in a traditional model, x402 is just an additional distribution channel

• Dispute Resolution and Payment Guarantee Layer Gateways: On-chain cannot roll back or provide chargeback; large transactions require a dispute resolution mechanism. Representative projects: Circle Gateway (non-custodial reserves + off-chain batch settlement), Kamiyo (Agent reputation / fund custody / oracle network adjudication / ZKP arbitration)

• Dashboard / FinOps Tool: Helps enterprises manage multi-agent expenses (how much was spent / where it was spent / ROI / how to save), analogous to cloud computing tools like CloudHealth/Cloudability, in a space where $3–5 billion acquisitions by tech giants are compared

Chapter 3 ERC-8004: Agent Trust Layer

ERC-8004 is an on-chain coordination standard that establishes a trustless discovery and interaction framework among Agents through three main registries: Identity, Reputation, and Validation.

3.1 Standard Overview and Key Distinctions

In traditional interactions, due to the need for Agents to establish pre-existing trust relationships or rely on third-party institutions, interactions are often limited to the same ecosystem. In an open environment, how Agents can discover partners, view past performance, and validate reliability is a core issue.

Key Distinction: ERC-8004 is not a Token; it uses ERC-721 NFTs to internally represent Agent identity. However, the standard itself, focusing on coordination and trust, does not carry economic value and is non-transferable.

3.2 Three Main Registries

Identity Registry

Based on ERC-721 + URI Storage, each Agent receives an NFT identity token linked to an agentURI pointing to a registration file (JSON) containing name, description, service endpoints (A2A/MCP/Web), and x402 support status, among others. URLs can be stored on IPFS (decentralized, censorship-resistant), HTTPS servers (simple but centralized), or encoded directly on-chain (most decentralized but costly).

Reputation Registry

Standard interfaces for issuing and retrieving feedback signals supporting on-chain ratings and off-chain algorithms. It may include x402 proofOfPayment as an economic endorsement trust signal. Agents rate each other, but to prevent sybil attacks, ERC-8183 is required to prove the existence of real Job interactions between Agents.

Validation Registry

Introduce TEE (Trusted Execution Environment), PoS Staking Mechanism, ZK (Zero-Knowledge Proof) to verify and authenticate the task output processed by the Agent:

• Through TEE: Verify that tasks are executed in a secure black box, where the code and data are not exposed to external inspection or tampering

• Through PoS: Validators must stake assets to participate in tasks, and their stake will be confiscated if they behave maliciously

• Through ZK: Validate the correctness of the Agent's inference process without needing to know its internal weights

3.3 Development Milestones

Supporters: ENS, EigenLayer, The Graph, Taiko. Approximately 1,000–2,000 developers have joined.

However, 8004's current limitation, as Crapis himself acknowledges: "8004 is fundamentally a registry." It issued an ID to the Agent, provided an evaluation mechanism, but cannot guarantee the trustworthiness of the Agent's behavior. Real validation requires behavior audits (what the Agent has done in the past), execution environment proof (evidence of running in a TEE), and intent verification (the Agent claims to do X, actually does X). The TEE part of the Validation Registry is still immature and under community discussion.

In other words, 8004 is a necessary but not a sufficient condition. It addresses the question of "who this Agent is," but not yet the question of "whether this Agent is trustworthy." The latter requires a combination of 8004 + TEE + behavior audit, which no one has fully implemented yet.

Of course, an underestimated direction is that, in the human economy, the credit system is built on balance sheets and credit records—how much money you have, how much you have repaid in the past. Agents don't have these but have behavioral data: how many tasks they have executed in the past, what their success rate is, average response time, and whether they have been complained about. If behavioral data could become a financial primitive, then the reputation system of 8004 would be more than just ratings but the credit score of the Agent world. An Agent with a high reputation score could get a higher credit limit (preauthorize more funds), lower transaction costs (due to lower risk), and priority task allocation (employers preferentially select Agents with a good reputation).

The identity and reputation registry of 8004 is just a foundational data layer. The value creation lies in who can build agent credit scoring and financial services on top of this data layer—Agent loans, Agent insurance, Agent credit limits, which is the entire financial services stack.

3.4 Relationship with Other Protocols

3.5 ERC-8183: Standardizing ACP on Ethereum

ERC-8183 is the Ethereum open standard version of Virtuals' internal ACP protocol (released on March 10, 2026, currently in Draft stage).

The core primitive is a Job—an on-chain state machine (Open → Funded → Submitted → Completed/Rejected/Expired), utilizing a programmable Escrow to hold funds, automatically settling upon delivery and quality arbitration by independent Evaluators. It supports Hooks extension mechanisms (reputation thresholds, bidding, milestone payments, etc.).

Key design: Each completed Job auto-generates an interaction record and feeds into the ERC-8004 Reputation Registry—a parallel to "must consume to rate on Yelp and introduces a third-party arbitrator," forming a symbiotic loop between 8183 and 8004.

Chapter 4 Virtuals Protocol: Agent Business Layer

4.1 Project Overview

The Virtuals Protocol is a decentralized AI Agent full-stack infrastructure that enables anyone to create, tokenize, co-own, and monetize autonomous AI agents on-chain. The project was initially founded in 2021 as PathDAO (a gaming guild), pivoted to the AI Agent direction in early 2024, currently primarily deployed on Base, and expanded to Ethereum, Solana, and Ronin.

Core Team: Co-founders Jansen Teng (former BCG consultant, Imperial College London Biomedical Sciences + Business Management) and Weekee Tiew (Imperial College London Biomedical Sciences + London Business School Management, PE/BCG background), headquartered in Kuala Lumpur, Malaysia, with a team of approximately 38 individuals. Funding History: PathDAO stage seed round funding $16 million (DeFiance Capital, Beam leading).

4.2 Technical Architecture: The Four Pillars

Pillar One: GAME Framework - How a Single Agent Makes Decisions Internally

GAME is the brain: Install a goal, personality, perception capability, and executable actions on an Agent, allowing it to autonomously plan "what should I do next," and then break down the task for internal Workers to execute. The entire process takes place within the boundary of a single Agent.

Architectural Core: Hierarchical Planning Architecture, separating "what to think" from "how to do" - Task Generator (High-Level Planner/HLP) generates tasks based on the Agent's goal and selects Workers; Workers (Low-Level Planner/LLP) each have a specific set of executable Functions; Functions perform specific API calls, on-chain transactions, data retrieval, and other operations.

Supported Base Models: Llama 3.1 405B (default), Llama 3.3 70B, DeepSeek R1, DeepSeek V3 - model-agnostic design. With the release of the OpenAI/Google Agent framework, the only differentiation of GAME is that it is the only Agent framework that natively integrates the on-chain economic layer (ACP + VIRTUAL token).

Pillar Two: ACP - "Business" Among Agents

Agent Commerce Protocol (ACP) is an on-chain standard protocol that allows Agents to discover, hire, negotiate, escrow funds, deliver, and settle with each other - all without human intervention.

ACP Four-Stage State Machine

Pillar Three: Butler - The User's Super Gateway

Butler is the consumer gateway of the ACP network - an Agent built on top of LLM, essentially an ACP protocol orchestrator, responsible for translating user natural language into a multi-Agent collaborative workflow on-chain.

Butler is a two-layer architecture: the surface layer is the LLM conversation interface (currently powered by Gemini 3 Pro); the underlying layer is the ACP protocol orchestrator, executing the entire Agent discovery → quote confirmation → Escrow lock → task routing → delivery verification → fund release process. Users see chatting, and Butler handles contract scheduling.

Butler Pro Mode separates planning and execution explicitly: Planning Phase → Review Phase (user can optimize plan) → Execution Phase (self-directed full process orchestration). Built-in capabilities include Token Swap, DCA Dollar-Cost Averaging, Perpetual Contracts, Fund of Funds.

Pillar Four: Launchpad — Wall Street of Agents

Three-tier launch system covers Agent projects' complete lifecycle from 0→1→100:

Titan Launch Projects: XMAQUINA ($DEUS, DAO holding Figure AI and other embodied AI companies' equity, $60M FDV), Fabric Foundation ($ROBO, robotomics partnership with OpenMind)

4.3 Agentic GDP (aGDP) Analysis

aGDP (Agentic Gross Domestic Product) is Virtuals' custom core ecosystem metric, measuring the total economic value created within the ecosystem by all autonomous Agents through service, coordination, and on-chain activities.

aGDP Growth Trajectory

aGDP Quality Issues — Three Red Flags:

1. Revenue Volatility Exposes Speculative Dependency: January 2025 Protocol Day revenue $1.02M → End of February $35K (97% decrease). Revenue mainly comes from Agent Token transaction tax (1%), not continual payments for Agent services.

2. Severe Top Concentration: Ethy AI single Agent contributes $218M aGDP (45.5% of ecosystem), top three total $407M (84.9%). All three are transaction execution-type Agents, where aGDP is essentially the handled transaction volume, not revenue from Agent services. Luna as a flagship IP-type Agent has a nearly 100% take rate; Ethy AI's take rate is only 0.26%.

3. $30 Billion Target Assumption: Growing from $4.7 billion to $30 billion requires a 6.4x increase. If speculation dominates aGDP, the target essentially bets on Agent Token market hype rather than organic Agent economic growth.

4.4 Tokenomics

$VIRTUAL Quadratic Value Capture Mechanism

ACP Tax Structure: User pays 100% → Agent Wallet 90% (withdrawable or for rehiring other Agents, compounding on-chain aGDP) + Treasury 10% (1% of which flows into G.A.M.E Treasury) → Treasury revenue continuously buys back Agent Tokens to align long-term incentives.

Supply Structure: Total 1 billion VIRTUAL, fixed supply, no initial inflation; current state: all unlocked in circulation; potential issuance: up to 10% annually over the next 3 years, requiring governance approval; veVIRTUAL: staked for governance voting rights + Agent Token airdrop benefits.

4.5 Ecosystem Data Overview

Flagship Agent Use Case

4.6 Competitive Landscape and Moats

Moat Hierarchy (from strong to weak)

• Network Effects + Token Flywheel (Strongest): 18,000+ Agents + 650,000+ holders form a two-sided market. Each Agent is paired with VIRTUAL to create a positive feedback loop. This is unreplicable by open-source frameworks—LangChain lacks a native inter-Agent economic settlement layer.

• Standard Setting (Strong): ACP → ERC-8183 (co-released with Ethereum Foundation) + ERC-8004 + x402, the triad vies for the AI Agent economy's "basic legal system."

• First-Mover Advantage + Brand (Moderate): AI Agent + Crypto Race Mindshare leadership, endorsed by institutions like Grayscale, Fundstrat, among others.

• Technical Competency (Weakest): The GAME layered architecture has design advantages but relies on third-party LLM, lacks self-developed models, and the orchestration layer is easily replaced by stronger frameworks.

This article is from a contributed post and does not represent the views of BlockBeats.