TessPay proposes a 'verify-then-pay' architecture that only releases escrowed funds after cryptographic verification of an autonomous agent's completed work, aiming to fill foundational trust gaps in agent-driven commerce; the design links canonical agent registries, PoTE evidence, and chain-agnostic settlement rails but lacks empirical validation or deployment evidence.

Main Finding

TessPay is a unified infrastructure design that closes a foundational trust gap preventing widescale agentic commerce. By enforcing a "Verify-then-Pay" workflow and separating control/verification from settlement, TessPay replaces implicit trust with cryptographic evidence and gated escrow, enabling accountable, auditable, chain‑agnostic transactions between autonomous agents and stakeholders.

Key Points

• Trust gap in agentic commerce arises across three stages:
  • Task Delegation: no robust way to translate intent into scoped, verifiable mandates or to discover and authorize agents.
  • Payment Settlement: payments are typically made before verifiable evidence of task completion exists, creating moral hazard.
  • Audit Mechanisms: lifecycle evidence is missing or fragmented, making dispute resolution and accountability difficult.
• TessPay architecture:
  • Two planes: a control/verification plane (handles intent, agent identity, evidence) and a settlement plane (escrow and rail adapters).
  • Before execution: agents are registered in a canonical registry; user intent is captured as verifiable mandates (signed, machine-interpretable).
  • During execution: funds are locked in escrow; agents produce cryptographic Proofs of Task Execution (PoTE) using primitives like TLS notaries, Trusted Execution Environments (TEEs), signed logs, etc.
  • At settlement: PoTE is verified against predicates; funds are released only if verification passes; modular rail adapters enable chain- and rail-agnostic settlement.
  • After settlement: tamper-evident audit trails are preserved for accountability and dispute resolution.
• Design goals: reduce prepayment risk, enable automated dispute resolution, support heterogeneous payment rails, and create a canonical record of agentic transactions.
• PoTE is modular and extensible — different tasks can rely on different evidence types and verification policies.

Data & Methods

• Nature of work: architectural/protocol proposal (conceptual design). The paper specifies system components and verification primitives rather than reporting large-scale empirical datasets.
• Core components described:
  • Canonical registry: identity and reputational anchoring for agents and stakeholders.
  • Verifiable mandates: structured, signed statements encoding scope, success predicates, authorization and liability.
  • Escrow layer: conditional locking of funds until verification criteria are satisfied.
  • PoTE generation primitives: TLS notary proofs, TEE attestations, signed transcripts, cryptographic timestamps, and other verifiable logs.
  • Verification engine: policy-driven validators that run predicates over PoTE artifacts to decide settlement.
  • Rail adapters: modular connectors to on-chain smart contracts, payment processors, banking rails, and custodial services.
  • Audit store: tamper-evident ledger (could be blockchain-backed or append-only signed logs) for lifecycle preservation.
• Validation and evaluation methods suggested (or implied) for future work:
  • Security analysis of attack vectors (forgery of PoTE, registry poisoning, collusion).
  • Economic simulations to measure effects on transaction costs, dispute rates, and adoption thresholds.
  • Prototype implementations and pilot deployments across different rails to measure latency, cost, and interoperability.
  • Empirical metrics: verification false-positive/false-negative rates, escrow time, dispute frequency/resolution time, user/provider churn, and throughput across rails.
• Assumptions and dependencies:
  • Availability and trustworthiness of PoTE primitives (e.g., TEEs, TLS notary services).
  • Willingness of payment rails to integrate via adapters and honor conditional settlements.
  • Standardization of verifiable mandate formats and verification predicates.

Implications for AI Economics

• Reduced moral hazard and lower transaction costs: Replacing prepayment with PoTE-gated escrow aligns incentives, reducing fraud and opportunistic behavior by autonomous agents and lowering monitoring costs for principals.
• Market expansion for agentic services: Lowered trust barriers and interoperable settlement rails can unlock new agent-driven markets and microtransactions (e.g., automated procurement, information-as-a-service, delegated scheduling).
• New economic roles and intermediaries: Canonical registries, PoTE providers (notaries, attestation services), verification-as-a-service vendors, and rail adapters become economic actors; competition and pricing for these services will shape market structure.
• Pricing and contract design: Verifiable mandates enable finer-grained contingent contracts and outcome-based pricing tied to machine-verifiable predicates, potentially changing bargaining and insurance models.
• Network effects & standards: The value of the infrastructure grows with registry coverage and adapter availability. Standardizing mandate formats and PoTE semantics is critical to reduce coordination failure.
• Cross-border and regulatory impacts: Chain-agnostic escrow plus cryptographic audit trails simplify cross-rail settlement but raise regulatory questions about custody, KYC/AML, and legal status of machine-signed mandates.
• Labor and substitution effects: As agentic commerce becomes more trustworthy, tasks currently performed by humans may be automated; however, new jobs in verification, audit, and rail integration may emerge.
• Risks and limits:
  • Reliance on PoTE primitives introduces single-point failures if attestations are spoofed or TEE vendors are compromised.
  • Privacy trade-offs: richer audit trails improve accountability but can expose sensitive data unless designed with privacy-preserving proofs (e.g., zero-knowledge).
  • Liquidity and capital costs: escrow holding increases working capital needs; economic analysis must account for capital friction and fee structures for verifiers and adapters.
  • Governance: canonical registries and policy choices (verification predicates, dispute rules) can centralize power; careful governance and open standards are required.
• Measurable economic outcomes to track in deployments: reduction in dispute rate, change in average contract size and frequency, adoption curve of agentic transactions, per-transaction verification costs, and welfare effects across principals, agents, and intermediaries.

Overall, TessPay offers a principled, modular architecture that could materially reduce trust frictions in agentic commerce. Empirical validation (prototypes, pilots, and economic simulations) and attention to governance, privacy, and integration costs will determine practical economic impact.