This paper surveys the intersection of AI agents and blockchain, framing it as "Web 4.0" and organizing the design space around bidirectional trust: blockchain providing trust infrastructure for agents (identity, permissions, intent execution, token economies) and agents participating in blockchain mechanisms (security audits, consensus, governance). The authors catalog 70 Ethereum standards, 20 industry projects, and 118 papers, assessing each along five dimensions (verifiability, trust minimization, expressiveness, composability, maturity). They find that agent-specific standards are immature, intent architectures lack formal analysis, and there's no unified security framework treating AI as a first-class protocol actor.
Main takeaways:
- Organizes AI-blockchain interaction into two directions: blockchain as trust layer for agents (B→A) and agents as participants in blockchain protocols (A→B).
- Reviews 70 Ethereum standards (EIPs/ERCs), 20 projects, and 118 papers across identity, delegation, intent systems, auditing, consensus, and governance.
- Finds major gaps: agent standards are overwhelmingly immature, intent systems lack formal security models, and no unified framework for AI as protocol-layer actors.
- Verifiable computation (ZK proofs, TEEs, on-chain inference) underpins both directions but involves trade-offs between trust minimization, overhead, and readiness.
- Proposes a three-dimensional taxonomy and identifies nine open research questions.