The Boundless Frontier of AI + Crypto: Navigating the Present and the Future

Web3CN_Pro avatar
Web3中文
@Web3CN_ProNov 27, 2025

The Spaces brought together builders at the AI–blockchain intersection to examine real synergies, verification, standards, and the road to adoption. Host Pamela guided a panel featuring Martin (Chain Aware), Aiden (Vishua), Raul Harad (DeepSafe AI), Henry Lee (Kite AI), and Marsha/Maxwell (Termix AI). Consensus formed that AI contributes intelligence (prediction, automation) while blockchain supplies integrity (provenance, identity, auditability). Concrete use cases highlighted included fraud detection, agentic commerce with crypto rails, cross-chain institutional execution via ZK-verified proofs, and decentralized verification networks. Each team outlined distinct approaches: Chain Aware’s predictive fraud/rug-pull, growth analytics, and portfolio tooling; Vishua’s trustless interoperability for institutions (keeping assets in custody and proving state cross-chain; 700M committed; 600 BTC live); DeepSafe’s decentralized attestation layer for verifiable computation and events; Kite AI’s agent identity, wallets and spend guardrails (Kite Passport) to build agent trust; and Termix’s natural-language DeFi strategy execution and an “agent–protocol” standard layer. The panel assessed X402 (micropayments/agent payments) and ERC-8004 (on-chain agent identity/reputation) as complementary. A major theme was AI trusted verification—combining data/model provenance, ZK proofs, TEEs, and attester networks—to move AI from black box to auditable. Looking ahead, the next wave hinges on verifiable infra, economic alignment, usability, identity, and safer agent architectures.

The Boundless Frontier of AI x Crypto — Roundtable Notes

Participants and Roles

  • Pamela (Host, Web3 CN Pro)
  • Martin (Co-founder, Chain Aware)
  • Aiden (Weishuo/Vishua team; building trustless AI infra for autonomous finance)
  • Raul Harad (CEO, DeepSafe AI)
  • Henry Lee (Ecosystem Product Lead, Kite AI)
  • Marsha (Co-founder, Termix AI; referred to as Maxwell by the host in closing; partner Okenzhou absent due to time zone)

Theme and Context

  • Focus: Opportunities and challenges in building the AI and blockchain ecosystem; where real synergy emerges; what infra and standards are needed (e.g., X402, ERC-8000/8004); and how to achieve trusted AI verification on-chain.

How AI and Blockchain Complement Each Other

  • Martin (Chain Aware)
    • Unorthodox view: Many “AI x blockchain” projects merely issue tokens or resell LLM outputs (OpenAI/Gemini) without truly using on-chain data or building proprietary models.
    • Real synergy requires: training and inference on blockchain-native data, building/using MCPs to expose blockchain data to agents, and shipping predictive analytics that affect real outcomes (growth and fraud detection).
  • Aiden (Weishuo/Vishua)
    • TradFi rails (e.g., SWIFT) are not designed for AI agents to execute financial actions. Blockchain plus trustless interoperability is necessary for autonomous execution at scale.
    • ZK proofs can verify asset state wherever it lives; proofs can be brought to counterparty chains so agents can act seamlessly across heterogeneous ecosystems (SVM, EVM, MoveVM).
  • Raul (DeepSafe AI)
    • AI provides intelligence (insights, predictions, automation) but lacks transparency; blockchain brings integrity (provenance, immutability, verifiability). Together: blockchain supplies traceable sources, identity, and auditable logs; AI supplies analysis, scaling, optimization.
    • Key synergy areas: trusted data pipelines for AI, verifiable AI outputs, on-chain-accountable autonomous agents, and decentralized infra optimized by AI.
  • Henry (Kite AI)
    • Internet is shifting from mobile-first to agent-first. Agents will transact by default; crypto rails (esp. stablecoins) are best suited (cross-border, programmable, low-fee, microtransactions, high frequency).
    • Agentic commerce needs programmable, low-latency, low-cost payments—hard in Web2, natural on crypto.
  • Marsha (Termix AI)
    • AI results are powerful but hard to verify; blockchain offers tamper-proof transparency.
    • Combining them enables verifiable AI (models can show how they reached conclusions without revealing sensitive data). AI, in turn, optimizes blockchain ops (liquidity, risk, governance, UX).

How AI Can Accelerate DeFi Adoption

  • Martin: Massive reduction of user acquisition costs (UAC) via predictive wallet analytics and one-to-one targeting. Move away from mass marketing to growth tech grounded in on-chain behavior prediction; improves protocol revenues and survivability.
  • Aiden: Before advanced intelligence, we need reliable, trustless execution across chains (interop solved with proofs). Only then can domain-specific AI (risk, strategy) flourish on top.
  • Henry: Agent-driven microtransactions and cross-border flows require programmable, low-cost rails—crypto unlocks agent-native commerce models (e.g., pay-per-datapoint) impractical in Web2.

Project Deep-Dives: Problems, Approaches, and Solutions

  • Chain Aware (Martin)
    • Problem clusters: (1) Fraud detection, (2) Growth technology (acquisition at scale), (3) Portfolio construction.
    • Solutions (12 products; 13th coming):
      • Predictive fraud detection (before transactions); predictive rug-pull detection; transaction monitoring for compliance.
      • Growth agents that analyze on-chain user behavior, generate resonant content/CTAs, and enable one-to-one targeting.
      • MCPs to help other AI agents access blockchain data; portfolio MCP plus portfolio rebalancer/calculator/constructor/tree-balancer tools.
    • Claims: Fraud is materially undercounted (Chainalysis says ~0.4% of addresses are fraud; they estimate 15–20%); advocates predicting fraud pre-transaction.
  • Weishuo/Vishua (Aiden)
    • Main challenge: Assets are scattered and largely idle; interop risk and compliance hurdles block institutions.
    • Approach: 3-layer design where client assets remain in-place (custody-compliant), with ZK proofs transporting attestations (ownership/terms/duration) to counterparty chains for verifiable execution across SVM/EVM/MoveVM.
    • Outcome: A trustless ground layer for humans/agents to execute; execution traces (success/failure) become high-quality data to refine institutional agents. Early traction: ~$700M capital committed; initial batch included ~600 BTC.
  • DeepSafe AI (Raul)
    • Problem: Lack of cryptographically verifiable trust across digital operations (AI outputs, cross-chain bridges, CEX on-chain records).
    • Solution: Decentralized verification infrastructure—a distributed network of attester nodes providing signatures, attestations, and verifiable computation across service/compute/verification layers.
    • Use cases: Validating cross-chain transfers, AI agent identities, auditing data availability, and integrating verifiable proofs into existing stacks. Millions of verifications processed.
  • Kite AI (Henry)
    • Problem: Users don’t trust agents beyond low-stakes tasks due to opacity and unpredictability; yet agents need to transact to be useful.
    • Solution: Kite Passport—assigns identity and wallet to agents, with policies/guardrails and spending rules (e.g., only trade assets >$500M FDV; cap per-transaction spend). Provides a risk floor and control primitives for agentic payments/commerce.
    • Thesis: Agentic internet requires crypto rails for micro, programmable, cross-border payments.
  • Termix AI (Marsha)
    • Problem: Complex DeFi strategies are error-prone and inaccessible to newcomers; executing across many dApps/protocols is risky and costly.
    • Solution: An AI engine that turns natural-language intents into optimized, verifiable, secure on-chain action plans. Simulates a plan, presents it for approval, then executes.
    • Unique angle: Building the underlying “DeFi Llama for agents”—a standardized protocol layer enabling any AI agent to understand and interact with DeFi protocols, thus composing and executing strategies across 100+ protocols with institutional-grade security (partners mentioned: BNB Chain, DODO, GoPass, etc.).

Emerging AI-Agent Standards: X402 and ERC-8000/8004

  • Important note: The discussion referenced both “ERC8000 for” and “ERC8004” interchangeably; context indicates Ethereum proposals around agent identity/provenance.
  • Martin: Strongly favors X402 for pay-per-use over monthly subscriptions and plans to support it. Hasn’t deeply reviewed the Ethereum proposal yet.
  • Aiden: Both are useful enablers for trustless, programmable agent interactions and payments. They have implemented X402 to support B2B-like payments; combined with their trustless interop, clients can source liquidity optimally (wherever it is cheapest) and then use X402 to complete payments (e.g., bills, equipment purchases). Notes off-ramp frictions remain.
  • Raul: Sees X402 standardizing verifiable AI interactions (agent comms, identity proofs, on-chain anchoring) and ERC-8004 enhancing data/model provenance in Ethereum. DeepSafe’s network can serve as the attestation layer for such standards; aims for ecosystem-wide, shared verification norms rather than one standard dominating.
  • Henry: Coinbase Ventures invested in Kite AI; Kite is standards-agnostic. Intends to be the neutral governance/trust layer across agent comms/payment standards (e.g., A2P/A2A, MCP), enabling controlled, policy-bound agent behavior regardless of the protocol used.
  • Marsha: Frames X402 as a universal micro-payment/info “HTTP-like” protocol for instant transfers; frames ERC-8004 as agent-focused identity, validation, and reputation for high-stakes cooperation. Envisions synergy where ERC-8004’s reputation can reference X402 payment proofs to anchor real completed transactions.

AI Trusted Verification: Concepts and Approaches

  • Martin
    • Two verification layers are needed: (1) Model-level performance verification (e.g., backtests for trading agents); (2) Transaction-level cryptographic verification.
    • Calls for public repositories of backtests; Chain Aware published its fraud prediction model and verification script for public reproducibility.
  • Aiden
    • Priority is verification of data quality and execution traces to mitigate hallucinations and domain drift. Domain-specific knowledge is crucial; “GN” models (their term) can outperform LLMs for quantitative tasks.
    • Execution outcomes (success/failure, routes) should be verified and accumulated to create institutional “moats,” evolving agents from a static map to a live “Google Maps”-like decision fabric.
  • Raul
    • Goal: Make AI decision pipelines provable end-to-end (inputs→model→outputs), not just believable. Requires data provenance, model provenance, and execution attestation.
    • Tooling mix and trade-offs: ZK proofs (math guarantees), TEEs (runtime integrity), decentralized attester networks (independent confirmation). Balance depends on use case.
    • Unlocks: Auditable AI-driven DeFi decisions and oracles; verifiable AI-assisted governance; content/data provenance against misinformation. DeepSafe positions as the foundational verification network.
  • Henry
    • The need for on-chain verification grows with the stakes of agentic decisions. As agents manage finance and other high-impact domains, verification becomes essential. Early days, but trend is inevitable.
  • Marsha
    • Trusted verification is foundational to move beyond demos. When AI manages assets/governance, outputs must be provably correct; blockchain validation plus incentives reinforce honest AI behavior and safe automation.

State of the Industry, Catalysts, and Infra Gaps

  • Martin
    • We’re still in experimentation. Next wave hinges on delivering real, value-adding use cases (not hype or wrappers).
  • Aiden
    • Early stage; identity and authorization for agents are critical for both retail and institutions. Sees complementary roles: Kite handles high-frequency consumer-grade payments/identity; Weishuo addresses institutional, large-volume liquidity and interop.
    • Having tackled native Bitcoin interop, they’re confident about tokenized blue-chip assets being agent-managed. Tokenization is inevitable; infra must make it safe and compliant.
  • Raul
    • Groundwork phase with fragmented POCs. For breakout growth, three conditions must align:
      1. Verifiable infrastructure (low-friction proofs of computation, data provenance, agent identity)
      2. Economic alignment (sustainable incentive models for contributors/verifiers/data providers)
      3. Usability (developer tooling and standards making AI-chain integration as simple as a smart contract)
    • Foundational needs: scalable ZK systems, decentralized attester networks, standardized AI provenance metadata, robust off-chain compute ↔ on-chain verification bridges, and governance/dispute frameworks for autonomous agents.
  • Marsha
    • Still early. Trigger will be when model outputs are trusted and validated at scale. Needs: more powerful compute (GPUs), better data pipelines, safer agent architectures, and strong transparency/control primitives.

Notable Highlights and Data Points

  • Chain Aware: 12 shipped predictive analytics products; 13th imminent. Focus areas: predictive fraud/rug pull detection, transaction monitoring, growth agents, MCPs, and portfolio tools.
  • Weishuo/Vishua: ~$700M capital committed; initial on-chain execution batch included ~600 BTC; supports SVM/EVM/MoveVM with ZK-verified interop.
  • DeepSafe AI: Distributed attester-node network across service/compute/verification layers; millions of verifications processed for on-chain and cross-chain use cases.
  • Kite AI: “Kite Passport” introduces identity+wallet+guardrails for agents; standard-agnostic stance; Coinbase Ventures investor and Coinbase listing noted by Henry.
  • Termix AI: NL-to-execution DeFi agent engine; standardized protocol layer for agent↔DeFi interoperability across 100+ protocols; institutional-grade security; ecosystem partnerships mentioned (BNB Chain, DODO, GoPass).

Key Takeaways

  • Complementarity is real: AI adds intelligence/automation; blockchain adds integrity/verifiability. The combination is essential for autonomous finance.
  • Trust is the bottleneck: Verification of data, model, and execution is mandatory for high-stakes agentic use cases (payments, asset management, governance).
  • Interoperability first, intelligence second: Trustless, verifiable execution across chains is the substrate upon which domain-specific AI can compound value.
  • Standards are emerging: X402 (agent micro-payments and interactions) and ERC-8004/“8000 for” (agent identity/provenance) are converging to formalize agent behavior and trust. Ecosystem players favor neutrality and layered integration.
  • Business traction hinges on real use cases: Fraud prevention, user acquisition/growth, and portfolio construction in DeFi are immediate, data-rich opportunities.

Open Challenges and Risks

  • Data and backtesting transparency for AI agents (especially trading) remains sparse; public, reproducible evaluations are needed.
  • Off-ramp/on-ramp frictions still hinder seamless agentic commerce.
  • Balancing verification toolchains (ZK, TEEs, attesters) with cost, latency, and privacy is non-trivial; the right blend will be use-case dependent.
  • Governance, accountability, and dispute resolution for autonomous agents require shared frameworks.

What Could Trigger the Next Wave

  • Mature, low-friction verification layers for agent identity, data provenance, and execution.
  • Clear, sustainable incentive models that reward verifiers, data providers, and infra operators for real value, not speculation.
  • Developer ergonomics: standard SDKs, schemas, and MCP-like access to on-chain data and off-chain tools that make building verifiable agentic apps straightforward.
  • Production-grade interop that keeps assets in-custody while enabling trustless, cross-domain execution with cryptographic assurances.