Rob Hadick

As digital assets, tokenized deposits, and on-chain financial infrastructure move closer to the regulated banking system, a new control layer is emerging: agentic artificial intelligence. Unlike traditional automation or static smart contracts, agentic AI systems can interpret intent, evaluate conditions, and take action autonomously within defined guardrails.

For banks, this evolution raises fundamental questions about execution, risk, governance, and accountability. Who—or what—makes decisions in an always-on, programmable financial environment? How can autonomy be introduced without sacrificing control, compliance, or trust? And where does agentic AI become a competitive advantage versus a systemic risk? If finance is becoming programmable and always-on, the real question isn’t whether banks will use agentic AI—it’s whether they’ll control it, or be forced to react to it.

This panel discussion will explore how agentic AI could reshape on-chain finance and digital asset operations, from liquidity and treasury management to compliance, custody, and market structure—and what bank leaders must do now to prepare. The in-depth conversation will include:

• Why rule-based automation is insufficient at institutional scale.
• How agentic AI introduces decision-making, not just execution.
• Where autonomy adds value—and where it becomes dangerous.
• Who is accountable when an autonomous agent acts?
• Human-in-the-loop vs. policy-in-the-loop models.
• Designing kill-switches, escalation paths, and auditability.
• Real-time liquidity, settlement, and reconciliation with programmable money.
• Why tokenization without orchestration breaks at scale.
• Continuous, real-time compliance vs. post-transaction controls.
• Embedding AML, sanctions, and policy enforcement on-chain.
• Model risk, explainability, and regulatory defensibility.
• Coordinating liquidity across DEXs, CEXs, and tokenized markets.
• The role of banks as stabilizers in on-chain markets.
• Policy-driven smart wallets and automated approvals.
• Secure key management with agent-assisted controls.
• Making self-custody viable for regulated institutions.
• Emergent behavior between interacting agents.
• Concentration risk if agents rely on similar models or signals.
• Where to experiment safely (sandboxes, pilots, limited scope).
• How to engage regulators early and credibly.