Before you approve an agent that moves money.
Stablecoin University · Executive programAgentic Banking & Stablecoin Agentic Payments is a two-day program for the executives who sign that approval. It covers how agents connect to core banking systems, how they settle over stablecoin rails, and what has to be proven before one goes live.
An agent that succeeds on 95% of runs completes eight in a row 66% of the time.
A payments customer gets one attempt, so the second number is the one they live with. The program teaches the people who sign off how to ask for it, how it is tested, and what else to require before an agent touches production liquidity.
Who should be in the room.
Who it is forC-suite and the heads of payments, treasury, digital banking, risk and compliance, and enterprise architecture. Two days of architecture briefings, demonstrations and tabletop exercises.
What you take back to the board.
What you leave withAn architecture blueprint
For an agentic treasury and liquidity sweeping system: which steps stay fixed workflows, which go to an agent, and where a person approves.
A reliability threshold
The consistency measure to demand before an agent touches production liquidity, and how to have it tested automatically.
Payment limits enforced in code
Session keys, per-transaction and velocity caps, allow-listed counterparties, and delivery-versus-payment settlement.
A decision
Day 2 closes with a boardroom and red-team exercise on a live stablecoin B2B payment pilot. You approve it or decline it, with reasons.
Two days, hour by hour.
AgendaEach topic is taught through retail and commercial banking, core ledger integration and stablecoin settlement.
Day 1: Agentic banking architecture, rails and orchestration
How tool-using agents connect to core banking systems, where deterministic workflows must stay in charge, and how agents coordinate across institutions.
- 08:30 – 09:45Module 1
From chatbots to agentic banking
Workflows versus autonomous agents, compound AI systems in banking, and why reliability matters more than model size.
- 09:45 – 11:00Module 2
Financial-grade LLMs and grounded knowledge
Structured output (ISO 20022, JSON), context engineering, and agentic retrieval (RAG) over regulatory and policy text.
- 11:15 – 12:30Module 3
Tool use, MCP and banking APIs
Connecting agents to core ledgers, KYC and AML engines, and wallet infrastructure through the Model Context Protocol (MCP).
- 13:30 – 15:00Module 4
Durable orchestration and design patterns
Workflows versus ReAct agents, and durable execution for long-running wire, trade finance and settlement flows.
- 15:15 – 16:30Module 5
Agent memory and multi-agent networks (A2A)
Episodic and semantic memory for relationship banking, and agent-to-agent protocols for interbank and B2B commerce.
- 16:30 – 17:30Workshop
Day 1 tabletop: architecture blueprint
Scoping an agentic treasury and liquidity sweeping system.
Day 2: Stablecoin agentic payments, risk, evals and governance
How autonomous stablecoin payments work, how they are tested for consistency, how they are attacked, and what a board needs before it approves a pilot.
- 08:30 – 10:00Module 6
Stablecoin agentic payment rails
Machine-to-machine micropayments, programmable escrow, smart-wallet session keys, and round-the-clock stablecoin and tokenized-deposit settlement.
- 10:15 – 11:30Module 7
Data flywheels and cost/accuracy optimization
Turning transaction traces and human overrides into proprietary data, and when to optimize prompts versus fine-tune weights.
- 11:30 – 12:45Module 8
Evaluation and zero-tolerance reliability
Designing banking benchmarks, LLM-as-judge scoring, and why consistency across runs (pass^k) beats best-of-k (pass@k) in payments.
- 13:45 – 15:15Module 9
Agent safety, guardrails and financial liability
Indirect prompt injection (poisoned invoices), sandboxing, spend-limit guardrails, Know Your Agent (KYA), and liability allocation.
- 15:30 – 16:30Module 10
Proactive personal banking and software agents
Mixed-initiative personal finance agents and coding agents for legacy core-banking modernization.
- 16:30 – 17:30Workshop
Day 2 capstone: boardroom and red-team exercise
Approving a live autonomous stablecoin B2B payment pilot.
Five readings, one page each.
Pre-readParticipants receive a one-page brief on each reading, so the sessions can stay on strategy and architecture.
- System strategy
Zaharia et al., "The Shift from Models to Compound AI Systems" (Berkeley AI Research, 2024)
Why banking AI is a systems-engineering problem, not a model-shopping problem. bair.berkeley.edu
- Design discipline
Schluntz and Zhang, "Building Effective AI Agents" (Anthropic, 2024)
When a simple deterministic workflow is the right answer and when an autonomous agent is. anthropic.com
- Connectivity standard
Model Context Protocol specification (2025)
How agents connect to enterprise banking APIs and wallet infrastructure through one interface. modelcontextprotocol.io
- Reliability and governance
Grace et al., "Demystifying Evals for AI Agents" (Anthropic, 2026)
Board-level reliability metrics, including pass^k, for financial-grade deployment. anthropic.com
- Risk and privacy
Shao et al., "PrivacyLens" (NeurIPS 2024); Cemri et al., "Why Do Multi-Agent LLM Systems Fail?" (NeurIPS 2025)
Information leakage and coordination failures to understand before launching customer-facing agents. arxiv.org/abs/2409.00138arxiv.org/abs/2503.13657
Links checked 21 November 2026.
The full lessons are in the workspace.
Self-paced courseMembers get every module in depth: worked banking scenarios, knowledge checks and both workshop guides.
Education, not legal, regulatory or investment advice. Nothing here claims that live third-party provider execution is generally available on StablecoinRoadmap. Regulatory references are current as of 21 November 2026.
Bring the people who will sign.
Executive programTell us who would attend and the decision they are preparing for. We reply by email to agree format, dates and scope.