Agentic AI for Treasury, Capital Markets & Trade Finance

Agent value in markets and treasury is decided by clock speed and reversibility, not by model capability. This module maps where agents genuinely pay in front, middle and back office, then establishes the perimeter that actually binds them in 2026 — RTS 6, MAR, DORA and the GCC principles regime — and confronts the model risk vacuum left by SR 26-2.

• 1.Three Clocks: Where Agents Genuinely Pay in a Markets and Treasury Stack10m
• 2.What Actually Binds an Agentic Trading Desk: RTS 6, MAR, DORA and the SR 26-2 Gap10m
• 3.Lab: Building an Agent Perimeter Register for a Treasury and Markets Function25m

This module puts reinforcement learning agents inside the limit order book, where impact, adverse selection and non-stationarity decide whether a policy is a strategy or a simulator artefact. It covers reward design, agent-based simulation and off-policy evaluation, then maps the resulting system onto RTS 6 obligations and the current absence of any supervisory instrument covering agentic trading.

• 1.Market Microstructure as the Agent's Environment: Order Books, Impact and Adverse Selection10m
• 2.Reward Design, Agent-Based Simulation and Off-Policy Evaluation for Trading Agents10m
• 3.Lab: Build a TWAP-Benchmarked Execution Agent and Its RTS 6 Approval Dossier20m

This module sets the boundary between what a language model may decide and what deterministic pricing code must compute in a derivatives workflow. It covers volatility surface construction as a tool, hedge narrative generation as a regulated output, and the model governance gap left by SR 26-2's exclusion of agentic AI. The lab produces a Hedge Recommendation Pack for a Gulf airline exposure.

• 1.Pricing Engines as Tools: Wiring an Agent to QuantLib and the Volatility Surface10m
• 2.Hedge Narratives, Suitability and the Model Governance Gap After SR 26-210m
• 3.Lab: Build a Hedging Advisory Agent for a Gulf Airline's EUR and INR Receivables20m

Treasury automation fails on data and definitions far more often than on modelling. This module separates intraday, operational and structural forecasting, grounds cash positioning in the Basel and CBUAE liquidity constraints that actually bind, and settles where an agent's authority should stop before a payment leaves the building.

• 1.Cash Flow Forecasting Agents: Horizons, Baselines and Where Accuracy Stalls10m
• 2.Cash Positioning, Intraday Liquidity and the Limits of Agent Authority15m
• 3.Lab: A Thirty-Day Liquidity Forecast Agent with a Variance Attribution Pack25m

Extends the treasury agent stack from cash positioning into currency risk, covering how an agent aggregates and nets FX exposure across ERP, TMS and unbooked contracts before recommending anything. It sets the boundary between recommendation and execution, and grounds hedge choices in policy, IFRS 9 documentation and Gulf market realities including the dirham and riyal pegs.

• 1.Where FX Exposure Actually Lives: Transaction, Translation and Economic Risk in a Pegged Market10m
• 2.Instrument Selection, Hedge Ratios and the Guardrails That Make an FX Agent Deployable10m
• 3.Lab: Build an FX Exposure Aggregation Agent and Produce a Hedge Recommendation Pack20m

Fixed income is where agentic systems meet thin data: most bonds do not trade daily, spreads are model outputs as often as observations, and the ALCO reads a paragraph rather than a grid. This module builds pricing, duration and surveillance agents that keep valuation deterministic, treat provenance as an output, and handle rate of return risk on sukuk as a structural problem rather than a flag.

• 1.Bond Pricing Agents: Curve Construction, Spread Decomposition and Stale Quotes10m
• 2.Duration Management and IRRBB: EVE, NII and the Sukuk Repricing Problem10m
• 3.Lab: Build a Bond Surveillance Agent That Produces a Defensible EVE Pack20m

Documentary credit checking is a legal determination under UCP 600 with a five-banking-day clock and a preclusion penalty, not a data-extraction problem. This module sets out where agents genuinely help — cross-document reconciliation and evidence assembly — where they must not decide, and how financial crime screening and the current supervisory gap shape the build.

• 1.What UCP 600 Actually Asks a Document Checker to Decide15m
• 2.Extraction, Cross-Document Reconciliation and the Refusal Path10m
• 3.Lab: Building a Discrepancy Adjudication Pack for a Sight LC Presentation25m

Supply chain finance is the open-account counterpart to the documentary work of Module 7, and it removes the examiner from the loop entirely. This module treats the resulting problem as one of data control and priced judgement: what makes an invoice eligible, what a discount offer is actually worth to three different parties, and where an agent's authority has to stop.

• 1.Reverse Factoring versus Dynamic Discounting: Whose Cash, Whose Risk15m
• 2.Pricing the Offer: Discount Curves, Supplier Risk and Adverse Selection10m
• 3.Lab: Building a Dynamic Discounting Offer Policy and Invoice Exception Register25m

Reporting is where a markets desk's data quality becomes a supervisory matter. This module treats MiFIR Article 26, EMIR Article 9 and the CFTC's Part 43 and Part 45 obligations as three projections of one trade record, and locates the agent where the manual cost actually is: reportability determination, break triage and disciplined error remediation.

• 1.Three Regimes, One Trade: MiFIR Article 26, EMIR Article 9 and CFTC Parts 43 and 4515m
• 2.Eligibility, Breaks and the Error Ladder: Where a Reporting Agent Earns Its Keep15m
• 3.Lab: Build a Cross-Regime Reporting Control Pack for a Single Product25m

Module 9 built agents that report to regulators; this one builds agents that publish to clients, which is the harder conduct problem. It works through what an automated earnings note legally is under MAR and the MiFID research rules, how numeric provenance is enforced, who certifies the output, and what the reversal of research unbundling does to the business case.

• 1.From Filing to Publishable Note: Provenance and the Investment Recommendation Line15m
• 2.Certification, Conflicts Disclosure and the Economics After Unbundling Reversed15m
• 3.Lab: An Earnings Note Production Standard with Provenance Ledger and Conflicts Register25m

Surveillance inverts the seat the course has occupied since Module 2: you are now the firm explaining its own order flow to someone who already finds it odd. This module works through what MAR and the Dodd-Frank anti-disruptive-practices provisions actually require, where exchange rules go further, and why an agent's honest output is an evidence pack rather than a verdict.

• 1.Reading Intent from an Order Book: MAR Article 12 and CEA Section 4c(a)(5)15m
• 2.From Threshold Alerts to Agentic Triage: Designing the Alert-to-STOR Pipeline15m
• 3.Lab: Building a Spoofing Triage Agent and a Surveillance Evidence Pack25m

On-chain agency is constrained less by what a model can reason about than by capital treatment, key custody and the fact that a settled transaction cannot be recalled. This module fixes the current perimeter under MiCA, the UAE regimes and Basel SCO60, then works through irreversibility, oracle dependence and the authority architecture a regulated institution can actually defend.

• 1.Where the Perimeter Sits: MiCA, the UAE Regimes and the Basel SCO60 Capital Wall15m
• 2.Irreversibility, Key Custody and Oracle Dependence: The Three Failure Modes10m
• 3.Lab: Building an On-Chain Agent Authority Matrix and Pre-Flight Control Sheet25m

SR 11-7 was rescinded on 17 April 2026, and its successor SR 26-2 expressly excludes generative and agentic AI from scope. This module works from that gap rather than around it: what the narrowed model definition does to your inventory, how to run effective challenge on systems that will not reproduce, and where the EU and CBUAE texts cover what the American letter now leaves out.

• 1.The April 2026 Rescission: What SR 26-2 Covers and What It Deliberately Does Not15m
• 2.Effective Challenge Without a Ground Truth: Validating Non-Deterministic Quant Agents15m
• 3.Lab: Building a Scope Determination Memo and Agentic System Risk File25m

Settlement compression is a governance problem before it is a modelling one. This module fixes the real calendar — T+1 in force in the US, the UK, EU and Switzerland moving together on 11 October 2027, T+0 optional only in India — then works through instant payment rails where a completed transfer cannot be recalled, and settles where an agent's authority must end.

• 1.The Settlement Calendar: T+1 in Force, T+0 Optional, and Where the Slack Went10m
• 2.Instant Payment Rails and the Point of No Return: Aani, sarie and the EU Instant Payments Regulation15m
• 3.Lab: A Fail-Risk Scorecard and Irreversibility Gate for a Settlement Agent20m

Brings the course together by rebuilding the treasury agent stack for an entity where permissibility is a hard system constraint. It re-scores the module 1 perimeter register on Islamic reversibility, restricts module 4's buffer ladder and module 5's hedging set to instruments the ISSC has authorised, and settles the governance case left open by SR 26-2 in module 13.

• 1.Sharia Constraints as System Requirements: What an Islamic Treasury Agent May Not Propose15m
• 2.Architecture of an Islamic Treasury Intelligence Agent: Constraint Engine, Evidence and the Governance Case15m
• 3.Lab: Build the Treasury Intelligence Agent Design Dossier for a GCC Islamic Bank25m