Skip to main content

AI Portfolio Assistant Demo

Intent and context names below match the FDC3 interop catalog.

  1. Select a portfolio (observe)

    • The Portfolio app broadcasts an fdc3.portfolio context.
    • The AI Assistant receives the selected portfolio and aligns with the desktop.

  2. Analyse portfolio risk (analyse)

    • The user asks: “What are the biggest risks in this portfolio?”
    • The AI Assistant raises demo.GetPortfolioRisk with fdc3.portfolio.
    • The Desktop Agent intercepts the intent and checks it with the Governance Gateway.
    • Governance returns ALLOW.
    • The Risk app performs the deterministic analysis and returns demo.portfolioRisk (with FDC3 Security provenance).
    • Optionally, the user asks for a stress test; the Assistant raises demo.RunStressScenario with demo.stressScenarioRequest and receives demo.stressScenarioResult from the Risk app.

  3. Gather evidence and explain (explain)

    • The AI Assistant raises demo.GetResearch with fdc3.instrument, fdc3.instrumentList, or fdc3.portfolio.
    • Governance returns ALLOW.
    • The Research app returns demo.researchBundle (signed for provenance).
    • The Assistant combines demo.portfolioRisk and demo.researchBundle into an explanation for the user.
    • Optionally, supporting View* intents (ViewAnalysis, ViewNews, ViewInstrument, and similar) open the underlying specialist UIs without changing the analysis itself.

  4. Request an investment recommendation (recommend)

    • The user asks: “What should I do about the energy concentration?”
    • Governance evaluates applicable controls, such as:
      • approved AI model;
      • current portfolio data;
      • valid risk analysis and research evidence;
      • permitted recommendation capability.
    • Governance returns ALLOW or CONDITIONAL ALLOW.
    • The Assistant produces an on-screen portfolio recommendation — a recommended portfolio change with rationale and supporting evidence, not an executable order.

  5. Create an order proposal (propose)

    • The user says: “Reduce my energy exposure to 15%.”
    • The AI Assistant raises demo.CreateOrderProposal with demo.portfolioRecommendation.
    • Governance verifies that the AI is permitted to construct, but not execute, a proposed trade.
    • The Order Entry app returns demo.orderProposal — a draft basket for human review (signed for provenance).

  6. Human review in Order Entry

    • The draft demo.orderProposal is presented in the Order Entry app.
    • Governance treats this as ALLOW — HUMAN APPROVAL REQUIRED.
    • The user inspects the proposed trades. A reviewed basket may be represented as an fdc3.orderList for any subsequent execution path.

  7. Attempt autonomous execution (execute — denied)

    • The user (or the Assistant under user pressure) attempts: “Just execute those trades.”
    • The AI Assistant raises demo.SubmitOrders with fdc3.orderList.
    • The Desktop Agent checks with the Governance Gateway.
    • Governance determines that the AI has no delegated trade-execution authority.
    • Governance returns DENY.
    • The Desktop Agent does not forward the intent to Execution / Order Entry.

  8. Inspect the governance decision

    • The Governance app displays the decision and its traceability:
      • governed activity (including the named FDC3 intent);
      • actor;
      • applicable policy/control;
      • evaluation results;
      • evidence;
      • decision;
      • required human action.
    • The denied demo.SubmitOrders demonstrates that governance changes system behaviour rather than merely recording it.

  9. Connect the runtime decision to ISO 42001

    • Show how the controls and evidence relate back to the organisation's AI governance requirements.
    • Demonstrate the chain from:
      • ISO 42001 guidance;
      • risk;
      • control/policy;
      • sensitive activity;
      • evaluation;
      • evidence;
      • runtime decision.

  10. Show the authority progression

    • Summarise the increasing level of delegated authority:

      observe → analyse → explain → recommend → propose → execute

    • Map the demo intents onto that chain:

      • analyse: demo.GetPortfolioRisk, demo.RunStressScenario
      • explain: demo.GetResearch
      • propose: demo.CreateOrderProposal
      • execute (blocked): demo.SubmitOrders

    • The key point: Ten-Factor governance determines how far the AI may progress along that chain, based on explicit, machine-readable governance.

  11. Opportunities for further expansion

    • Here, we are just showing how we can use FDC3 to govern an AI application.
    • We could apply the same governance gateway to MCP.
    • Also, the governance runtime itself could be an AI agent running to observe and constrain the assistant.