AB-731: AI Transformation Leader

Build the strategic vocabulary a business leader needs to talk about generative AI with confidence. You'll distinguish generative AI from earlier AI paradigms, learn how pretrained and fine-tuned models are chosen for business problems, unpack the cost drivers (tokens, inference vs training) that determine what an AI project actually costs, and put common failure modes (fabrications, bias, over-reliance) into a mitigation-ready framework. By the end of this lesson you will be able to recognize when generative AI is the right tool for a business need and defend the choice to a skeptical CFO.

Move from "what generative AI is" to "how do we deploy it safely at Halcyon." You'll learn prompt-engineering discipline every leader must recognize, retrieval-augmented generation (RAG) as the enterprise grounding pattern, Microsoft Foundry's Content Safety primitives (prompt shields, groundedness detection, protected material detection), and the data-handling boundary that determines what data reaches the model, what stays in your tenant, and what is contractually excluded from training. By the end you can evaluate any "safe AI" claim from a vendor or internal team.

Move from "AI in general" to "Microsoft's AI stack in particular." You'll cover what Microsoft 365 Copilot actually does inside Word, Excel, PowerPoint, Outlook and Teams; how Business Chat grounds on your work data via Microsoft Graph; and how Copilot Studio turns the out-of-box Copilot into a set of Halcyon-specific agents with topics, actions, and knowledge sources. By the end you can hand Marta a shortlist of the first three Halcyon-specific agents to build.

Understand Microsoft Foundry — the platform where Halcyon builds AI capabilities that outgrow Copilot Studio. You'll cover the Foundry portal, its model catalog and deployments, the Foundry Tools stack (prompt flow, evaluations, tracing, content safety), the Foundry agent service, and the decision framework for when to reach for Foundry vs Copilot Studio vs out-of-box Copilot. By the end you can describe Foundry to a CIO in five sentences and defend a "should we build on Foundry?" recommendation.

Close the "which lever should we pull?" arc with a rigorous framework for choosing between buying a productized AI capability from Microsoft, extending a Microsoft platform with Halcyon-specific configuration, and building bespoke on Foundry or open-source stacks. You'll learn the signals, the trade-offs (cost, time-to-value, control, differentiation), and the anti-patterns (buying to look modern, building for prestige, extending to avoid a decision).

Move from "what to build" to "how to be trusted." You'll cover Microsoft's six Responsible AI principles as they apply at Halcyon, the composition and remit of the AI Council, the risk-classification framework that determines review depth, the workflows for high-risk decisions, and the specific regulatory posture a FINRA-registered firm must be ready to defend. By the end you can answer any audit-committee or examiner question about AI oversight without guessing.