FA-0708AI for Leaders & BusinessAgentic & Generative AI

Understanding AI, Tools and Prompt Engineering

A practical one-day introduction for analysts and business professionals

Introduction

Why this course

Explore AI foundations, compare selected tools and practise prompt and context design for everyday analytical work. This one-day introductory workshop combines short explanations, guided demonstrations and a small prompt-refinement exercise.

Compare capabilities, data access, reliability, cost and human oversight rather than assuming that every tool or subscription offers the same features. Examples use approved sample data; no coding or production deployment is required.

Learning outcomes

Distinguish rule-based systems, machine learning, generative models and agentic workflows.
Explain LLM training, adaptation and inference at an introductory level, including uncertainty and hallucination.
Compare conversational assistants, Power Platform AI capabilities, agentic tools and source-grounded notebook assistants.
Evaluate a proposed use case against accuracy, permissions, privacy, integration and cost constraints.
Build and evaluate a reusable prompt with relevant context, examples, output constraints and human verification.

Prerequisites

Comfort with business analysis, reporting or everyday office software.
Basic familiarity with cloud applications; no programming or machine-learning background required.
Access to the approved demonstration tools where available; account, licence and administrator restrictions are checked before practical work.

Training outline

5 modules

01Module 1 — AI foundations and practical limits4 topics

Rule-based systems versus supervised, unsupervised and reinforcement learning; neural networks and deep learning in context.
LLMs: an accessible overview of model architecture, training, fine-tuning and inference; fine-tuning is discussed conceptually, not promised as a hands-on service.
Generative, predictive, prescriptive and conversational applications; distinguish an assistant response from an agent performing a sequence of actions.
Potential help with routine tasks and insight generation; examine bias, hallucination, missing context and over-reliance.

02Module 2 — Agents, tools and MCP3 topics

Observe, plan, act and evaluate: trace a bounded multi-step workflow and identify the human approval points.
Tool orchestration and Model Context Protocol: connecting an AI application to permitted data and tools through an open protocol.
Interoperability does not grant data access or guarantee trustworthy actions; review authentication, permissions, auditability and action limits.

03Module 3 — Compare AI tool categories5 topics

ChatGPT for drafting, summarisation, questions and selected data or code assistance; check outputs against supplied evidence.
Connected apps and plugins can add tools and reusable workflows, subject to account and workspace permissions; distinguish these end-user capabilities from developer API tool calling.
Microsoft Power Platform: distinguish AI Builder, Copilot capabilities in individual products and Copilot Studio rather than treating them as one universal AI hub.
Agentic assistants, including Manus as a comparison example: evaluate task scope, reliability and supervision without assuming a particular internal multi-agent architecture.
Source-grounded notebook assistants from Google, including Gemini Notebook: use supported documents and web sources, follow citations and compare generated summaries with the originals. Public YouTube sources require captions; arbitrary video ingestion is not assumed.

04Module 4 — Guided tool exploration5 topics

Use two representative tasks, such as a document brief and a small reporting question, to compare available tools.
Observe Power Platform automation or application examples and discuss connectors, setup, governance and feature availability.
Refine a prompt, inspect an answer and identify where the assistant needs better context or a reliable source.
Review an agentic workflow demonstration and a source-grounded research brief; record unsupported statements, missed evidence and supervision needs.
Broader product comparisons are demonstrations or discussion, not a promise of hands-on accounts for every platform.

05Module 5 — Prompt and context engineering7 topics

Define the task, audience, constraints and output format; compare zero-shot instructions with a small set of relevant examples.
Use roles and clear structure where helpful, but do not assume role labels improve factual accuracy. Adapt prompts to the selected model and test alternatives.
Request concise explanations and verifiable evidence rather than hidden chain-of-thought; reasoning models do not generally need step-by-step prompting.
Permit uncertainty and ask for citations where sources are available, then check the citations and claims; these instructions do not eliminate hallucination or bias.
Select task-relevant documents and context, separate reference content from instructions, and manage stale or changing context. Distinguish supplying context from embedding-based retrieval.
Refine one analyst prompt against sample cases using agreed quality criteria and human review; retain a reusable template with its assumptions and limits.
Discuss iterative coding-assistant trends as context, not a coding lab or an attributed guarantee; choose a bounded next experiment.

A programme built around your team.

Share your training goals and requirements.