AI Agents Immersion Workshop — Participant Workbook

An AI Agent is software that perceives its environment, reasons about a goal, and takes actions — autonomously or with human guidance — to achieve a defined outcome.

Key Idea: AI Agents are not just chat interfaces — they are goal-oriented systems that can perceive, reason, and act.

1. P Perceive — Reads data from email, SharePoint, Teams, databases, or external systems
2. R Reason — Uses an LLM to interpret instructions, context, and goals
3. A Act — Sends messages, updates records, calls APIs, generates documents
4. L Learn — Improves with feedback, memory, and grounding in your knowledge

• Custom Copilot persona with a specific name, role, and tone
• Scoped to specific knowledge sources (SharePoint, files, URLs)
• Built in minutes using Copilot Studio Lite — no coding required
• Deployed in Microsoft Teams, SharePoint, or Copilot Chat
• Configured via natural language instructions

When to Use: You have docs/SOPs that employees constantly ask about, you want a scoped Q&A bot without coding, or you need an HR policy / compliance / training materials assistant.

No configuration required. Available in Copilot Chat, Teams, and M365 apps.

1. 1 Instructions — System prompt, persona, scope, tone, guardrails
2. 2 Knowledge — SharePoint, files, URLs, Dataverse, web search
3. 3 Tools & Actions — Send email, update records, call APIs, create tasks
4. 4 Connectors — Power Platform, Microsoft Graph, custom REST APIs
5. 5 Triggers — User message, schedule, event, form, workflow

Goal: Discover what pre-built agents exist in Microsoft 365, interact with at least two, and document your observations.

You will need: Microsoft 365 login with Copilot license • Edge or Chrome browser

1. 1 Open Microsoft 365 at m365copilot.com and sign in with your work account.
2. 2 Open Copilot Chat. Ask it to summarize a document or draft an agenda for a team meeting.
3. 3 Open the Agents/Agent Store experience and browse available agents. Click the agent gallery icon (top right of the chat bar).
4. 4 Select one pre-built agent relevant to your work (e.g., Researcher, Analyst, or SharePoint Agent). Launch it.
5. 5 Ask the agent 3 questions that relate to your day-to-day tasks. Note how it responds.
6. 6 Try a second agent. Compare its behavior, tone, and capabilities to the first.

Goal: Create your first declarative agent, ground it with a knowledge source, and test it with real questions.

You will need: Microsoft 365 login • Edge or Chrome • Access to a SharePoint site or sample document (provided) • Copilot Studio access (trial or licensed)

1. 1 Open Agent Builder: Go to m365copilot.com and click the sparkle/agent icon, then “Create an agent.”
2. 2 Click “Create an agent” to open the configuration panel.
3. 3 Name your agent — choose something relevant (e.g., “Franchise FAQ Agent,” “Guest Recovery Helper,” “Drive-Thru Coach”).
4. 4 Write your system instructions. Tell the agent: who it is, what it helps with, its tone, and any constraints. Example: “You are a helpful franchise operations assistant. You answer questions from restaurant managers about policies, procedures, and brand standards. Always be concise and practical.”
5. 5 Add a knowledge source using the options available in your tenant (SharePoint site or uploaded document) that supports your scenario.
6. 6 Test your agent in the right-hand preview pane. Ask it 3 questions a real user might ask.
7. 7 Refine the instructions based on the responses you see. Note what you changed and why.

Take 5 minutes to reflect on both exercises before the hackathon begins:

• What worked well?
• What surprised you?
• What do you want to explore further?

Use this framework to identify the right agent scenario for your team:

1. Pain Point: What process is painful, slow, error-prone, or repetitive?
2. Who is Affected: Which role, team, or brand experiences this most?
3. Current Solution: How is it handled today? (manual process, spreadsheet, email chain, phone calls, nothing…)
4. What Would an Agent Do Differently: How would an agent perceive the situation, decide what to do, and act on it?
5. Agent Type: Pre-built / Declarative / Custom engine
6. Priority Score: Rate on three dimensions (1–5 each):
   • Business Impact (1 = Low, 5 = Critical)
   • Feasibility (1 = Hard, 5 = Easy)
   • Speed to Value (1 = Months, 5 = Days)
   Max: 15 points. Higher score = better fit for today’s build.

• ☐ Scenario selected and documented on Agent Design Canvas
• ☐ Agent type confirmed with facilitator
• ☐ Agent created in the correct tool (Agent Builder or Copilot Studio)
• ☐ System instructions written and tested
• ☐ Knowledge source connected and indexed
• ☐ Agent tested with at least 5 representative questions
• ☐ Edge cases and off-topic questions tested
• ☐ Demo script prepared and rehearsed

Monitors KPI dashboards across brands. Alerts franchise managers when a location falls below thresholds for sales, speed of service, or guest satisfaction.

Handles inbound guest complaint messages from email or social channels. Triages severity, drafts a response, and escalates high-risk cases to the Regional Operations Manager.

Guides new restaurant team members through Day 1–30 onboarding. Answers policy questions, delivers training reminders, tracks completion, and escalates issues to the GM.

Monitors inventory levels against par standards. Detects low-stock situations, recommends reorder quantities based on projected sales volume, and initiates purchase orders with suppliers.

Generates brand-consistent campaign briefs for local store marketing. Incorporates brand guidelines based on the franchise region and brand identity.

Guides shift managers through food safety checklists (temperature logs, holding times, cleaning schedules). Flags non-compliance, generates corrective action tickets, and logs audit results.

Complete this canvas as a team before you start building:

1. S Start Simple — A working MVP beats an unfinished complex agent every time.
2. I Instructions Matter — Be specific about persona, tone, scope, and what to avoid.
3. K Quality Knowledge — Use clean, well-structured docs as your knowledge source.
4. T Test Adversarially — Ask off-topic questions and edge cases.
5. B Business Value First — Tie your agent to a real outcome. Quantify if possible.
6. D Document As You Go — Note design decisions for the demo.
7. G Think About Guardrails — What should the agent NOT do?
8. C Collaborate — Ops people know the pain points; tech people know the tools.

You have 5 minutes to demo + 2 minutes for Q&A. Prepare your story here.

The most important thing you can do right now is commit to a next step:

• The agent scenario I want to pilot: Name the agent, the problem it solves, and the target user.
• My immediate next step: What is the ONE thing you will do in the next 5 business days? Who do you need to involve?
• I will have this done by: Set a date.
• Who I will loop in: Name the stakeholders, technical contacts, or teammates you need.
• My biggest question going forward: What do you still need to figure out?