← Course kit · Day 3 outline · for the program owner

Day 3 Outline: Advanced Agents with Copilot Studio

The third day of Applied AI at TIYA Services. Day 2 built agents that wait for a person to ask. Day 3 builds agents that start themselves: on an incoming email, on a schedule, and by handing work to other agents. It also covers how to run them safely once no one is watching.

Day 3[Date TBD] Hours8:00 AM – 4:00 PM ToolMicrosoft Copilot Studio InstructorsAlex Djalali, Alex Rhodeen

1

Audience and prerequisites

  • Who: TIYA staff who completed Day 2 and will design or own agents for their team: facilities leads, the work order desk, business management, and supervisors. About 30 per cohort, with limited remote seats.
  • Prerequisite: Day 2, Agentic AI with Microsoft Copilot. Day 3 assumes participants can write agent instructions from the template, write tests before building, and rate an agent's risk tier. It opens with a 15-minute Day 2 review and doesn't depend on any agent built on Day 2.
  • Accounts: a Copilot Studio license and maker access to a training environment (section 5).

2

Learning outcomes

On completing Day 3, participants will be able to:

OutcomePracticed in
LO1Decide whether an agent should run on demand, on an event, or on a schedule, and design its trigger, conditions, actions, and human check.Builds 2 and 3
LO2Build and test event-triggered and scheduled agents in Microsoft Copilot Studio that hand their output to a person for review.Builds 1–3
LO3Design and build a multi-agent solution with an orchestrator and specialist agents, and judge when one agent is enough.Build 4 · One agent or many?
LO4Operate autonomous agents safely, including untrusted input, credentials, cost, monitoring, and a documented off switch.Red team · Runbook · Shutdown check

3

The day, 8:00 to 4:00

About 5¼ hours at the keyboard. Every build ends with a person checking what the agent produced.

TimeBlockTopicsActivityLO
8:00 – 8:30Welcome & Day 2 in Review
30 min
Copilot Studio login check · the autonomy ladder, rungs 4 and 5 · what changes when no one is watching: untrusted input, real actions, real costLogin check · show of hands on agents built since Day 2
8:30 – 9:45From Agent Builder to Copilot Studio
75 min · hands-on
A tour of Copilot Studio · tools and connectors as the way an agent acts · general-knowledge and web settings as a lock · the activity view · publishing for other agentsBuild 1 SOP Navigator, rebuilt and retested against the Day 2 tests, then publishedLO2
9:45 – 10:00Morning break
10:00 – 11:45Event-Triggered Agents
105 min · hands-on
Trigger, condition, payload, action, human check · agent vs. payload instructions · the email is data, not instructions · fixed tool inputs · leaving out the send-email toolBuild 2 Work Order Intake: eight sample emails, triage cards posted to Teams for a Facilities LeadLO1 · LO2 · LO4
11:45 – 12:30Lunch
12:30 – 1:30Scheduled Agents
60 min · hands-on
Scheduled prompts in Copilot Chat vs. recurrence triggers in Copilot Studio · let the flow count, let the agent write · testing on a short schedule, then turning it offBuild 3 Monday Backlog Briefing from an Excel table, checked against the answer keyLO1 · LO2
1:30 – 2:45Multi-Agent Orchestration
75 min · hands-on
Orchestrators and specialists · child vs. connected agents · descriptions drive routing · the cost of a team of agentsBuild 4 Facilities Desk with six routing tests · "One agent or many?" sortLO3
2:45 – 3:00Afternoon break
3:00 – 3:45Operating Autonomous Agents
45 min
Five new risks: injection, credentials, silent failure, cost, chained errors · the off switch · risk tiers for triggered agentsRed-team the intake agent with three attack emails · write its runbookLO4
3:45 – 4:00Closing & Assessment
15 min
What to propose at TIYA, and who approves itKnowledge check · trigger shutdown check · evaluation · certificates

4

The four builds

Each build adds one capability on top of the last, using the same fictional SOPs and work order data as Day 2. Step-by-step instructions are in the Day 3 workbook.

Build 1 · 8:30

SOP Navigator (Studio)

  • Rung: 3, on-demand
  • Knowledge: the four sandbox SOPs
  • Teaches: the Copilot Studio workspace, settings as locks, publishing for reuse
  • Human check: the person asking
Build 2 · 10:00

Work Order Intake

  • Rung: 4, event-triggered
  • Trigger: a new email with [WO-TRAIN] in the subject
  • Tools: post to Teams, to the owner only; no send-email tool
  • Teaches: payloads, untrusted input, tools as locks
  • Human check: a Facilities Lead reads every card
Build 3 · 12:30

Monday Backlog Briefing

  • Rung: 4, scheduled
  • Trigger: recurrence; 5 minutes for testing, then Monday 7:00, then off
  • Tools: read the Excel table, post to Teams
  • Teaches: schedules, cost, checking AI arithmetic
  • Human check: the lead verifies numbers before sharing
Build 4 · 1:30

Facilities Desk

  • Rung: 5, orchestrated
  • Agents: SOP Navigator (connected), Priority Advisor and Notice Drafter (child)
  • Teaches: routing by description, child vs. connected, when not to split
  • Human check: the person asking; notices are drafts

5

Environment requirements

Day 3 needs more of TIYA's tenant than Days 1 and 2 did. Confirm each item with TIYA IT, in writing, at least 14 days before.

If Copilot Studio isn't available

Run Builds 2–4 as instructor demos on screen. Participants complete the trigger canvas, triage log, routing tests, and runbook on paper from the demo output. Build 1 can run in the Copilot agent builder instead. The outcomes shift from "build" to "design and evaluate".

6

Assessment and certificate

  • Formative: an eight-question knowledge check, self-scored. Instructors collect anonymous per-question scores.
  • Evidence of work: the triage log from Build 2, the red-team log, and the runbook.
  • Certificate (proposed): full attendance (up to 30 minutes excused), Build 2 run against the sample inbox, a completed runbook, and every trigger shown turned off before leaving.

7

Materials

ItemFor
Day 3 slides (HTML, PowerPoint, PDF)Instructors · participants
Day 3 workbook: Day 2 review, four builds, sample inbox, red-team log, runbook, shutdown checklistParticipants
Day 3 agendaEveryone
Day 3 assessment kit: certificate criteria, answer keys, knowledge check, exit checklist, evaluationInstructors
Facilitator guide, Day 3 section: run sheet, talking points, demos, watch-fors, debriefsInstructors
Sample inbox (Word): eight request emails and three red-team emailsParticipants
Work order export as an Excel table (WorkOrders, 80 rows)Participants
The four sandbox SOPs, also on one web pageParticipants

8

Decisions for the program owner

  1. Date. Day 3 has no date yet. Every Day 3 document shows "[Date TBD]".
  2. Copilot Studio access. Day 3 depends on it (section 5). If TIYA can't license participants, decide between instructor-demo delivery and postponing.
  3. Who attends. Day 3 suits people who will own agents, not everyone who took Day 2. Consider a smaller cohort of 15–20.
  4. Who approves triggered agents at TIYA. The closing asks participants to name the person who would approve a triggered agent. TIYA should name that role before the day.
  5. Certificate criteria. Confirm the proposal in section 6.