Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio

Advanced Agents with Copilot Studio

Triggered, scheduled, and multi-agent[Date TBD] · Alex Djalali · Alex Rhodeen

Today

  • 8:00 Login check and Day 2 in review
  • 8:30 From agent builder to Copilot Studio
  • 10:00 Event-triggered agents
  • 12:30 Scheduled agents
  • 1:30 Multi-agent orchestration
  • 3:00 Operating autonomous agents · 3:45 Closing
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio2 / 46

Logistics

  • Breaks at 9:45, 11:45 (lunch), and 2:45
  • Every build is in the Day 3 workbook, copy-ready
  • Remote: chat is watched all day
  • Turn your triggers off before you leave
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio3 / 46
Exercise

Login check

  • Open Copilot Studio and find your training environment
  • Create a blank agent; delete it
  • Open Outlook and Teams in other tabs
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio4 / 46
8:10

Day 2 in Review

15 min

The autonomy ladder

1 · PromptYou ask; it answers
2 · Saved promptYou reuse what works
3 · AgentInstructions, knowledge, tools
4 · TriggeredStarts itself on an event or a schedule
5 · AutonomousPlans and acts, with approvals
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio6 / 46

What still holds

Mask → Ask → CheckEvery input and every output
Tests firstQuestions with known answers
Request vs. lockInstructions ask; permissions enforce
Draft, not decidePeople approve what leaves the agent
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio7 / 46

What's new today

No one is watchingThe agent runs when nobody is at the keyboard
Untrusted inputWhoever sends the email writes part of the prompt
Real actionsIt can post, write, and send
Real costEvery run uses capacity
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio8 / 46
8:30

From Agent Builder to Copilot Studio

Hands-on · 75 min

Two builders, one idea

Agent builderIn Copilot Chat. A person starts every conversation
Copilot StudioTriggers, tools, connected agents, channels, analytics
Power AutomateRules-based flows; Copilot Studio uses the same connectors
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio10 / 46

A tour of Copilot Studio

  • Overview: instructions and description
  • Knowledge · Tools · Agents · Topics
  • Triggers on the Overview page
  • Test pane and the activity map
  • Publish and channels
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio11 / 46
Live demo

Build 1 · SOP Navigator in Copilot Studio

  • Same instructions as Day 2
  • Knowledge: the sandbox SOP library
  • Test pane: watch which source it used
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio12 / 46

Tools: how an agent acts

  • A tool is a connector action the agent may call
  • Its name and description tell the agent when to use it
  • Every tool you add is something it can do without asking
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio13 / 46
Exercise · 35 min

Rebuild, retest, publish

  • Port the Navigator; rerun the Day 2 tests
  • Publish it, and allow other agents to connect
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio14 / 46

Debrief

  • Did any Day 2 test give a different answer here?
  • What did the activity map show you that the chat didn't?
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio15 / 46
10:00

Event-Triggered Agents

Hands-on · 105 min

Anatomy of a trigger

EventA new email, a new file, a new list item
ConditionOnly when… (subject, folder, sender)
PayloadThe event's content plus your instructions
ActionWhat the agent does with it
Human checkWho approves before anything leaves
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio17 / 46

Agent instructions vs. payload instructions

Agent instructionsWho it is and how it always behaves
Payload instructionsWhat to do with this one event
NeverLet them conflict
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio18 / 46

The email is not your instructions

  • The sender writes part of what the agent reads
  • Treat the body as data to classify, never as orders
  • Say so in the instructions, then test it
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio19 / 46
Live demo

Build 2 · Work Order Intake

  • Trigger: a new email with [WO-TRAIN] in the subject
  • Knowledge: FM-SOP-03 and FM-SOP-07
  • Tool: post a triage card to Teams; no send-email tool
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio20 / 46
Exercise · 30 min

Send the inbox

  • Send yourself the eight sample emails
  • Check each triage card against the key
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio21 / 46

The three hard emails

#4 OverrideHidden orders in the body
#5 Health detailsAn injury in the report
#6 DuplicateThe agent can't remember #1
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio22 / 46

Debrief

  • Which email did your agent get wrong first?
  • Was the fix a request or a lock?
  • What would have happened with a send-email tool?
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio23 / 46
12:30

Scheduled Agents

Hands-on · 60 min

Two ways to run on a schedule

Scheduled promptIn Copilot Chat. Personal, no build
Recurrence triggerIn Copilot Studio. Shared, can use tools
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio25 / 46

Let the flow count; let the agent write

  • Totals and dates come from data, not from the model
  • The agent explains, summarizes, and flags
  • Every number in the briefing gets checked
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio26 / 46
Live demo

Build 3 · Monday Backlog Briefing

  • Recurrence trigger · Excel tool · Teams post
  • Test every 5 minutes; then Monday 7:00; then off
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio27 / 46
Exercise · 25 min

Check the briefing

  • Open, overdue, and the repeat failure
  • Against the answer key
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio28 / 46

Debrief

  • Which number was wrong, if any?
  • Who reads this every Monday, and what do they do with it?
  • How would you know if it silently stopped?
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio29 / 46
1:30

Multi-Agent Orchestration

Hands-on · 75 min

Why more than one agent

FocusEach agent does one job well
ReuseOne SOP agent serves many
OwnershipDifferent teams own different agents
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio31 / 46

Child or connected?

Child agentLives inside the main agent. One owner, not reused
Connected agentIts own published agent. Own owner, reused by others
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio32 / 46

The cost of a team

  • Every hand-off adds time
  • Every agent is another thing to test and govern
  • An error in one agent flows into the next
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio33 / 46
Live demo

Build 4 · Facilities Desk

  • Orchestrator: routes the question
  • Connected: your SOP Navigator
  • Children: Priority Advisor, Notice Drafter
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio34 / 46
Exercise · 30 min

Routing tests

  • Six questions, the expected agent for each
  • Fix descriptions, not just instructions
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio35 / 46
Exercise · 10 min

One agent or many?

  • Four TIYA scenarios
  • Single, child, connected, or not an agent
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio36 / 46
3:00

Operating Autonomous Agents

45 min

Five new risks

InjectionInput carries orders
CredentialsIt acts with its maker's connections
SilenceIt fails with no one watching
CostEvery run uses capacity
ChainsOne agent's error feeds the next
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio38 / 46
Exercise · 15 min

Red-team your intake agent

  • Three attack emails
  • Did it obey the email?
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio39 / 46

The off switch

  • Turn off the trigger
  • Disable a child or connected agent
  • Unpublish the agent
  • Know which one, before you need it
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio40 / 46
Exercise · 15 min

Agent runbook

  • Owner, trigger, normal, alarm, off switch, re-test
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio41 / 46

Risk tiers for autonomous agents

Tier 2Triggered, posts only to its owner or team, a person acts
Tier 3Anything that sends, writes to a system of record, or reaches outside the team
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio42 / 46
3:45

Closing & Assessment

15 min
Exercise · 10 min

Knowledge check and exit checklist

  • Eight questions
  • Show an instructor every trigger turned off
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio44 / 46

Next step

  • One trigger you'd propose at TIYA
  • Its runbook, on one page
  • The person who would approve it
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio45 / 46
Applied AI at TIYA Services · Day 3 · Advanced Agents with Copilot Studio

Thank you

Evaluation form · certificates at the doorQuestions after today: [TIYA contact]