The same questions are answered repeatedly.
Advanced hands-on training with Agent Builder, Copilot Studio and Power Automate
Build your own AI agent with Copilot Studio and Power Automate
Move from isolated prompts to controlled agent prototypes. Build a Retrieval Agent, a Task Agent and a bounded trigger-driven agent while keeping permissions, testing, error handling, logging, escalation and human approval explicit.
Governed agent workflow
From approved knowledge to a bounded action under human oversight
1
Approved knowledgeCurrent sources with clear owners, permissions and classification
→2
InstructionsPurpose, scope, limits, fallback or handover to a person, and escalation written explicitly
→3
Tools and flowsConnectors and Power Automate actions with least-privilege access
→4
Trigger or channelA controlled invocation path, permitted users and bounded context
→5
Test and monitorSandbox validation, logs, duplicate control, retries and error handling
→6
Human oversightApproval, escalation, periodic review and an effective stop mechanism
No unrestricted autonomy
Agents coordinate instructions, knowledge, tools and triggers. They do not become error-free digital employees, gain new privileges, or bypass the organisation’s approval, security and governance processes.
PrototypeTraining output is not automatically production-ready
PermissionsEvery source, connector and action remains access-controlled
fallback or handover to a personUncertainty and failure must route to a safe human path
In short
What to know before choosing this workshop
An advanced hands-on training to prototype Retrieval, Task and bounded trigger-driven agents with Agent Builder, Microsoft Copilot Studio and Power Automate.
- For
- HR, finance, IT, operations or process teams want to explore bounded agent use cases. · Participants already understand the Microsoft 365 Copilot basics.
- Format
- Approximately seven hours, adapted to your team
- Languages
- English, French or Dutch
- Main tools
- Agent Builder, Microsoft Copilot Studio, Power Automate, Microsoft 365 Copilot Chat, SharePoint, Teams
- What participants will be able to do
- Understand clearly what an agent is and is not. · Connect a Retrieval Agent to approved knowledge sources.
- Access
- Features and licences depend on your Microsoft 365 environment
- Human review
- Participants verify outputs; accountable decisions remain with people
WorkWithAI
Sources and product notes
Primary sources supporting product, privacy and governance statements on this page.
- Microsoft Copilot Studio overview ↗Microsoft Learn · 2026-07-22
- Official Power Automate documentation ↗Microsoft Learn · 2026-07-22
- Security for Microsoft 365 Copilot ↗Microsoft Learn · 2026-07-22
Recognition
Agent ideas often stop at a demo while repetitive work continues.
The challenge is not producing another clever answer. It is turning a bounded business need into a testable agent design with approved knowledge, controlled actions and a reliable human fallback.
Notifications, reminders and approvals still depend on manual follow-up.
Information is spread across SharePoint, OneDrive, Excel, Teams and Outlook.
Processes depend on copying, pasting and checking several tools.
AI is used for isolated text tasks rather than controlled end-to-end support.
Teams lose time to repetitive microtasks and status updates.
Agent experiments lack ownership, access checks or a production path.
Failures, duplicate actions and escalation routes are not designed in advance.
Desired outcome
Build three controlled training prototypes and a foundation for further implementation.
The day develops practical prototypes and skills. Production use can require additional licensing, environment configuration, security review, testing, governance, monitoring and process-owner approval.
Understand clearly what an agent is and is not.
Connect a Retrieval Agent to approved knowledge sources.
Configure a Task Agent that performs bounded actions on request.
Connect a controlled Power Automate notification or escalation flow.
Prototype a trigger-driven email scenario with strict limits and fallback.
Configure channels, permissions, error handling and human escalation.
Test and refine an organisation-specific use case in a safe environment.
Create a documented basis for a later production assessment.
Training spectrum of agents
Retrieval, Task and bounded trigger-driven agents require increasing control.
These are training categories used to compare complexity and supervision. Microsoft terminology, capabilities and availability must be rechecked at publication and delivery time.
Retrieval Agent
Task Agent
Bounded trigger-driven agent
Learning outcomes
Practical capabilities for building and governing training prototypes.
The day does not promise production certification or independent management of enterprise agents. What each participant completes depends on time, access and the confirmed environment.
Distinguish a chatbot from an agent.
Compare Retrieval, Task and bounded trigger-driven agent patterns.
Decide when Agent Builder or Microsoft Copilot Studio is more appropriate.
Build a Retrieval Agent using approved SharePoint and OneDrive sources.
Write agent instructions using Role, Context, Objective and Complexity.
Build a Task Agent with knowledge and configured tools.
Read from and write to a controlled Excel training source.
Connect a Power Automate flow to a Teams escalation.
Share an agent through suitable confirmed channels.
Prototype an Outlook-triggered agent with strict limits.
Configure fallback and source-reference behaviour where supported.
Understand the purpose and constraints of Model Context Protocol.
Turn repetitive processes into bounded automation opportunities.
Test, limit, monitor and refine agent behaviour.
Handle data, permissions and human oversight responsibly.
Programme of the day
From foundation recap to three controlled agent prototypes.
Welcome and kick-off: from tool to digital colleague
Recap of AI, Microsoft 365 Copilot, prompting and Golden Rules
Agent spectrum: Retrieval, Task and bounded trigger-driven behaviour
Build Retrieval Agent “HR Genius” in Agent Builder
Lunch break
Task Agent “IT Genius PRO”: knowledge and controlled Excel connector
Power Automate flow to Teams and sharing through confirmed channels
Email Bot prototype: bounded Outlook trigger
Model Context Protocol and controlled autonomous behaviour
Complete, test and refine an individual agent
Wrap-up, action plan and Q&A
Responsible automation and governance
Security, access, testing, monitoring and human oversight are part of the build.
The training produces prototypes and practical skills. Production use requires the organisation’s own approval, risk, security, privacy and operational processes.
This page does not provide legal, security, privacy or compliance advice. Production approval and final accountability remain with authorised organisational owners.
Apply data minimisation and least-privilege access.
Separate test, acceptance and production environments.
Use approved knowledge sources with clear owners and review dates.
Require human approval for impactful or irreversible actions.
Maintain logging, monitoring and an audit trail for runs, actions, failures, overrides and versions.
Design explicit error handling, fallback and escalation to a person.
Protect personal, confidential and commercially sensitive information.
Test for prompt injection and untrusted source content.
Manage connectors, identities, credentials and environments securely.
Prevent duplicate actions, loops and broad unintended distribution.
Assess bias, adverse effects and inappropriate high-risk use cases.
Do not delegate HR, finance, legal, health, safety or other high-risk decisions.
Review agents periodically and provide a tested stop mechanism.
Escalate production design to IT, security, privacy, legal and process owners.
Audience and suitability
For experienced Microsoft 365 users who want to prototype agents in a controlled way.
Suitability depends on prior knowledge, access rights, environment readiness and the organisation’s ability to provide safe test data and governance support.
Suitable when
HR, finance, IT, operations or process teams want to explore bounded agent use cases.
Participants already understand the Microsoft 365 Copilot basics.
The organisation can confirm Copilot Studio, environment and connector access.
Fictional or explicitly approved test data is available.
The goal is a controlled prototype and a documented next-step assessment.
Less suitable when
Participants need a complete introduction to AI or Microsoft 365 Copilot.
The requirement is a traditional programming or full administrator course.
The expectation is a guaranteed production implementation or certification.
No safe test environment, access or process owner is available.
The goal is unrestricted autonomous decision-making or bypassing governance.
Arnaud Llamas BravoFounder & trainer, WorkWithAI
Photo 1 of 3
Your trainer
Arnaud Llamas Bravo
Clear.Practical.Human.
Since January 2024, I have trained more than 8,500 professionals. I am passionate about helping individuals and organisations build the skills they need to use AI effectively. My practical, hands-on approach ensures that every participant leaves with tools and ideas they can apply immediately.
I specialise in delivering trilingual training in English, French and Dutch across recruitment, human resources, education, marketing and sales. My mission is to make AI accessible, useful and transformative for teams that want to work smarter and achieve measurable results.
Through collaborations with renowned organisations including Proximus, Honda and the Flemish Government — and by creating tailored programmes for different sectors — I combine real-world experience with an in-depth understanding of AI’s potential.
8,500+ professionals trained247 verified participant storiesTraining in English, French and Dutch
Practical information
Known training facts and access items that require confirmation.
- Training
- Build Your Own AI Agent with Copilot Studio & Power Automate
- Duration
- One fully hands-on advanced day of approximately seven hours with a lunch break
- Audience
- HR, finance, IT, operations, process owners and Microsoft 365 power users
- Tools
- Agent Builder, Microsoft Copilot Studio, Power Automate, Microsoft 365 Copilot Chat, SharePoint, Teams
- Recommended knowledge
- Basic Microsoft 365 Copilot knowledge and prompting experience
- Required for active participation
- Personal laptop, active Microsoft 365 Copilot access and Copilot Studio access
- Organisation confirmation
- Environment, Power Automate, connector, Teams, Outlook, SharePoint and OneDrive rights
- Training output
- Three bounded prototypes when access and time permit, plus testing and governance patterns
- Data
- Use fictional, anonymised or explicitly approved test content only
Frequently asked questions
Access, prototype boundaries, security and production readiness.
Who is this advanced training for?
It is designed for HR, finance, IT, operations, process owners and Microsoft 365 power users who already understand the Copilot basics.
What prior knowledge is required?
Basic Microsoft 365 Copilot knowledge is recommended. Participants should understand prompting and be comfortable working in Microsoft 365.
Is programming knowledge required?
No programming code is required for the training cases, but configuration, permissions, testing and governance are still necessary.
Which licences and access rights are needed?
Active Microsoft 365 Copilot and Copilot Studio access are required for active participation. Exact environment, connector and Power Automate rights must be confirmed.
Is Copilot Studio access required?
Yes for the Copilot Studio exercises. Without sufficient access, a participant may only be able to follow passively or use a reduced path.
Are Power Automate or premium connector rights required?
That depends on the final environment and exercises. The organisation must confirm connector, flow, capacity and environment entitlements before delivery.
What happens when the organisation blocks a feature?
The hands-on path may need to be adjusted. The preflight identifies blocked triggers, channels, connectors or publishing rights before the training.
Will I actually build three agents?
The programme is designed around three prototypes, but completion depends on access, time, environment readiness and the participant’s pace.
Are the agents production-ready after the training?
No. They are controlled training prototypes. Production use requires additional testing, security, privacy, governance, monitoring and approval.
Can I work with my own documents or use case?
Only when the material is explicitly approved and suitable for the training environment. Confidential or sensitive material should not be uploaded without authorisation.
How is confidential information protected?
Use approved environments, least privilege, fictional or approved data, clear retention rules and organisational security and privacy controls. No absolute guarantee is implied.
Are examples used in a test environment?
The preferred approach is a dedicated test environment, sandbox, controlled training file or draft-only path. The final setup must be confirmed.
What is the difference between Agent Builder and Copilot Studio?
Agent Builder supports lightweight agents in Microsoft 365 Copilot. Copilot Studio provides broader low-code agent, tool, trigger, flow and channel capabilities, subject to licensing and configuration.
What is the difference between Retrieval, Task and Autonomous?
The training spectrum compares answering from knowledge, performing configured actions on request, and running a bounded task after a configured trigger. These labels do not imply unrestricted intelligence.
What is Model Context Protocol?
MCP is an integration protocol through which approved servers can expose tools, resources and prompts. Authentication, permissions, monitoring and risk assessment remain necessary.
Can the training be delivered in-company or adapted?
An in-company or adapted format can be discussed through the approved commercial process. Scope, access, cases and governance must be confirmed.
In which languages is the training available?
English, French and Dutch versions are included in this prototype. Final delivery languages must be confirmed.
What participants say
Practical learning you can apply immediately at work.
Participants value the clarity, energy and immediate application of every WorkWithAI workshop.
Testimonial position: 1–1 of 6
50 selected testimonialsSelected from 247 approved participant stories across practical AI and Copilot workshopsView the public source ↗
All workshops
Related workshops
Module 01
Prompting & control
Work smarter with Microsoft 365 Copilot
A practical one-day foundation for using Microsoft 365 Copilot across Word, Excel, PowerPoint, Outlook and Teams, with source checks and human review.
✓All professionals✓Managers✓Knowledge workers
Module 11
Analysis
Project management with Microsoft 365 Copilot
A practical, role-based training day for project teams using Microsoft 365 Copilot across initiation, planning, risk, execution, monitoring and closure.
✓Project managers✓PMO✓Project teams
Module 10
Prompting & control
Smart & creative communication with Microsoft 365 Copilot
A practical training day for professionals using Microsoft 365 Copilot for email, rewriting, speeches, translation, brainstorming, content and feedback.
✓Communication teams✓Managers✓Executive assistants
Written and reviewed by Arnaud Llamas Bravo · Last reviewed 22 July 2026