Microsoft AI

Creating Micro-Agents with Copilot M365

2 - 4 Hours 4 Modules

Overview

Main Theme: Microsoft AI / Copilot

Duration: 2 - 4 Hours

Level: Beginner

Target Audience: Managers, Team Leaders, Business Analysts, and HR/Finance Departments

ABOUT THE TRAINING

The standard Copilot is a powerful but general-purpose tool. This training teaches how to create Declarative Copilots (Micro-agents) that reside within Microsoft 365 and are specialists in your business. Instead of generic answers, your team will learn to configure agents that consult directly with the company's manuals, products, and rules, ensuring that AI speaks the organization's language.

Objective

Empower managers and professionals to independently create, without the need for programming (No-Code), intelligent assistants that solve specific departmental pain points, citing internal policies and adhering to the company's security guidelines.

TRAINING MODULES

Knowledge and Data Sources Ingestion

Learning how to connect the micro-agent to specific repositories. The student will learn to select SharePoint folders, internal sites, or external URLs for the agent to exclusively use these sources as 'absolute truth', eliminating answers based on generic public data.

Persona and Behavior Architecture

Advanced techniques for defining the agent's role. How to set up the Role (e.g., 'Senior Data Scientist') and the Task (e.g., 'Transform raw data into predictive models'), ensuring that the AI adopts the necessary posture and expertise for the assigned role.

Logical Guidelines and Tone of Voice

Implementation of conduct and reasoning rules. The student will learn to use the Chain of Thought technique so that the agent explains its strategy before delivering the result, in addition to defining restrictions and the appropriate tone of voice (Analytical, Direct, or Motivational) for each internal audience.

Validation Laboratory and Context Refinement

Practical exercise of 'Stress Test' of the agent. The challenge is to test the micro-agent against real-use cases, adjusting the instructions and data sources to eliminate hallucinations. The focus is to ensure that the agent responds with 100% accuracy based solely on the provided knowledge base, validating the reasoning logic before the final deployment.

Additional materials

Agent configuration

To create agents in Copilot M365, the interface fields should be filled in with technical precision in mind. The name should be functional and the description should summarize the agent's specialty. The instructions should contain a step-by-step of how the agent is meant to work. This is also where you turn on tools such as the code interpreter and the image generator.

Knowledge and publishing

Under knowledge you connect the agent to specific URLs, or restrict its search to trusted sources only. Finally, you can set up predefined titles and messages to guide the first-time user. That works as a strategic shortcut: with a single click, the user fires off complex commands.