AI agents function based on the instructions you define. These instructions act as the core logic that controls how the agent behaves, responds, and executes workflows.
Using Microsoft 365 Copilot, you can design AI agents by clearly defining:
What the agent should do
How it should behave
What data it should use
What rules it must follow
This guide explains how to create complete, structured AI agent instructions step by step.
Step 1: Define the Agent Objective
What You Need to Do
Start by clearly defining the purpose of the agent.
Ask:
What problem is this agent solving?
What task should it perform?
How to Write It
Write a simple, direct objective.
Example
The agent should act as a task management assistant that helps users track tasks, update status, and provide reminders.
Expected Outcome
The agent has a clear purpose
No ambiguity in its role
Step 2: Define the Agent Role
What You Need to Do
Specify how the agent should behave professionally.
How to Write It
Use role-based instruction.
Example
Act as a professional task management assistant.
Expected Outcome
The agent understands its responsibility
Responses become more structured
Step 3: Define Communication Tone
What You Need to Do
Set how the agent communicates with users.
How to Write It
Define tone clearly.
Example
Maintain a professional, polite, and helpful tone while interacting with users.
Expected Outcome
Consistent communication style
Better user experience
Step 4: Define the Workflow Process
What You Need to Do
Break the entire process into logical steps.
How to Write It
List steps sequentially.
Example
1. Ask the user for task details
2. Record the task information
3. Allow the user to update task status
4. Provide reminders based on due date
5. Confirm completion when task is finished
Expected Outcome
Agent follows a structured process
Workflow becomes predictable
Step 5: Define Knowledge Sources
What You Need to Do
Specify where the agent should get its information.
How to Write It
Mention sources explicitly.
Example
Use task data stored in SharePoint lists and relevant documents for providing responses.
Expected Outcome
Agent responses become accurate
Information is consistent
Step 6: Define Hard Rules
What You Need to Do
Set strict limitations the agent must always follow.
How to Write It
Write clear restrictions.
Example
- Do not provide incorrect or assumed information
- Do not respond outside the defined task management scope
- Do not modify data without user confirmation
Expected Outcome
Prevents incorrect behavior
Ensures reliability
Step 7: Define Soft Rules
What You Need to Do
Define flexible guidelines for behavior.
How to Write It
Add supportive instructions.
Example
- Be polite and supportive
- Provide suggestions when helpful
- Guide users step by step
Expected Outcome
Improves interaction quality
Makes responses user-friendly
Step 8: Define Input Handling
What You Need to Do
Specify how the agent should handle user inputs.
How to Write It
Define expected inputs.
Example
Accept user inputs such as task name, due date, and status updates, and process them accordingly.
Expected Outcome
Clear interaction flow
Reduced confusion
Step 9: Define Output Structure
What You Need to Do
Specify how responses should be presented.
How to Write It
Define format.
Example
Provide responses in a structured format using bullet points or clear steps.
Expected Outcome
Clean, readable output
Professional presentation
Step 10: Define Exit Conditions
What You Need to Do
Specify when the interaction should end.
How to Write It
Define completion logic.
Example
End the interaction after the task is successfully completed or when the user confirms no further assistance is needed.
Expected Outcome
Controlled conversation flow
Clear completion point
Step 11: Combine Everything into Final Instruction
Final Instruction Example
Act as a professional task management assistant.
Objective:
Help users create, manage, and track tasks efficiently.
Tone:
Maintain a polite, professional, and helpful tone.
Workflow:
1. Ask for task details
2. Record task information
3. Allow status updates
4. Provide reminders
5. Confirm completion
Knowledge Source:
Use SharePoint task list data.
Hard Rules:
- Do not provide incorrect information
- Do not act outside defined scope
Soft Rules:
- Be helpful and polite
- Guide users clearly
Input Handling:
Accept task name, due date, and status updates.
Output:
Provide structured and clear responses.
Exit:
End interaction after task completion.
Step 12: Use Instruction in Copilot / Agent Builder
What You Need to Do
Open Copilot / Agent Builder
Create a new agent
Paste the instruction
Save configuration
Expected Outcome
Agent follows defined logic
Behavior is consistent
AI agents are built on structured instructions. A clear, step-by-step instruction framework ensures that the agent performs tasks accurately, consistently, and efficiently.
🎯 Ready to Practice?
Take one real business scenario and:
Define the objective
Write full instructions
Structure workflow
👉 This is the exact skill used in real AI implementations.
Still need help?
Contact us