Fundamentals: Understanding Copilot Chat and Agents
- What agents are and how they work
- Capabilities and limitations of agents in Copilot Chat
- Typical practical mistakes—and how to avoid them
Agent Architecture
Key components of an agent at a glance:
- System instructions (role)
- Knowledge base
- Behavior and rules
Clear separation of tasks, role, and objectives
Creating Agents: From Idea to Implementation
- Creating a new agent
- Assigning a unique agent name
- Defining the specific use case
- Writing a clear and user-friendly agent description
Defining Roles and Objectives Clearly
- Defining functional roles (e.g., HR, sales, IT)
- Delimiting the scope of responsibilities
- Describing desired outcomes in a goal-oriented way
- Avoiding vague or contradictory instructions
Writing Professional System Instructions (System Prompts)
Structure of an effective system instruction:
- Role
- Tasks
- Desired behavior
Writing clear and unambiguous instructions
Using examples to guide agent responses
Common prompt design errors in practice
Controlling Response Behavior
- Defining tone (formal, neutral, casual)
- Setting the level of detail and response depth
- Controlling output structure
(bullet points, continuous text, tables) - Handling unclear or incomplete user requests
- Defining rules for follow-up questions by the agent
Rules, Constraints, and Safety
- Restricting the agent to defined topic areas
- Handling missing information
- Prohibiting speculation or assumptions
- Defining “non-responses”
Testing, Optimization, and Quality Assurance
- Testing with simple and complex queries
- Verifying goal achievement
- Evaluating response quality
- Iteratively improving system instructions
- Comparing different prompt versions
- Documenting changes and improvements
Management and Reuse
- Saving and organizing agents
- Reusing existing agents
- Maintaining oversight of multiple agents
