Designing effective agents
- By Lisa Crosbie
- 2/25/2026
Practice tasks
No practice files are necessary to complete the practice tasks in this chapter.
Choose the type of agent
Review the agent use case you developed in Chapter 1 and consider the following questions to decide which type of agent you should build.
What is the best user experience for this agent? Could this be an autonomous agent that acts in the background, or is it a conversational agent with which the user interacts directly?
How broad and varied is the domain knowledge or process that your agent needs to cover? Could it be more effective as an agent with multiple child agents handling specific parts of the process, or can a single agent handle everything required?
Select the agent’s orchestration model
Navigate to the Surface Support Guide agent you built in the first chapter and complete the following tasks.
In the test pane, ask the question: What is the difference between a 12-inch and 13-inch Surface Pro?
Review the answer given and the activity map on the left side of the screen, which shows how the agent reasoned over the input and which knowledge source it used to generate the response.
Select Overview from the top menu to return to the agent configuration screen.
Switch the generative orchestration toggle to Disabled and wait a couple of minutes for confirmation that the change has been saved. Your agent is now using classic orchestration.
Refresh the test pane.
Ask the same question in the test pane: What is the difference between a 12-inch and 13-inch Surface Pro?
Review the difference in the quality of the answer compared with the generative orchestration mode. Note that you don’t see the activity map in this mode.
Switch the generative orchestration back to Enabled.
Design the agent’s capabilities
Navigate to the Surface Support Guide you built in Chapter 1 and perform the following tasks.
Review the description of the agent. It describes the tone of voice of the agent as “friendly and enthusiastic.”
Refer to the Microsoft brand voice guidelines here: https://learn.microsoft.com/en-us/style-guide/brand-voice-above-all-simple-human.
Copy the section that describes the “three voice principles”—warm and relaxed, crisp and clear, and ready to lend a hand.
Go back to your agent and select the Edit button at the top of the Details section on the main configuration page.
Replace the section of the description that describes the tone of voice with the Microsoft voice principles copied in the previous steps.
Select the Save button to save the changes to the description.
Refresh the test pane and ask the question: What is the difference between a 12-inch and 13-inch Surface Pro?
Compare the tone of the response with the one you got at the beginning of the previous exercise. What differences do you notice?
Understand and calculate agent costs
Consider the Surface Support Guide agent you built in Chapter 1, which uses generative orchestration and web-grounded answers connected to the Microsoft Surface website.
Download the latest Power Platform licensing guide from https://aka.ms/powerplatformlicensingguide.
Go to the section in the document for Microsoft Copilot Studio. Your agent is using generative orchestration. Review the billing rates for web-grounded answers and agent actions (AI-led orchestration). Imagine your agent gets 1000 single-turn question and answer interactions each day.
Calculate the message consumption for 1000 web-grounded answers.
Calculate the AI-led orchestration message consumption for 1000 agent actions.
Add these numbers together. This is the total estimated number of messages your agent will consume for this scenario.
Visit the Copilot Studio agent usage estimator at https://microsoft.github.io/copilot-studio-estimator/.
Enter an estimate for the number of users and the number of times per month your users will interact with the agent, and review the estimated message consumption.
Make some changes and review the changes to the estimate.