Domain 1 of 3 · Chapter 17 of 22

Configure advanced agent responses with custom prompts

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing Designing and Building Integrated AI Agent Solutions in Copilot Studio premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

  • What a custom prompt is, and which surface owns what
  • Where the prompt runs, and who picks the moment
  • Writing the instruction: the task, the context, and a way out
  • Inputs: text, image or document, and Power Fx
  • Grounding a prompt in your own business data
  • The 100-second budget and what fills it
  • Choosing the model, and what a model choice commits you to
  • The settings panel: temperature, moderation, and code interpreter
  • Giving the response a shape downstream logic can bind to
  • Testing the prompt, then testing the agent around it
  • Where a rule belongs: the prompt, the agent, or a topic
  • How these choices show up in questions

Prompt model categories, as Microsoft groups them

ConsiderationMiniGeneralDeep
PerformanceGood for most tasksSuperior for complex tasksTrained for reasoning tasks
SpeedFaster processingMight be slower due to complexitySlower, as it reasons before responding
Published use casesSummarization, information tasks, image and document processingImage and document processing, advanced content creation tasksData analysis and reasoning tasks, image and document processing
Billing rate of the listed modelsBasic rateStandard ratePremium rate
Pick it whenThe task is bounded and latency or cost mattersThe task is complex or multimodal and accuracy matters more than speedThe task genuinely needs multi-step reasoning

Decision tree

Must the rule hold for every answer the agent gives? Agent instructions Overview page; governs every generated answer Yes No Must the wording be exact and unchanging? Topic with a Message node Fallback system topic for the default reply Yes No: one task, so a custom prompt Does it run inside automation, with nobody waiting? Run a prompt action in an agent flow fixed step of a business process Yes No Must it run at one authored step? Agent-level tool orchestrated call: the orchestrator picks it No Yes: an explicit call Will another topic need the same prompt? Agent-level tool, dynamic use cleared authored timing, reuse kept Yes No Topic-level tool reachable only by that topic
Cheat sheet

Cheat sheet

  • Add a custom prompt at the scope where it must run
  • Define prompt inputs for dynamic runtime context
  • Configure the prompt's model and generation settings deliberately
  • Test the custom prompt with sample data before integrating it
  • Place custom prompts at the required execution level
  • Test prompts with representative sample data
  • Write a prompt with a specific instruction and sufficient context
  • Include an explicit unsupported or not-found response
  • Configure prompt output formatting when downstream nodes need a contract
  • Use agent instructions for a rule that must apply to every generated response
  • Set temperature for consistency or variety

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. Prompts overview
  2. Prompt assistant
  3. Create a prompt
  4. Use prompts to make your agent or agent flow perform specific tasks
  5. Add tools to custom agents
  6. Prompts performance and execution
  7. Get started with prompt library
  8. Add inputs to your prompt
  9. FAQ for prompts FAQ
  10. Add knowledge to your prompt
  11. Change the model version and settings
  12. Write agent instructions
  13. Prompt model availability by region and updates
  14. Bring your own model for your prompts
  15. Use code interpreter in a prompt to generate and execute Python code
  16. JSON output
  17. Test your agent - Microsoft Copilot Studio
  18. Configure high-quality instructions for generative orchestration