Unstable Copilot 365 agent performance

Question

Hi, I created M365 agent to retrieve email based on my keyword prompts but i see many issues. Could you please help? Thanks vm!

1. Retrieval Consistency
   • Are there recommended practices to improve retrieval consistency when the same prompt need to submit multiple times. Agent can pull out correct email only after few rerun even prompts not changed. So it should not be due to configuration. What is the reason exactly? Search engine at backend not stable functioning?
   • Is there a way to reduce variation across sessions or users?
2. Indexing Delays
   • What are the expected indexing timelines for newly received emails before they become available to Copilot agents?
   • Are there methods to identify or monitor indexing status when troubleshooting retrieval issues?
     3. Agent Instruction Limitations
   • Are there plans to increase the current instruction size limit for Copilot agents?
   • Are there alternative approaches for maintaining larger retrieval rules, scenarios, or reference guidance outside the instruction field?
   • Are there recommended prompting strategies or configuration techniques to improve natural language retrieval success?