What signs should managers look for to choose departmental champions for Copilot rollout pilot programs?

Best regards from Germany. I hope my questions are relevant to this topic and look forward to receiving answers.

1. How can I interact with an AI agent via a Power Automate cloud flow?
2. Will the limits for AI agents—currently a maximum of 20 agents and a 25 MB file size—be increased?
3. Why isn't there an easier way to interact with AI agents created in Copilot Studio via Power Automate? Currently, handling larger documents—where text and images need to be extracted—requires going through MS Azure and AI Builder.
4. In my opinion, the process for sharing AI agents with company members is currently very complex. Will this be simplified?
5. How can I get a Copilot AI agent to improve itself during the process?
6. What is the best way to save prompts and view the history of how a prompt has evolved? The initial prompt usually has room for improvement, after all. I’m missing a prompt archive feature—something like Apple’s Time Machine.
7. How can we effectively measure efficiency gains from AI support and credit consumption in order to gauge productivity gains and allocate credits sensibly?

8. Feature suggestion: Once a chat has resulted in a solution, a button could be used to condense the conversation into a single initial prompt that further optimizes the output. Otherwise, the LLM has to navigate through all the unstructured inputs and follow-up instructions. Currently, I eventually start a new chat using that optimized prompt.

Addition to Question 1: I would like the AI ​​agent to automatically optimize its knowledge base and initial instructions with each iteration to improve output quality. However, I also want the ability to disable this feature once a certain level of quality has been reached.

Many thanks.