Question 18 — AB-620 Case Study Set
Given that the system invokes the Foundry policy agent after selecting a channel-specific card solely to “Confirm card compatibility for the active channel,” what should Paideffort change to improve efficiency? (Select one answer)
Remove the agent invocation and use deterministic channel and schema compatibility logic before sending the card.
Keep the agent invocation and add a second model call to validate whether the first compatibility decision was correct.
Replace the channel conditions with a FunctionTool that asks the model which Adaptive Card schema to use for each request.
Send every card to the Azure OpenAI endpoint first so the model can normalize its schema before Copilot Studio renders it.
Correct answers
Remove the agent invocation and use deterministic channel and schema compatibility logic before sending the card.
Explanations
Correct Answer:
Remove the agent invocation and use deterministic channel and schema compatibility logic before sending the card.
Explanation:
Paideffort already knows each target channel's supported schema ceiling, making compatibility evaluation deterministic. Invoking the HR policy agent for this decision adds avoidable latency and model activity. The Foundry agent should remain focused on policy reasoning rather than a fixed compatibility lookup.
Wrong Answers:
Keep the agent invocation and add a second model call to validate whether the first compatibility decision was correct.
This compounds the inefficiency. Paideffort can determine the result from channel and schema metadata without any generative reasoning.
Replace the channel conditions with a FunctionTool that asks the model which Adaptive Card schema to use for each request.
A deterministic function could perform the lookup, but asking the model remains unnecessary. Paideffort already maintains explicit channel compatibility values.
Send every card to the Azure OpenAI endpoint first so the model can normalize its schema before Copilot Studio renders it.
The Azure OpenAI endpoint isn't required to perform a fixed schema-compatibility decision and would introduce additional latency, cost, and nondeterminism.