Exam AB-100 Topic 1 Question 60 Discussion

Explanation:
Custom knowledge sources # Must be uploaded to the agent; AI general knowledge # Must be disabled for the agent The requirement says the solution must minimize the risk of generating inaccurate responses . In generative AI business solutions, the best way to reduce hallucinations or loosely grounded answers is to rely on approved, scoped, domain-specific knowledge rather than broad general knowledge.
That means the agent should use:
* custom knowledge sources that are intentionally provided to the agent
* AI general knowledge disabled , so the agent does not answer from broad open-ended model knowledge when enterprise-specific guidance is needed In finance and operations scenarios, this is especially important because users may ask about:
* internal business processes
* financial controls
* operational procedures
* policy-specific guidance
* company-configured ERP workflows
These are areas where inaccurate answers can create operational or compliance risk. Grounding the agent on uploaded custom sources provides a more controlled knowledge base.
Why custom knowledge sources must be uploaded
If the goal is in-app help and guidance for Dynamics 365 finance and operations apps, the safest and most reliable approach is to provide the agent with curated documentation such as:
* internal process guides
* approved SOPs
* finance workflow documentation
* policy documents
* ERP-specific instructions
Uploading these custom knowledge sources ensures the agent answers from enterprise-approved material rather than from generic model knowledge.
From an AI governance perspective, this improves:
* factual grounding
* relevance
* auditability
* trustworthiness
* alignment to company-specific processes
Why AI general knowledge must be disabled
General AI knowledge can be useful in broad assistant scenarios, but in a finance and operations help context it increases the risk that the agent may generate:
* overly generic responses
* answers not aligned to company policy
* incorrect interpretations of internal process steps
* content that sounds plausible but is not operationally correct
Because the question explicitly says to minimize the risk of inaccurate responses , the safer design is to disable general knowledge and constrain responses to approved custom sources.