Skip to main content
cancel
Showing results for 
Search instead for 
Did you mean: 

Fabric Ideas just got better! New features, better search, and direct team engagement. Learn more

Expose User Prompt and Agent Response Logs for Fabric Data Agents

Idea:
As an engineer responsible for managing and improving Microsoft Fabric Data Agents, I currently have no way to view how users are interacting with the agents I’ve built. While it’s likely that prompt and response data is being captured internally by Microsoft for telemetry purposes, there is no mechanism that makes this data available to those of us maintaining these solutions.

Proposal:
Introduce a feature within Fabric Data Agents that provides engineers with controlled access to interaction logs, including:

  • User-submitted prompts

  • Agent responses

  • Feedback indicators (e.g., thumbs up/down, optional comments)

  • Outcome metadata, such as whether the interaction was helpful or resulted in follow-up

Why this is critical:
Without access to this data, engineers are operating in the dark. We cannot evaluate how our agents are performing, where they are failing, or what users are actually trying to accomplish. This limits our ability to iterate, improve relevance, and ensure that Fabric Data Agents deliver real value.

Key Benefits:

  • Understand real-world usage and identify unmet needs

  • Improve agent behavior by refining instruction sets and context

  • Proactively address recurring failure patterns

  • Increase user trust and adoption by continuously tuning responses

  • Align agent capabilities with actual user workflows

Governance & Privacy:
Access should be controlled by tenant-level permissions and comply with Microsoft’s data governance policies. Logs can be anonymized and filtered to protect sensitive content, with visibility limited to approved roles (e.g., workspace admins or agent developers).

Summary:
Microsoft Fabric needs to provide engineers with visibility into how their Fabric Data Agents are being used. Access to prompt and response history, combined with user feedback, is essential to improving agent quality and delivering successful AI-driven experiences.

Status: New
Comments
OldDogNewTricks
Advocate II
I would expand this to Q & A as well as Copilot on Home (in preview). Ideally this would all be in one place.