Audience: analytics engineers, BI architects, and platform teams building on Microsoft Fabric and Power BI · Reading time: ~8 min TL;DR At scale, manual analytics work does not add up - it m...
Updated 1 month ago
Version 1.0
Blog Post
harry-arce This is the blueprint we’ve been waiting for. "Agents design at the edges; a deterministic core guards the center"
The "one service layer, two doors" pattern - REST for humans, MCP for agents - is exactly how you avoid logic drift at scale. Love that the contract is just a plain spec that’s Git-diffable. That makes review + CI so much easier.
The opt-in, read-only by default + approval-gated writes posture is spot on for a system of record. And putting BPA/DAX/Accessibility checks in the deterministic auditor before any publish is how you make AI safe in production.
Quick question: In the reference implementation, how do you handle conflicts when both a human and an agent propose changes to the same spec before publish? Is the operation_id + audit trail enough to serialize it, or is there a lock mechanism?
Thanks for open-sourcing fabric-autopilot. Cloning this to test the semantic-model-authoring skill this week.