Explicit Measures Podcast 504: Living in a Direct Lake Only World
Feb 27, 2026
A lively debate on whether Direct Lake replaces Import for Power BI and when each approach still matters. They unpack Direct Lake basics, V-ordering performance tricks, and how semantic models link to lakehouses. Licensing trade-offs and model size thresholds get scrutiny. The hosts also cover MCP’s role connecting LLMs to models and how AI could automate notebooks and refactoring.
AI Snips
Chapters
Transcript
Episode notes
Direct Lake Is A New Middle Ground
- Direct Lake is a new middle ground between Import and DirectQuery that auto-syncs semantic models with lakehouse tables for near real-time freshness.
- Tommy and Mike note it avoids repeated import compute by reading delta/Parquet files and using V-ordering for efficient access.
Monitor Fabric Capacities And Avoid Renaming Trials
- Monitor Fabric capacity and CU usage closely using the Microsoft Fabric Capacities app to catch runaway models and cost spikes.
- Mike warns renaming trial capacities can break metrics in the capacity app, so avoid renaming trials until fixed.
MCP Server Turned 3,000 Lines Into Model Docs
- Tommy and Mike describe using Model Context Protocol (MCP) agents to inspect and change Power BI models, generating documentation and bulk edits fast.
- Mike used an MCP to map 3,000 lines of external SQL/metrics view docs to a Power BI model and auto-generate column descriptions in minutes.