What's New · Issue #008 · AI & Copilot · May 2026 · by Raushan Ranjan, MCT
Copilot DAX generation reaching GA is genuinely useful — I've tested it in training sessions with 500+ learners. Here's the honest MCT assessment: it works well for common patterns, fails on complex ones, and will never replace understanding DAX fundamentals.
🤖 What Copilot DAX Can Do
- Time intelligence — "Revenue same period last year" → generates correct SAMEPERIODLASTYEAR pattern
- Running totals — "Cumulative revenue from start of year" → TOTALYTD correctly
- Simple CALCULATE — "Revenue for product category Laptops only" → basic CALCULATE with FILTER
- Percentage of total — "Sales as % of all products" → DIVIDE with ALL() correctly
- Basic RANKX — "Rank products by revenue" → RANKX with ALL
⚠️ What Copilot DAX Cannot Do (Yet)
- Complex CALCULATE with multiple REMOVEFILTERS — often generates incorrect filter context
- Nested SUMX with RELATED — frequently wrong row context handling
- Dynamic RLS patterns — USERPRINCIPALNAME() combinations are unreliable
- Bidirectional relationship scenarios — ignores cross-filter direction constraints
- Custom time intelligence with non-standard fiscal calendars
Rule: Always validate Copilot DAX against known test values before using in production.
💡 MCT Recommendation
Use Copilot DAX as a starting point and learning tool, not a final answer. When training learners: ask Copilot to generate the measure, then explain why it works or why it's wrong. This builds intuition faster than writing from scratch. But never ship AI-generated DAX without validation — especially for financial or compliance reporting.
For PL-300: you still need to understand DAX deeply. The exam tests whether you know why a measure works, not whether you can prompt an AI to write it.