AI copilots for SQL and Power BI in practice

Writing a database query used to mean memorizing table names, joins, and filter syntax. Microsoft’s new GitHub Copilot features for the SQL Server extension in Visual Studio Code now pull in the structure of the connected database so the assistant can draft code against your actual schema instead of guessing from a blank prompt. (learn.microsoft.com) That schema is the map of the database: which tables exist, which columns live in each table, and how records connect. Microsoft’s code-generation guide says Copilot can generate context-aware Structured Query Language, or SQL, and object-relational mapping code that reflects the connected database’s structure inside Visual Studio Code. (learn.microsoft.com) Microsoft is also pushing a “smart query builder” that turns plain-English requests into SQL while checking the database context. Its quickstart says the assistant is designed to build “efficient, accurate, and secure” queries in either raw SQL or an application framework format. (learn.microsoft.com) The same pattern is showing up in Power BI, Microsoft’s dashboard tool. Microsoft’s current Copilot overview says users can prompt it to create report pages, summarize data, and work inside reports, while some experiences are already generally available and others remain in preview as of late March 2026. (learn.microsoft.com) Power BI has its own hard part: Data Analysis Expressions, or DAX, which is the formula language behind many metrics and calculations. Microsoft’s DAX Copilot documentation says Power BI Desktop can now help write DAX queries, and a February 3, 2025 product update says Copilot can use semantic model descriptions, synonyms, and sample column values to do it. (learn.microsoft.com) (powerbi.microsoft.com) A semantic model is the business dictionary that tells Power BI what “revenue,” “customer,” or “margin” actually mean. Microsoft says Copilot works better when that model already contains descriptions and metadata, which means the assistant is leaning on curated definitions rather than inventing them on the fly. (learn.microsoft.com) (powerbi.microsoft.com) That is the split in these demos. A trained analyst can save time if the assistant already knows the database layout or the semantic model, but an ungoverned prompt can still produce the wrong join, the wrong filter, or a metric that looks polished and is numerically off. (learn.microsoft.com 1) (learn.microsoft.com 2) Microsoft’s own Power BI prompt guidance quietly says the generated page might not be exactly what the user wants and should be treated as a starting point. That is a useful reality check: the software can draft the first version of a dashboard, but someone still has to inspect the visuals, filters, and measures before anyone uses it in a meeting. (learn.microsoft.com) The practical winners are likely to be teams that already did the boring work first. If a company has clean table names, documented relationships, access controls, and approved metric definitions, Copilot can act like a fast junior analyst with perfect recall and no fear of opening documentation. (learn.microsoft.com 1) (learn.microsoft.com 2) If a company did not do that work, the assistant mostly speeds up the production of plausible-looking mistakes. These tools are getting good at drafting SQL and assembling Power BI pages, but the part that scales is still the old part: a database people can trust and a metric layer that means the same thing every time it is asked. (learn.microsoft.com 1) (learn.microsoft.com 2)