You write the query, run it, catch a missing condition, tweak the join, test again—repeat until it’s right. SSMS 21 brings Copilot into that loop.

TL;DR:

• SSMS 21 now includes built-in Copilot—no extensions. Just require connection to your Azure OpenAI deployment (like GPT-4o).
• Works only with SQL Server connections (not SSIS, SSAS, or SSRS).
• Runs in read-only mode—won’t execute any DML or modify data.
• Can generate queries, fix broken SQL, and suggest optimizations.
• Supports prompts like: “fix this query,” “optimize for performance,” or “generate insert script.”
• Ideal for SQL developers who want help writing, debugging, or learning better query patterns—without giving up control.

Installation: Copilot Comes Built-in

The installer for SSMS 21 clearly shows that Copilot is not an afterthought—it’s front and center. Once you launch the installer, you’ll notice the “Copilot Integration” checkbox enabled by default. This means no manual setup is required post-install. It’s part of the core experience now.

First Run: Setting Up Your Azure OpenAI Keys

Once SSMS launches, you’ll be prompted to connect Copilot to your Azure OpenAI setup.

In the first prompt, you’ll be asked to provide:

• Azure OpenAI service endpoint
• Azure OpenAI deployment
• Azure OpenAI API key

If you haven’t already created this in Azure, it’s a quick process: deploy a model, generate a key, and you’re good to go. we see how this information is pasted into SSMS.

It’s a one-time step, and once entered, Copilot starts responding instantly within SSMS.

To get an Azure OpenAI deployment name for SSMS Copilit, we need Gen AI model. That’s the engine powering your Copilot. GPT-4o is optimized for speed and real-time interaction, making it well-suited for something as dynamic as SQL writing and analysis.

If you already use GPT-based models in Azure, you can reuse the same deployment. If not, create a new one and pick GPT-4o or other depending on availability.

Which Connections Are Supported?

As of now, Copilot in SSMS works only with SQL Server connections—whether it’s a local database or Azure SQL.

It does not support:

• SQL Server Integration Services (SSIS)
• SQL Server Analysis Services (SSAS)
• SQL Server Reporting Services (SSRS)

So if you’re connected to an SSIS or SSAS instance, you won’t see Copilot suggestions. It’s focused purely on Transact-SQL (T-SQL) scenarios for now.

Read-Only Mode – No Live Data Changes

Copilot works in a read-only mode inside SSMS for GOOD.

It will not execute updates, deletes, or any data-modifying queries on its own. It doesn’t have permissions to alter databases or run DML commands. You remain in full control.

That said, it can generate update, insert, and delete queries for you, which you can review, tweak, and run manually. This helps reduce the risk of accidental changes while still speeding up development.

Quick Demo: Fixing a Broken Query

I pasted a malformed SELECT query into SSMS. Copilot immediately offered a fix and even explained what was wrong.

Few Prompt Examples: Result Accuracy

In below example, we asked Copilot to generate a complex query involving filters, aggregates, and joins. The result was accurate and ready to run with minimal edits.

the same query was further optimized by Copilot after being prompted with: “Make it efficient.” It introduced indexing suggestions and added a WITH (NOLOCK) hint—something even experienced developers can forget.

In another example, we asked Copilot another question on a different query context (another database connection), and it just worked fine.

These examples show that Copilot isn’t just copying boilerplate SQL—it understands context and best practices.

Should You Use This?

If your daily work involves writing or reading SQL, then yes—this is probably the most useful update to SSMS in years.

Here’s what it helps with:

• Writing long queries faster
• Understanding joins and filters
• Fixing syntax issues
• Learning better SQL patterns
• Generating stored procedure templates
• Drafting DML scripts (but not running them automatically)

What It Won’t Do (Yet)

Let’s be clear:

• It won’t connect to SSIS/SSAS/SSRS
• It won’t execute anything that modifies data
• It won’t automatically run DDL/ DML code it suggests

This is intentional, for safety and audit reasons. Think of it as a smart assistant—not a trigger-happy automation engine.

A Note on Privacy and Control

Copilot integrates directly with your Azure environment, so you control what it accesses and shares. It processes your prompts and data securely within your subscription. According to Microsoft, it doesn’t store any query logs outside your environment, which helps meet compliance requirements for enterprise users.

SSMS 21 doesn’t just add Copilot—it integrates it naturally into the flow of writing, editing, and reviewing SQL. It’s fast, context-aware, and doesn’t get in the way.

Pro tips:
1. While it’s good to have Copilot like features in SSMS 21, it seems to have lost some very important features like Import Export Wizard for databases like Azure SQL.
2. Learn how to build data agent in Microsoft Fabric.

See more

The post SSMS 21 Copilot: Here’s How It Works appeared first on AzureOps.

In this post, I’ll walk you through how I built a Fabric Data Agent on top of the standard AdventureWorksDW dataset, and how you can too, even if you’re a complete beginner.

TL;DR

Microsoft Fabric’s Data Agent lets you ask questions about your data in natural language; no SQL, DAX, or visuals required. Think of it like a chat-based analyst that already knows your data model and never sleeps. This article walks you through building a Data Agent on top of AdventureWorksDW, writing agent instructions, adding data sources, and exploring advanced use cases like AI Foundry and multi-agent systems. By the end, you’ll be able to create your own interactive AI analyst in under 30 minutes.

Prerequisites:
1. You must have access to Microsoft Fabric Capacity specifically F2 or higher.
2. Only workspaces with Fabric-enabled capacity will show the “AI Agent” option in the “+ New” menu.
3.Your Fabric admin must enable the Data Agent preview in the Tenant Settings within the Microsoft Fabric Admin Portal. if you don’t have admin rights, contact your admin.

How to Create a Data Agent in Microsoft Fabric

If you’ve never done this before, don’t worry. Here’s a simple step-by-step guide.

Step 1: Open the Microsoft Fabric Workspace

1. Go to https://app.fabric.microsoft.com.
2. Select or create a new Workspace where your Lakehouse is located.

Step 2: Create a Lakehouse (if not already done)

1. Click New > Lakehouse and give it a name (e.g., demolakehouse).
2. Add your data using Shortcuts:
   • Go to the Tables pane inside the Lakehouse.
   • Click Add Shortcut > OneLake data.
   • Browse and select the relevant tables (e.g., from AdventureWorksDW).

I used the classic AdventureWorksDW dataset, the same one you’ve probably seen in BI demos. Instead of importing tables manually, I used OneLake Shortcuts, which connect existing data without duplicating it.

Step 3: Create the Data Agent

1. In your workspace, click New > Data agent (preview).
2. Choose the Lakehouse where your data is stored. You can also choose semantic model in case you have created semantic model on top of your lakehouse.
3. Give your agent a name (e.g., Sales Insights Agent).
4. Click Create — this opens the agent configuration screen.

Step 4: Add a Data Source to the Data Agent

Once the agent is created, you’ll need to give it instructions and connect it to your data.

1. Inside the agent editor, go to the Data Sources tab.
2. Click + Add data source and choose the Lakehouse as shown in the image below.

Step 5: Add the Agent Instructions

1. Go to the Agent Instructions tab.
2. Provide instuctions for the data agent about it’s goal and responsibilities.

Goal:
Assist users in analyzing sales, customers, geography, and product performance.

Key Relationships:
- FactInternetSales ➡ ProductKey ➡ DimProduct ➡ Subcategory ➡ Category
- FactInternetSales ➡ CustomerKey ➡ DimCustomer ➡ Geography
- FactInternetSales ➡ OrderDateKey ➡ DimDate

Ask me things like:
- Top-selling categories last year?
- Sales trend in Q1 2024?
- Which cities had highest average order size?

Answer in charts, tables, or summaries where helpful.

Step 6: Add the Data Source Instructions

1. Add the data source instruction to let data agent understand your data model.

yamlCopyEditname: AdventureWorksDW_Shortcut
type: Lakehouse
tables:
  - name: DimCustomer_shortcut
    description: Customer demographics and keys
  - name: DimDate_shortcut
    description: Calendar date dimension
  - name: DimProduct_shortcut
    description: Product attributes
  - name: DimProductCategory_shortcut
    description: Categories of products
  - name: DimProductSubcategory_shortcut
    description: Subcategories of products
  - name: DimGeography_shortcut
    description: Country, state, and city details
  - name: FactInternetSales_shortcut
    description: Sales facts including sales amount, quantity, and dates

Step 7: Test the Agent and fine tunning

Start asking questions to the agent like;

1. List top 10 cities by internet sales in last available 12 months.

2. List top 5 customers who made the highest purchases in the latest available year:

Notice that it also provides the SQL query it has executed on the lakehouse to get the required output.

If the data source and instructions are set correctly, the agent should respond with meaningful tables or summaries. In case you observe issues in the responses, you may need to fine tune the AI instructions and test again.

When Should You Use Data Agents?

If you’re already using:

• A well-modeled star schema
• Lakehouse tables or Delta format
• Power BI reports or semantic models
• Microsoft Fabric workspaces with clean relationships

Then a Data Agent will act like a bridge between your data and your business users. No technical barrier — just questions and answers.

Integrations and Advanced Use Cases

What makes Fabric Data Agents truly powerful is not just that they can answer questions — but that they can plug into larger AI and data ecosystems. Here are a few ways to take your agent to the next level.

1. Integration with Azure AI Foundry

Microsoft Azure’s AI Foundry allows you to orchestrate LLM-driven pipelines, combine ML models with data operations, and deploy custom copilots. Your Data Agent can be used here in multiple ways:

• As a ready knowledge base within Foundry apps
• To summarize and explain insights from streaming or batch pipelines
• As a modular AI skill within a larger orchestrated flow

For example, you could use the agent to automatically generate weekly sales summaries and send them via email to business users — no dashboards needed.

2. Use in Multi-Agent Systems with Copilot Studio

Microsoft recently announced integration of Fabric data agent with Copilot Studio. This means data agent also supports multi-agent patterns where different agents work together, each specializing in their role.

Your Data Agent can act as:

• A domain expert (e.g., Sales Agent, Inventory Agent)
• A knowledge base that other agents query
• A collaborative AI alongside others like a document reader, chatbot, or metric generator

Example: Multi-Agent Use Case

Imagine a scenario where you’ve built:

• A “Sales Insights Agent” (based on the Sales model)
• A “Customer Sentiment Agent” (that reads survey data or feedback documents)
• A “Decision Support Agent” (that aggregates insights from both)

Now, a user asks the Decision Support Agent:
“Should we invest more in red-colored bikes for Q3 promotions?”

Here’s what happens behind the scenes:

1. The Sales Agent looks up sales trends for red bikes from your Data Agent.
2. The Sentiment Agent checks recent customer feedback about product quality or color preferences.
3. The Decision Agent combines both and generates a recommendation — backed by data and context.

This makes your Data Agent a building block in a broader AI ecosystem, not just a standalone tool.

Final Thoughts

Data Agents in Fabric aren’t just experimental toys; they’re fully capable assistants for real business scenarios. Whether you’re a data engineer, analyst, or decision maker, you’ll find value in being able to converse with your data.

For me, this agent worked perfectly with the AdventureWorksDW model and started answering meaningful business questions within minutes, no DAX, no Power BI visuals, no joins to write.

This service is currently in preview. It does not currently render charts. Hopefully, future updates will enable it to generate Power BI level visuals on the fly.

See more

Pro tips:
1. Learn how to integrate Microsoft Copilot in SQL Server Management Studio 21.

The post Data Agent in Microsoft Fabric – Here’s How it Works appeared first on AzureOps.