AI-Assisted Greenfield Software Development, Part 12: Executing, Verifying, and Understanding the RegisterAcademic Slice

John Miller | 9/29/2026

John explains how the RegisterAcademic slice in the AIAGSD series turns reference data into a durable academic record while enforcing contract-first implementation, persistence ownership, and SQL Server-backed verification. He shows how the prompt acts as an execution plan for AI-assisted development: validating input, resolving rank, degree, university, and extension references through approved contracts, creating the academic aggregate atomically, and preventing concurrency issues such as double-claiming an extension. John's post also emphasizes traceability and testing across validators, handlers, routes, integration, and concurrency scenarios, framing RegisterAcademic as a delivery gate that must pass before downstream academic features can safely proceed.

AI-Assisted Greenfield Software Development, Part 11: How Reference Data Was Executed with AI Subagents

John Miller | 8/28/2026

John Miller explains how he used specialized AI subagents to implement two independent reference-data slices in parallel for Zeus Academia: `ManageUniversities` and `ProvisionExtension`. The post shows how narrowly scoped prompts, isolated worktrees, and explicit coordination rules enabled concurrent development while preserving ownership boundaries, canonical data rules, and testability. It also highlights the architectural issues that surfaced during execution—especially around application-host responsibility, persistence boundaries, and university identity—and describes the refactoring and verification steps used to resolve them before integrating the changes. The main takeaway is that AI-assisted development is most effective when prompts define clear slice boundaries, escalation paths, and evidence requirements, allowing teams to accelerate delivery without sacrificing control or consistency.

SQL Server 2025: AI With Your Data at No Ongoing Cost

Ginger Grant | 8/28/2026

Ginger Grant explains how SQL Server 2025 brings retrieval-augmented generation directly into the database, enabling organizations to ground AI on their own data entirely on-premises, without recurring cloud costs or exposing sensitive information outside the network. Vector search and embeddings solve a key limitation of traditional SQL by allowing semantically related terms—such as “quiet” and “low noise,” or “camera” and “webcam”—to be matched even when users do not know the exact schema or wording. Grant outlines a practical implementation using local tools like Ollama, Caddy, and SQL Server’s preview AI features, showing how to create embeddings, build vector indexes, and query them with natural-language similarity search. The overall message is that SQL Server 2025 makes enterprise AI more accessible, private, and cost-effective for managers and developers.

How to Validate Microsoft Fabric CI/CD Deployments Using SQL Server 2025 and SSMS 22

Ginger Grant | 7/31/2026

Ginger explains how SQL Server 2025 and SSMS 22 can help teams validate Microsoft Fabric CI/CD deployments by addressing a common gap in migration workflows: deployment tools move Fabric objects such as lakehouses, warehouses, and SQL databases between Dev, Test, and Production, but they do not automatically move the underlying tables, views, stored procedures, or metadata. She shows that by using SSMS 22 schema comparison through SQL endpoints, teams can compare environments, identify missing database elements, and generate scripts to recreate them as part of an automated release process.

AI-Assisted Greenfield Software Development, Part 10: ManageDegrees

John Miller | 7/27/2026

Part 10 of the AIAGSD series explains how the ManageDegrees implementation prompt turns a small reference-data slice into a reliable foundation for downstream academic workflows.

Cut Your Fabric MCP Costs: Why I Switched from GitHub Copilot to Claude Code

Ginger Grant | 7/1/2026

Since GitHub Copilot's June 1 pricing shift to metered tokens, Ginger compares VS Code and Claude Code for working with the Fabric MCP Server. She breaks down setup steps, pricing, model options, and Fabric Capacity CU usage, explaining why Claude Code came out ahead on cost and simplicity.

AI-Assisted Greenfield Software Development, Part 9: Manage Ranks

John Miller | 6/29/2026

ManageRanks is the first reference-data slice in the Zeus Academia build, and it sets the pattern every later slice will follow. The post walks through defining canonical rank codes, enforcing uniqueness at both the application and database layers and documenting the rank-to-access-level mapping that registration, access control and reporting will all depend on going forward.

Why Local LLMs Matter

Philipp Bauer | 6/3/2026

Do you use tools like Copilot, ChatGPT, or Claude.ai knowing your prompts may be training someone else's model? In this post, Philipp Bauer makes the case for running large language models on your own hardware: your data stays private, your workflow survives cloud outages and API deprecations, and your inference costs drop to zero. From a practical hardware guide covering Apple Silicon, Nvidia, and AMD GPUs, to a model landscape comparison of today's top open-weight contenders, Bauer gives developers everything they need to decide whether local LLMs belong in their stack.

AI-Assisted Greenfield Software Development, Part 8: Slice 1 (Durable Database Invariants)

John Miller | 6/1/2026

In this post, John shows how the first implementation wave in Zeus Academia forced a shift from code-only verification to durable schema enforcement, migration-centric evidence, and more explicit multi-agent handoffs.

MCP: The Open Standard Turning AI Chatbots Into AI Agents

Ginger Grant | 5/29/2026

Ginger Grant explains that Model Context Protocol (MCP) is the open standard enabling AI to move beyond chat and into agentic execution of real workflows. Rather than being an AI model itself, MCP provides a common way for LLMs, apps, and external systems such as GitHub, SQL Server, Microsoft Fabric, and Microsoft 365 to communicate through secure, standardized servers and tools. Grant highlights how MCP can streamline developer and analyst work by chaining tasks—such as creating lakehouses, querying databases, or generating pull requests—within familiar environments like VS Code or Copilot.

Plan: The Most Interesting Thing That Came Out of Microsoft Fabric Community Conference in April 2026

Ginger Grant | 4/24/2026

Microsoft unveiled **Plan** at FABCON — now the second-largest tech conference with 8,000 attendees — as a standout Fabric feature for budget and project planning. Born from a 2024 partnership with Lumel's EPM tool, Plan lets users import data from semantic models or Excel, apply machine learning to forecast outcomes, run what-if analyses, and validate projections against historical data. Available to all Fabric license holders, it offers a compelling reason to adopt the platform.

AI-Assisted Greenfield Software Development, Part 7: Implementation Prompts

John Miller | 4/23/2026

Planning without execution guidance leaves AI agents guessing. In Part 7 of this series, John Miller completes the planning phase for Zeus Academia by creating implementation prompts that translate architectural intent into disciplined, slice-by-slice coding instructions. Learn how execution plans, agent role definitions, and structured prompt templates give both AI tools and human contributors a shared delivery contract — and set the stage for real implementation work to begin.

AI-Assisted Greenfield Software Development, Part 6: Vertical Slices and Implementation Planning

John Miller | 3/31/2026

Turning architecture into working software requires more than good intentions — it requires a plan. In Part 6 of the AI-Assisted Greenfield Software Development series, John Miller teaches how to use meta-prompts and AI agents to generate the planning artifacts that make vertical-slice delivery predictable: custom-agent governance standards, product requirements documents (PRDs), workflow models, and dependency diagrams. The result is a structured, repeatable handoff from architectural decisions into implementation-ready slices.

Cloud Data Transformation in Fabric with Medallion Architecture

Ginger Grant | 2/27/2026

Ginger explores how medallion architecture addresses the limitations of legacy ETL (Extract, Transform, Load) processes to meet the demands of modern data science, AI, and analytical workloads. She highlights the brittleness of traditional ETL pipelines, which often resulted in lost data, poor governance, and untestable transformations, and contrasts this with the multi-layered medallion approach. The architecture organizes data into bronze (raw data), silver (validated and transformed data), and gold (analysis-ready data) layers, ensuring robust governance, scalability, and trustworthiness. Grant emphasizes the importance of testing processes, securing sensitive data, and collaborating with stakeholders to meet organizational needs.

AI-Assisted Greenfield Software Development, Part 5: Technology Guardrails

John Miller | 2/27/2026

In Part 5 of his series on AI-assisted greenfield software development, John delves into the creation of technology-specific instruction files that act as guardrails for AI agents generating code. These files ensure consistency, correctness, and alignment with architectural standards by explicitly defining coding conventions, implementation patterns, testing expectations, and compliance requirements for each technology in the stack. Using AI-generated prompts, John demonstrates how to systematically create instruction files tailored to technologies like Vue 3, ASP.NET Core, MediatR, FluentValidation, Pinia, and more, while maintaining coherence with the project’s overall architectural vision. He emphasizes the importance of keeping project documentation up-to-date and validating governance rules to prevent conflicts in AI-driven workflows. This approach provides a scalable, repeatable framework for leveraging AI in software development, enabling teams to deliver high-quality solutions efficiently.

Tool Installation Guide

This post provides a comprehensive guide for preparing tools required for CODE Training's AI-assisted Software Development class. You'll find step-by-step instructions for installing and setting up essential software, including GitHub, Visual Studio Code, the GitHub Copilot extension, GitHub Copilot CLI, and the Mob.sh mob programming tool, on both Windows and macOS platforms. CODE suggests that you have a GitHub account, preferably with a paid Copilot license, but free models can be used. Enjoy the class.

Do I Need a Database Anymore?

Ginger Grant | 2/2/2026

Are traditional databases necessary in 2026? Advances in storage compression, memory interfaces, and AI-powered retrieval have reduced reliance on costly databases that require extensive maintenance and tuning. Microsoft Fabric offers a transformative alternative, providing four flexible storage options that handle massive datasets using open-source Parquet files at lower costs than traditional cloud databases. Maybe your organization should consider adopting Fabric for future-ready, cost-effective data infrastructure.

What Happened to Power BI?

Ginger Grant | 12/29/2025

Ginger Grant explains that Power BI remains a healthy, actively developed product despite recent licensing changes and Microsoft's emphasis on Fabric; you can still use Power BI standalone, but Fabric provides compelling benefits for organizations that need scale, real‑time performance, and more modular development. By moving data engineering, security, and storage into Fabric’s OneLake and using features like Direct Lake, teams gain near‑instant, scalable reporting, better pipeline and notebook workflows, improved monitoring and validation, and easier collaboration than traditional Power Query approaches. In short, Fabric is not required, but it materially enhances Power BI for enterprise scenarios.

Microsoft Fabric: The New Method for Organizing Data in the Azure Cloud

Ginger Grant | 11/1/2025

Ginger Grant presents Microsoft Fabric as a unified, modern solution for organizing and centralizing organizational data in the Azure cloud (and beyond) with dramatically simplified setup and governance. She explains how Fabric replaces the traditional mix of disparate tools (Data Factory, Databricks, Data Lake, IoT/Event hubs, and databases) with a single, configurable resource that orchestrates data from multiple sources—including multi-cloud and Snowflake—through seamless referencing or mirroring rather than heavy data import. Fabric includes broad tool compatibility (PySpark, T-SQL, Power Query, notebooks) and a predictable fixed-hourly pricing model, plus a centralized storage layer called OneLake. Overall, Fabric addresses complexity, cost, and collaboration challenges by delivering an integrated, cross-provider data platform.

Self-Service Reporting: Setting Up Successful Reporting for Everyone

Ginger Grant | 10/7/2025

Ginger makes the case for giving business users the tools to create their own reports. When IT controls all data access, teams often resort to inefficient workarounds that create security risks and waste time. She shows how Microsoft Fabric and Power BI can solve this problem by giving users secure, well-governed access to the data they need. With the right setup, people can build their own reports without sacrificing data quality or creating chaos. The result? Less waiting on IT, more consistent data across the organization, and faster, smarter decisions.

The Strategic Case of Open-Source Data Engineering

Ginger Grant | 8/29/2025

Ginger explores how companies can break free from the expensive 10-year cycle of proprietary data transformation tools by embracing open-source technologies like Python and Spark, which Microsoft Fabric now fully supports through its adoption of vendor-agnostic approaches and Medallion Architecture.

When Database Alternatives Make Business Sense: A Microsoft Fabric Perspective

Ginger Grant | 7/31/2025

Ginger Grant examines the transition from traditional databases to Microsoft Fabric's analytics-focused architecture. Fabric uses Delta Parquet files in scalable lakehouses and warehouses, supporting familiar tools like T-SQL and Power BI while adding AI-powered natural language queries. While databases excel at fast transactions, Fabric offers cost-effective, petabyte-scale analytics with integrated AI tools, making it ideal for organizations prioritizing data analysis over transactional processing.