Design ALM process for AI-powered business solutions: Review

ALM first step for a custom AI model: I picked Source Control, they wanted Data Preparation. I picked 'Source Control,' and the correct answer was 'Data Preparation.' That pretty much sums it up: a high-level, process-oriented slideshow, not a technical deep dive. The entire 45-minute module operates at this 30,000-foot altitude. It’s just a catalog of Microsoft product names mapped to ALM stages, with diagrams showing labeled boxes for source control, model registry, and CI/CD pipelines, but without a single line of YAML or a mention of a specific tool beyond 'Azure DevOps' or 'GitHub.'

Its main virtue is that the content is fresh. You learn the proper names for integrating Copilot Studio agents versus Foundry agents versus custom models. It distinguishes between ALM for AI in Dynamics 365 Finance and AI in Dynamics 365 Customer Service. This is valuable if your job is to sit in architecture review meetings and sound like you've read the latest press releases. It provides the taxonomy.

But for anyone who actually builds these systems, this is hollow. There's no discussion of model versioning strategies, managing large datasets in staging, A/B testing frameworks for Copilot plugins, or handling drift. It's a flowchart of what ALM *is*, not how to *do* it within the Microsoft ecosystem. I finished feeling like I’d read a well-organized table of contents for a book that doesn't exist yet.