Jose’s Platform Engineering Notebook

Azure App Service doesn’t get much love. It’s not shiny. It’s not trendy. It doesn’t give you the satisfaction of saying “we’re fully containerized.” And yet, it’s where some of our most reliable production workloads live. After running App Services, Functions, and Container Apps side by side, I’ve reached a conclusion that feels almost unpopular in 2025: Boring infrastructure is often the best infrastructure. The Problem With Exciting Compute When teams evaluate Azure compute options, the conversation usually starts with features: ...

AI agents are getting good at browsing the web. But they still interact with most sites by scraping HTML and guessing at structure. That is not great for anyone. WebMCP changes that. It lets websites register tools that AI agents can call directly, the same way MCP works with Claude or other AI tools, except it runs in the browser instead of a local server. ...

Most networking issues in Azure are actually DNS issues. Here is why DNS is hard in Azure, and how we handle it. Public and Private DNS Do Not Play Well Together Azure resources have public endpoints by default. They resolve via public DNS. When you add a private endpoint, the resource gets a private IP. Now you have two IPs for the same resource: public IP (internet-facing) private IP (VNET-only) DNS must resolve to the private IP from within your VNET, and the public IP from outside. ...

Soft delete saved us once. Purge protection blocked us twice. Here is what we learned about Key Vault deletion safeguards. Soft Delete Is Enabled by Default Now Soft delete used to be optional. Now it is mandatory for all new Key Vaults. When you delete a vault, it is not really deleted. It is soft-deleted. It stays in a deleted state for 90 days by default. You can recover it during that time. ...

We thought more vaults meant better security. It just meant more complexity. Here is what we should have done instead. The Logic Seemed Sound Each app gets its own Key Vault. Perfect isolation. Clear ownership. No shared access. It sounded like best practice. We created: one vault per microservice one vault per environment one vault per team Within six months, we had over 100 Key Vaults. Management Became Impossible Every vault needed: ...

Because mistakes scale faster. Infrastructure mistakes scale with traffic. AI mistakes scale with usage and trust. That is faster. AI Errors Feel Plausible AI failures are dangerous because they often look reasonable. The output is coherent. The tone is confident. The result is wrong. That combination spreads errors quietly. When infrastructure fails, it fails loudly. Services return 500 errors. Dashboards turn red. Alerts fire. The failure is obvious and immediate. ...

At first, Terraform was just a way to create resources. It lived next to the platform. It supported it. It automated it. Then, quietly, Terraform became the platform. The Shift Was Subtle There was no announcement. No rewrite. No big migration moment. It happened when: new environments required Terraform first access flowed through Terraform definitions changes without Terraform felt unsafe platform discussions started with code Terraform stopped being an implementation detail. It became the source of truth. ...

Guardrails over guesses. We are deliberate about how we use AI. We do not ask it to invent solutions. We ask it to follow patterns. Patterns Encode Experience Patterns exist because something worked. They represent lessons learned, mistakes avoided, and decisions made. AI is very good at following rules. It is less good at choosing them. Patterns are distilled experience. A retry pattern with exponential backoff exists because someone learned the hard way that constant retries overwhelm systems. A specific logging format exists because it makes debugging easier. A particular testing structure exists because it catches common mistakes. ...

Enforcing consistency costs almost nothing now. Structure used to be expensive. Templates took time. Reviews took energy. Enforcement took social capital. AI changed that. Structure Is No Longer a Bottleneck We can now ask for: the same format the same sections the same tone the same constraints Every time. Without friction. That makes structure cheap. Before AI, enforcing structure was a people problem. You could create templates and style guides, but getting engineers to follow them required constant effort. ...

Boring wins beat flashy demos. The biggest wins from AI were not dramatic. They were boring. Structure and Consistency AI helped most with: consistent formatting predictable layouts standard language repeatable patterns Things humans are bad at doing repeatedly. That alone improved quality. Humans are inconsistent at boring tasks. We get tired. We forget details. We take shortcuts when deadlines loom. We apply standards differently depending on mood or urgency. AI does not have these problems. It applies rules mechanically and consistently. If you tell it to format all documentation with the same structure, it does. Every time. Without fatigue or shortcuts. ...