Azure Weekly

Issue #555 - 15th March 2026

A good mixture of official announcements and community content this week. In AI: With its latest Phi-4 reasoning model, Microsoft reckons bigger isn’t always better: Microsoft’s new Phi-4-Reasoning-Vision-15B model shows that compact, carefully trained multimodal models can rival much larger systems by prioritising high-quality reasoning data over raw scale, and ships with three switchable thinking modes so developers can tune compute to the task at hand. A closer look at Work IQ: Work IQ is the intelligence layer inside Microsoft 365 Copilot that personalises responses by combining secure access to your organisation’s M365 and Dynamics 365 data, a contextual memory built from chat history and activity patterns, and a growing library of specialised skills and tools, with a public API preview landing this month. Build a Fully Offline RAG App with Foundry Local: No Cloud Required: A hands-on walkthrough for building a fully air-gapped Retrieval-Augmented Generation support agent using Microsoft Foundry Local, Node.js, and SQLite; no cloud, GPU, or embedding model required, with the same OpenAI-compatible API that makes the code trivially portable to Azure when connectivity is available. From Prompt Engineering to AI Programming: Building Enterprise-Ready Generative AI Solutions: argues that closing the gap between an impressive AI demo and a reliable production system requires applying the same engineering disciplines used elsewhere: testable components, labelled evaluation datasets, systematic metrics, and observability — to LLM integrations rather than treating them as magic black boxes.

In Infrastructure: The Startup-Scale Landing Zone: Get Azure right from day one: Introduces the open-source Startup-Scale Landing Zone (SSLZ), an opinionated Bicep/Terraform template that gives early-stage startups production-ready Azure governance, RBAC, security defaults, and budget alerts in under an hour, with an explicit migration path to the full Azure Landing Zone when the team and workloads eventually outgrow it. Designing and Implementing a Tagging Strategy for Azure Update Manager: Presents a lean, policy-enforced tagging taxonomy for Azure Update Manager that uses dynamic scopes to automatically assign newly provisioned VMs and Arc-enabled servers to the correct patch schedule the moment the right tag is applied. Stop Burning Money in Azure Storage: A practical cost-optimisation guide that explains when to use each of Azure Blob Storage’s five access tiers, compares the trade-offs between Smart Tier and Lifecycle Management policies, and highlights the early-deletion penalties and small-file transaction costs that quietly inflate storage bills.

Finally, Tracing Packets in Azure Networks: A comprehensive troubleshooting toolkit for Azure network engineers, walking through firewall logs, Virtual Network Flow Logs, PaaS diagnostics settings, Network Watcher features, and on-demand packet capture — including a practical tip on uploading Wireshark JSON exports to an AI for rapid packet analysis. It’s Not Always Azure: A real-world troubleshooting case study that traces an Azure VM RDP failure through NSG checks, platform redeployment, and guest OS diagnostics before identifying the root cause as a third-party security product silently re-enabling and misconfiguring Windows Firewall after an update. Implementing the Backend-for-Frontend BFF / Curated API Pattern Using Azure API Management: Shows how Azure API Management’s policy engine can serve as a lightweight BFF layer by fanning out parallel calls to multiple backend services and returning a single, client-optimised composite response — without writing or deploying any additional aggregation code.

If you're interested in all things Microsoft Fabric - don't forget to sign up for our new newsletter - Fabric Weekly - which we'll start publishing in the next month or so. We'll be moving all Fabric content over from Azure Weekly to Fabric Weekly, just as we did with Power BI Weekly 7 years ago.