Azure Weekly

Issue #554 - 8th March 2026

Highlights this week include: Introducing GPT-5.4 in Microsoft Foundry: OpenAI's GPT-5.4 is now generally available in Microsoft Foundry, bringing improved reasoning consistency, stronger instruction alignment, lower latency, and integrated computer-use capabilities for building reliable, production-grade agentic workflows at scale. The Great Foundry Shift: Microsoft Foundry New vs Classic Explained: A comprehensive layer-by-layer comparison of Microsoft Foundry (New) and Foundry (Classic), covering the move to a simplified CognitiveServices resource model, multi-agent workflow orchestration, long-term memory, a 1,400+ tool catalog with MCP/A2A support, and clear guidance on when to migrate. Even Simpler to Safely Execute AI-generated Code with Azure Container Apps Dynamic Sessions: Azure Container Apps Dynamic Sessions now exposes a built-in MCP endpoint, allowing AI agents to execute Python, Node.js, or shell code inside millisecond-startup Hyper-V-isolated sandboxes, with new samples demonstrating integration with the Microsoft Agent Framework.

The Future of Identity: Self-Service Account Recovery Preview in Microsoft Entra: Microsoft Entra's new Self-Service Account Recovery preview eliminates the vulnerable human help-desk step by verifying user identity through government-ID document forensics, biometric liveness checks, and face matching, then automatically issuing a Temporary Access Pass to bootstrap a new device without any agent involvement.

Azure Databricks Lakebase is now generally available: Azure Databricks Lakebase reaches GA, offering a serverless Postgres database that decouples compute from storage, integrates with Unity Catalog governance, and is purpose-built to support AI-native patterns such as real-time feature serving and agent memory directly on the lakehouse. Scaling API Ingestion with the Queue-of-Work Pattern: By decomposing large API ingestion jobs into thousands of small independent work items processed by concurrent Azure Container Apps workers via Azure Storage Queue, one team cut their data ingestion time from 15 hours to under 2 hours while gaining automatic retry handling and fault tolerance at lower cost than traditional orchestration tools.

Finally, two interesting posts about Drasi: Change-Driven Architecture on Azure with Drasi: A deep-dive into why polling-based event detection wastes compute and introduces latency, and how Drasi's Change Data Capture approach: running continuous Cypher queries over a live PostgreSQL stream, which delivers near-instantaneous event detection with declarative, version-controlled logic. How Netstar Streamlined Fleet Monitoring and Reduced Custom Integrations with Drasi: Fleet solutions provider Netstar replaced a sprawl of bespoke integration pipelines with a single Drasi-powered change-driven architecture, using continuous queries over Azure SQL and Event Hub to deliver real-time alerts and unified Grafana monitoring for logistics customers including Maersk.