Introduction

Happy Friday the 13th. John Savill also published content around Microsoft’s “IQ” branding: Work IQ, Foundry IQ, Fabric IQ. The short version: models are becoming commoditized, and what actually differentiates your AI agents is the enterprise knowledge, entity relationships, memory, evaluations, and safety controls you wrap around them. Microsoft’s IQ components are their answer to that—business context for your AI.

Two things caught my attention this week: Azure Monitor Pipelines (finally, a way to filter and reshape telemetry before it gets ingested and costs you money) and the Entra source of authority change going GA. That last one matters for any organization shifting their primary IDP from on-prem AD to Entra.

Azure Monitor Pipelines (preview)

This is about controlling what telemetry data actually gets ingested and stored.

Today’s problem: you output telemetry, it hits your sink (Log Analytics, etc.), and you pay for whatever lands there. So you make cost decisions to limit the signals you send, which directly limits your observability. It’s always a trade-off between visibility and cost.

Monitor Pipelines sit between your telemetry sources and the sink. Before data gets ingested, before you pay for it, you can filter, aggregate, and reshape it. If your output format isn’t compatible with the sink, the pipeline transforms it. If you’re sending more data than you need, you can reduce it to just the useful parts.

It uses KQL for transformations, so any KQL operator you’d use in Log Analytics works here. Pre-built templates cover common scenarios, but you can write whatever transformation logic you need.

This is the kind of feature that should pay for itself quickly. If you’re currently over-ingesting data because filtering wasn’t practical, worth looking at.

Azure Disk vaulted backups (private preview)

Azure Managed Disks have had snapshot capability for a while, giving you point-in-time views of the disk. The issue: those snapshots live in the same storage cluster that backs the managed disk itself.

For serious backup strategies, you want a copy away from the primary storage. Vaulted backups give you exactly that—an isolated vault with its own access controls, immutability settings, and escalation requirements. If an attacker compromises your primary environment, the vaulted backup has a completely different set of protections.

You also get geo-replication with vaulted backups, so you can restore to the paired region. Private preview for now, so not broadly available yet, but it’s coming.

AKS App Gateway for Containers add-on (preview)

App Gateway for Containers is the load balancing solution built specifically for AKS, not something designed for VMs that got bent to work with containers.

The new managed add-on lets you enable App Gateway for Containers with a single command. Works for new deployments or existing ones, uses managed identity, and simplifies the networking setup.

If you’re already running AKS and want a load balancer that actually fits the container model, this is the easiest path there.

AKS Kubernetes 1.34 support (GA)

AKS now supports Kubernetes 1.34, which brings a mix of stable, beta, and alpha features.

The big one on the stable side: dynamic resource allocation. This lets you share and allocate GPUs, TPUs, NICs, and other hardware resources across pods. The design was inspired by the dynamic provisioning work for storage volumes, now extended to other device types.

If you’re running GPU workloads on AKS, this matters. Dynamic resource allocation going stable means the API won’t change under you, and you can rely on it in production.

Azure Front Door Premium now supports private link origins in the UAE North region.

The value hasn’t changed: your origin servers behind Front Door don’t need to be publicly accessible. Front Door creates private endpoints in its own virtual network and uses private link to reach your origins. Traffic stays private, your origins stay off the public internet, and Front Door handles the proxying.

One more region covered. If you’re serving traffic in the Middle East with Front Door Premium, this closes a gap.

Claude Opus 4.6 on Azure Databricks

Claude Opus 4.6 is now available on Azure Databricks, joining the models already accessible through Azure AI Foundry, Copilot Studio, and GitHub Copilot.

Within Databricks, it’s served through Mosaic AI model serving—the unified way to deploy, govern, and query AI models. Exposes a REST API and joins other models from OpenAI, Meta, and others already available in the Databricks environment.

If you’re building AI workloads in Databricks and want to use Anthropic’s strongest model, you don’t need to leave the platform.

Azure Databricks Supervisor Agent (GA)

Building on the Agent Bricks platform that went GA a few weeks ago, Databricks now has a supervisor agent.

You give the supervisor agent an outcome, and it figures out how to get there. Think customer service automation: answering policy questions, looking up account details, searching research reports, analyzing usage data. It handles the orchestration, so you define what you want, not how to get it.

Not something you’d build from scratch. That’s the point.

Azure SQL Database: 1-4 secondaries in failover groups

Azure SQL Database failover groups now support one to four secondaries instead of just one.

Each secondary can be in the same or a different region, maintains its own direct geo-replication with the primary, and you can fail over to any of them. The read-only listener can point to one of the secondaries, but put it in a different region from the primary if you want it to actually help.

More secondaries, more options for how you lay out your failover strategy. Simple improvement, especially useful if you operate across multiple regions.

New Thailand South region

A new Thailand South region was announced.

New regions matter primarily for data sovereignty. As regulations get stricter about where data can be stored and processed, having a region within a specific boundary becomes a hard requirement, not a nice-to-have. Some organizations can’t use Azure until there’s a region that satisfies their compliance needs.

Also the latency benefit for workloads serving users in the area.

Kimi-K2.5 in Azure AI Foundry

Moonshot AI’s Kimi-K2.5 is now available in Azure AI Foundry. It’s a multimodal mixture-of-experts model that takes text, images, and video as input, with dedicated tokens for image and video understanding. Not an afterthought — visual input is core to how the model works.

You can feed it an image of a website and it’ll build it, use it for deep research, report generation, and more. One more model in the Foundry catalog to evaluate.

Entra source of authority user change (GA)

This one matters for organizations migrating their identity strategy.

Today, many organizations have Active Directory Domain Services on-prem as the source of truth, syncing users into Entra ID via Entra Cloud Sync or Entra Connect. The source of authority determines where user attributes are managed.

With this change going GA, you can flip the source of authority from on-prem AD to Entra ID. The synced account becomes a cloud-managed account, so you get Entra governance, lifecycle management, and the rest of the cloud identity tooling that doesn’t work when AD is still the authority.

Practical step for organizations making Entra their primary IDP. If you’ve been waiting for this to go GA before planning your migration, it’s time.

Final thoughts

Azure Monitor Pipelines is the update I’m most interested in. The cost vs. observability trade-off is something every team deals with, and being able to filter and transform data before ingestion is the right answer. KQL-based approach means no new query language to learn.

The Entra source of authority change doesn’t sound exciting, but it unblocks a lot. If your organization still treats on-prem AD as the source of truth, this is the clean path to making Entra primary.

And AKS getting Kubernetes 1.34 with stable dynamic resource allocation is good timing, given how many teams are running GPU workloads now. If you’re doing AI inference on AKS, check if DRA simplifies your resource management.

Sources

  1. John Savill, “Azure Update - 13th February 2026,” YouTube, https://www.youtube.com/watch?v=gFh0kpFrND8