All Intune, AI and Azure Blog Posts
Every article on jannikreinhard.com, newest first. Practical notes, deep dives and field-tested ideas around Microsoft Intune, AI, automation, Azure and endpoint management.
Every team that builds AI agents ends up building the same thing: a retrieval pipeline that feeds company knowledge into the agent. And most of these pipelines are built again and again, per agent, per project. Foundry IQ is Microsoft's answer to this problem — a shared knowledge layer in Microsoft Foundry that any number of agents can plug into. In this blog post I explain how Foundry IQ works under the hood, how you create a knowledge base and wire it into an agent, and where the limits are. At the end you can decide if it should replace your custom RAG setup. The first Microsoft Foundry screen can feel like a wall of new words. Workspace, project, model and deployment appear almost at once, even though they do different jobs. If you are new to Foundry, start with this: a project is the concrete use case your team works on, a model is the capability you want to try, and a deployment is the named version that an app can call. The word workspace needs a little more care — I explain that below. You can create all three quickly. The useful part is making the first setup easy to understand when you return to it next week. In this post I show the small, practical structure I use before an experiment turns into an application dependency. The video below shows the practical click path in German. This article adds a beginner-friendly explanation around it, so you know what each choice means before you move on. In June 2026 (week of June 22), Microsoft changed something that can break your automation overnight. Multi Admin Approval is now also enforced on Microsoft Graph API calls made with app-only tokens. Until now, only interactive (delegated) admin actions were intercepted. From now on, service principals, PowerShell scripts and third-party tools that change protected resources like apps or scripts are intercepted too. If they don't handle the new approval flow, they fail with an HTTP 403 error. In this blog post I explain what changed, why your automation suddenly gets 403 errors, how the approval flow works for API calls, and how you can exclude your automation apps from the enforcement. An AI agent that works in the playground is maybe twenty percent of the job. The other eighty percent start when real users hit it and you need to answer questions like: why did this run take 40 seconds, which tool call failed, and did the answer quality drop since Tuesday? In this blog post I do a deep dive into observability for agents in Microsoft Foundry: tracing, monitoring dashboards, continuous evaluation on live traffic and alerting. This is the production counterpart to my post about evaluating AI agents in Microsoft Foundry — that one is about testing before you ship, this one is about watching after you ship. In this blog post I explain how I use a Microsoft Foundry Realtime voice agent to talk about information that is currently visible on my screen. I share an Azure page, ask a question with my voice, and receive a spoken answer without copying text into a chat. The useful part is the combination of both inputs. The agent can hear my question and see the screen I selected. I can then ask a follow-up question in the same conversation. The complete example is available in my Microsoft Foundry examples repository. Keeping third-party apps up to date is one of the most time-consuming tasks in app management. Until now we had to create a new app for every new version and connect it with a supersedence relationship. With the June 2026 service release (2606) this changes: Enterprise App Management auto-update is now generally available. In this blog post I explain what the feature does, how you can enable it, and where the classic Win32 supersedence is still the better choice. The Azure AI Inference SDK migration is now a real deadline, not a future cleanup task. Microsoft has deprecated the beta SDK and says it will retire on 26 August 2026. If an application still uses azure.ai.inference, ChatCompletionsClient, or a /models endpoint, I would plan the move to OpenAI v1 now. In this blog post I explain how I would find an old integration, move it to the stable OpenAI SDK, and test the result before the deadline. The goal is not only to make one request work. The goal is to know which deployment, endpoint and identity your production workload will use. Update 23.07.2026: Microsoft documents the Azure AI Inference beta SDK as deprecated and recommends the generally available OpenAI v1 API with a stable OpenAI SDK. The official migration guide is the source of truth for supported languages and endpoint details. Picking a model used to be a one-time decision: you chose GPT-4o, put it in the config and moved on. In 2026 the model catalog in Microsoft Foundry lists more than 1,900 models across OpenAI, Anthropic, xAI, DeepSeek, Meta and Mistral — and the honest answer to "which model?" is usually "it depends on the prompt". That is exactly the problem the model router solves. In this blog post I explain what the catalog looks like today, how the Foundry model router works, which knobs you have to control it, what it costs, and how I decide between router and fixed model. I already wrote a general take in AI models in 2026: what I would actually pick — this post is the Foundry-specific deep dive. The Microsoft Foundry June 2026 update is much bigger than the arrival of another model. It changes how we build, operate and distribute agents. Claude reached general availability. Agents can move into Microsoft 365 Copilot and Teams through one governed publishing path. Toolboxes, Routines, Memory and evaluation are becoming parts of the same operating model. I first want to go through the complete update and explain what each area changes. After that, I build something practical on top of it: an Intune Change Briefing that reviews a planned endpoint change with Claude in Microsoft Foundry. The goal is not to copy the release notes. The goal is to understand what we can now build that was difficult before. I use my Modern Dev Mgmt tenant for the screenshots and the live Azure checks. The screenshots were taken on 20 July 2026. The Python sample is syntax-checked with anthropic 0.79.0 and azure-identity 1.25.1. You still need your own Claude deployment before it can send a real request. AI agents in production are a security topic, not only an AI topic. An agent has an identity, network paths, data at rest and a behavior that you need to control. In this blog post I explain how I secure Microsoft Foundry end to end, from the IT admin perspective: network isolation, agent identities, RBAC, encryption and guardrails. This is the checklist I would hand to any team that wants to move a Foundry agent from playground to production. I think about it in four layers, and I will walk through them in this order. Many companies want to build AI agents right now, and the first question I get is always the same: Microsoft Foundry vs Copilot Studio — which one should we use? In this blog post I explain how both platforms work under the hood, compare models, hosting, pricing, and governance, and I am honest about quality — because that is where many Copilot Studio projects struggle. I also explain why, in the age of AI-assisted coding, the "full code" platform Foundry is often the faster way to a working agent. At the end you should know which platform fits your scenario — or if you should use both together. Note: Azure AI Foundry was renamed to Microsoft Foundry at Ignite 2025, and the new Foundry Agent Service reached general availability on March 16, 2026. In this post I always mean the new Microsoft Foundry. Advertisement: This post is the first episode of my unboxing series in partnership with Admin By Request. As always, the walkthrough, the opinions and the experiences are 100% my own. Welcome to the first episode of my unboxing series! In this series I look at the products of Admin By Request in detail — and I start with the product the company is best known for: Endpoint Privilege Management (EPM). In this blog post I explain what Admin By Request is, how the solution removes standing local admin rights without blocking people from doing their work, and what I would configure first. If you prefer watching over reading, the full video episode is here: https://youtu.be/KmS9_8aLL4A
Read Foundry IQ Deep Dive: Knowledge Bases for AI AgentsFoundry IQ Deep Dive: Knowledge Bases for AI Agents
Read Microsoft Foundry Setup: Project, Model and DeploymentMicrosoft Foundry Setup: Project, Model and Deployment
Read Intune Multi Admin Approval Now Enforced on Graph API CallsIntune Multi Admin Approval Now Enforced on Graph API Calls
Read Microsoft Foundry Observability: Trace AI Agents in ProductionMicrosoft Foundry Observability: Trace AI Agents in Production
Read Microsoft Foundry Realtime Voice Agent: Talk to Your ScreenMicrosoft Foundry Realtime Voice Agent: Talk to Your Screen
Read Enterprise App Management Auto-Update: Now Generally AvailableEnterprise App Management Auto-Update: Now Generally Available
Read Azure AI Inference SDK Migration: OpenAI v1 GuideAzure AI Inference SDK Migration: OpenAI v1 Guide
Read Microsoft Foundry Model Router and Catalog: A Deep DiveMicrosoft Foundry Model Router and Catalog: A Deep Dive
Read Microsoft Foundry June 2026: Build More Than a DemoMicrosoft Foundry June 2026: Build More Than a Demo
Read Secure Microsoft Foundry: Network, Identity and GuardrailsSecure Microsoft Foundry: Network, Identity and Guardrails
Read Microsoft Foundry vs Copilot Studio: Which Agent Platform?Microsoft Foundry vs Copilot Studio: Which Agent Platform?
Read Admin By Request Unboxed: My Full EPM WalkthroughAdmin By Request Unboxed: My Full EPM Walkthrough
Complete archiveAll articles by year
2026
2025
2024
2023
2022
2021
Instruction pages