I haven’t written a line of code since January. Not because I couldn’t. I stopped because spec-driven development showed me I was spending time on the wrong part of the job. Today, it is easy to type “build me XYZ” and get code in seconds. I did that too. Sometimes the result looked great. Then I ran the same prompt again and got something different. Logic shifted. Edge cases disappeared. The output stopped being predictable. The turning point came when…
“The dangerous part starts when people try to build one agent for everything. Then you don’t have automation anymore – you have a very polite chaos machine.” Fellow MVP and friend Marcel Lehmann commented that on my last post. And yes – he was describing exactly what I had built at first. Until I started to split my agents and recombine them again with a new feature. This is the final part of my Teams ITSM Copilot series. You’ve seen…
Your Teams channel is busy. Messages come in constantly – questions, announcements, status updates, and a well known “I have a quick question…” that turns into a long thread. But how do you know which ones are actually incidents? In my previous post, I introduced a multi-agent ITSM system built with Copilot Studio. One agent qualifies incidents, another tracks them in ServiceNow. But before any of that happens, someone needs to watch the channels. That’s the job of the Incident…
I was deep into spec-driven development with GitHub’s Spec Kit. Writing detailed specifications. Describing what I wanted AI agents to build. And I was typing. A lot. My fingers became the bottleneck. Then I thought: why am I typing when I could be talking? I opened Copilot Chat and looked for a microphone. Nothing. Turns out there’s a small extension you need: “VS Code Speech for Copilot.” I installed it and the microphone appeared. But then I discovered something better. The…
Recently, I was creating an agent with the goal to identify potential incidents in Teams community of practice channels. I started building this agent to find problems such as “I need help” and create ServiceNow incidents from its information. Testing it meant feeding realistic support cases into the system. But here’s the problem: real support data contains sensitive information. Project names. Client issues. Internal struggles. You can’t demo that publicly. You can’t even use it safely in testing. This isn’t…
The Agents topic is everywhere and omnipresent. Everybody tells you; this is the next big thing. You need an army of agents that finish your work. But how do I build such agents? Should I go for pro-code or for low-code? Should I use AI Foundry or Copilot Studio. Furthermore, what is a good example of building an agent? This was also the billion dollar-question for me. Today, I will showcase a small example to you where I will build…
How do you usually use Generative AI? That’s a question I get asked quite often. Most of the time, my answer is something like this: “I use it to summarize texts. I also draft ideas together with the AI, and it helps me write content more efficiently.” But this is just one piece of the puzzle. In automation scenarios within the Power Platform, I also utilize AI for data correction. Today, I will explain how this process works. In detail,…
In my last blog post, I explained you why data preparation for Copilot Agents is necessary. In other words, why we must improve the raw data for RAG (Retrieval Augmented Generation). Today I will delve more into the technical details of the preparation step. Let’s discover together what I can do to improve my raw PDF documents. First, I will explain to you how I extract content as Markdown from these documents with Azure AI Document Intelligence. I’m implementing this…
Working with Azure Open AI models that integrate text and images is cool. Correctly, I’m talking about GPT-4o and GPT-4o-mini. It is amazing to see the results of adding visual content to LLM prompts. Furthermore, it’s so easy with Prompty to define and review these prompts. But there is sometimes the devil in the detail, when when it comes to integration. Yes, I’m talking about using images in Prompt flow from generated Python code. Today I will take you on…
Working with Generative AI is cool. As a user, you mostly see the chatbot as frontend of an AI application with amazing results produced by Large Language Models (LLMs). However, as a developer, I need more insights during the development phase. Specifically, I want to know what information is being passed to my LLM. That’s why I usually enable tracing for my AI applications, but how can I do this in Prompt flow? That is not so difficult. So let…