I’ve been hands‑on with Generative Pages in Model‑Driven Power Apps. Short version: you get results fast. And when you already know Dataverse and the client APIs, they get really good, really quickly. Microsoft’s goal is simple, describe a page, pick your tables, and let AI generate React code plus the wiring. You will then refine in a conversation and ship. No Power Fx layer, but direct code generation. That matters for performance and iteration speed.
I put it to work on something fun: my Fantasy Football tracker. One table with all relevant NFL players. I wanted a richer list experience than a default view. Filters, quick access, and a detail peek without kicking the user out of flow.
So I built a Generative Page.
Full Generative Page of NFL Players
It took me minutes to get a decent page looking page. Especially when I asked it to look like a Model-Driven App, it immediately applied Fluint Ui. Then I nudged it with prompts and a bit of platform knowledge to make it just the way I want it.
We spent a good part of a decade turning citizen development into a program: portals for makers, office hours, Center of Excellence kits, and a polite line between business and IT. The Power Platform Community Conference blew a hole in that line. In Microsoft’s official “Build with Copilot” announcement, creation moved into the same canvas where people already work: Microsoft 365 Copilot. Building is no longer a destination. It is now embedded in your workspace, always within reach as part of your normal workflow.
Custom Agents aren’t “install and forget.” They’re not a shiny app you deploy and walk away from. They’re living systems that need care, context, and continuous improvement. We’ve already moved past the old waterfall playbook. With low-code and DevOps, we learned to ship faster, automate deployments, and iterate continuously. But here’s the catch: AI agents break even the DevOps mold. You will need an AI Agent Operating Model!
Why? Because agents don’t just run code. They reason, adapt, and depend on data and models that change outside your control. A pipeline alone won’t save you when a model update shifts behavior overnight or when your RAG index drifts. That’s why you need more than CI/CD. You need owners, guardrails, telemetry, and a cross-functional Agent Team that supervises, learns in production, and iterates quickly. The tech is moving weekly and your operating model must move with it.
In this post, I’ll share why the old playbook isn’t enough, it goes even further than DevOps. I will talk about a few principles that you need to govern and grow your agents safely.
Quite a few posts on my timeline are about honesty about what AI is. Apparantly too often the term AI Agents are used when they are basically automation in disguise. As I’m building and learning more about Agents with Copilot Studio I am often wondering what I am building. So here are some wondering thoughts about the subject.
Copilot Studio and Power Automate
And recently I read that no-one even knows what AI Agents are. I will not pretend I do know what they are, but I will add my 2 cents to the discussion. Especially since I love building AI Agents with Copilot Studio! Or am I building automation? AI Automation? I don’t know! So, let’s first try and get some definitions.
Definitions: Automate, Elevate, Delegate and Relegate?
Automation: Strictly speaking automation refers to the use of technology to perform tasks. It involves creating systems and processes that can operate independently, reducing the need for manual effort. When I automate, I do so with software development and to support business processes, to improve efficiency, quality and accuracy.
Facts and yes my definitions are opionized
AI Automation: Now AI Automation takes automation a step further by incorporating artificial intelligence (AI) into the process. This sounds simple, but it geniuinly pisses people off if you call automation which calls a bit of AI an Agent. I use the term AI automation when the task has a defined input and output and usually resolves around a singular tasks. And we have to use AI in our automation else it’s merely automation.
Agents: Agents are software entities that act on behalf of users to perform specific tasks. They can be simple bots that follow predefined rules or advanced AI agents that use large language models (LLM) to understand and respond to user needs. Agents can interact with users, gather information, and execute actions to achieve their goals. And in the near future, they might do so autonomously!
Mixing in Copilot Studio
For the few readers who do not know me, I’ve been playing with Agents for quite some time. Back in the day we called them chatbots, which we programmed each conversation step with the Bot Framework SDK. After a few iterations Power Virtual Agent saw the light, and we could use messages nodes. In that time we already used machine learning, natural language processing and entity recognition. Did we call them AI Agents? Most of us didn’t. We still referred to them as chatbots.
Back in my day
Only after the introduction of LLM we saw the hype of Agents and AI Agents. This is such a recent concept that Microsoft 2 years ago thought of Copilots, renamed my beloved Power Virtual Agent to Copilot Studio and now the original name might make more sense! Don’t get me wrong, I don’t mind the name changes at all, they make some sense most of the time. It’s just to illustrate how fast innovation and vision changes currently.
The thing is, how we build solutions is fundamentally changing. No longer will we have to rely on large “if, then, else” statements to eventually get to the result we want. However with Copilot Studio solutions we still use large areas of deterministic outcomes. Sometimes I question if I am building AI Automation or an Agent.
But does it matter what type of Agents we build?
But now that I’ve given it some thought. Does it actually matter how we call what agents we build? My answer? It depends! Yes I’m still a consultant. But let me answer the question from both sides, with a yes and no.
It Depends
Yes, it matters because sometimes we are better off building automation using classic automation tools because it’s cheaper than using LLM tools. Automation tools can be more cost-effective for certain tasks, providing a practical solution without the need for advanced AI capabilities. Automation tools can also be more robust as they will follow your entire logic chain.
No, it does not matter because it’s vital to start learning. Start building a small agent, learn, iterate, and continue. The key is to begin with simple agents, understand their performance, and what works well for your users to gradually improve them. This iterative approach allows you to refine your agents based on real-world feedback. Whether we use basic automation or advanced AI, the goal is to continuously enhance our agents to meet user needs effectively.
Another year, another release wave! Once again I dive into the documents and highlight my favorite features coming to the platform. Last time I did not do it and I have to say, I missed doing this. Diving into the release wave is something I do anyways. Making a blog post about it forces me to think a bit deeper about them as well. Another benefit is that I am sharing this with you, who hopefully enjoys reading my ramblings.
Features are rolled out to the Power Platform constantly now so the release waves mean less and less. However it is good practices to identify what’s coming, what’s relevant to you and what direction the platform is taking. For instance, this release wave mentions quite a bit about how Dataverse is preparing to scale what’s coming with agentification with everything build for Copilot Studio.
You can find the release plan on Microsoft Learn where you can also download the PDF or you can use the Release Planner. Often times features are already in preview and just finalized in this wave. In this article I will point towards the official documentation if available, else I will point to the Release Planner
Bigger features are also not announced through the release wave. They deserver their own bigger podium. However the Power App learn pages do mention 3 key investments. These investments definitely tickle my interest as one is quite new to me and the others are going to be incredibly powerful!
Continuing on with my series about building a HR Copilot with Copilot Studio. Previously I’ve been advocating of making a multiple choice option with prefilled options to steer the answer in the right directions. But with the rise of Generative AI we can look different at how we program our apps. We can now do things significantly different. In my last post we changed our manually authored topic to an action. We lost the ability to steer the user in the right direction by presenting a multiple choice option set. And instead of trying to fix this, I want to show you a different way. Let Generative AI figure out the meaning of the provided answer by the user. No complex (Azure) Open AI connections needed, we can use AI Builder for this!
Now before I dive deep how to set this up, I want to talk a little bit about the future of programming. More and more I see examples of how Generative AI can solve tasks which in the future would require a lot of programming. Tasks such as find the nearest shop based on your location. Previously this would be a tedious programming exercise. Now we can ask ChatGPT to do this for us, if we provide it with enough information. The challenge for me is, to think about solutions like the previous and my upcoming example. As my default still is to program it out in (low) code. But being creative with specific instructions to a large language model can often be a better fit.
We’ve been building our Copilot Studio topics manually in my example of the HR Copilot. But in this world of generative AI there is a another (better?) way. You can utilize plugin actions to create wrappers around connectors, Cloud Flows and skills. The AI will recognize the intent of the user and will pick the right action to fulfill that intent. If there are input parameters that needs to be filled in, the AI will query the user about this. In this blog we go through the steps in making our first action where a user can submit a leave request.
This post is part of a series of blogpost about Copilot Studio in which I create a HR Copilot and show various options available within Copilot Studio. Here are my other posts about this topic:
When building out topics manually for Power Virtual Agents we have all the control in the world. This means we can also present information in a way that makes sense and is appealing. In this blogpost I will continue on with my example topic of requesting days off. We will start by talking about how to make the request to our external system. After I’ll will utilize the ability to add rich media to our response to wish someone a happy holiday.
This post is part of a series of blogpost about Copilot Studio in which I create a HR Copilot and show various options available within Copilot Studio. Here are my other posts about this topic:
Continuing on with my series about Copilot Studio I want to show you something cool. In questions the Copilot asks the user we can presents a option set for the user to choose from. However this list is usually static, such as in my request a day off example. What if we HR adds a leave type option in our HR system? They will probably forget that they have to update the Copilot as well! Let me show you how we can add Dynamic Options to our Copilot.
This post is part of a series of blogpost about Copilot Studio in which I create a HR Copilot and show various options available within Copilot Studio. Here are my other posts about this topic:
Another year, another release wave! This year I once again dive into what I think are the most interesting features. Reading these features will give us some insights on where the product team is investing in. Obviously the bigger and most game changing features will be reserved for a larger stage, nonetheless we can find some cool stuff coming our way!
Now I have to be honest, I am getting less excited about the release waves. Thinking of why I am less excited, I have 2 thoughts that might cause this. One is good and for the other the jury is still out. The good thing is that the Power Platform is moving towards a continuous improvement model. I also highlighted this last release wave. Lot’s of features are rolling out that are silently added to the release plan website. In fact my first highlight is that the monthly channel is coming to general availability in April 2024. Nice thing about this, is that you can set overwrite this on an App by App basis as well!
Monthly Channel Selectable per App
The second reason is that development of all the various Copilots. They kinda suck the oxygen out of the room. The investment made into building, refactoring and improving these features leaves less money for other product features. Frankly, some of these first iterations of these Copilots are lackluster. I do think they will eventually be really helpful, but sometimes I imagine a world where investments are made into further convergence of Canvas and Model Apps. I do hope it’s not my age turning away my passion from the product 😉
By 
