The choice I didn't have before
It started as a recognisable task: digitizing a registration and follow-up process. This can be done with a Canvas App or SharePoint Site with Power Automate Flows behind it, both logical and both feasible.
Until the scope changed. The user group also appeared to be external. And then the math changes: licenses per unique user for something that is used occasionally, the costs add up quickly.
So I watched again. What really fits into the Power Platform here? Power Pages proved to be the right solution. Suitable for external users, cost-effective when used sporadically, fully integrated into the Power Platform ecosystem.
But the default user interface didn't match what I had in mind. Too generic. And then you are faced with a choice: do you accept the limitation, or do you bring in a developer?
This time, I chose neither.
What vibecoding is — and what it isn't
Vibecoding is building in flow. Do not plan and then execute, but start, test, discard, repeat. AI is not an end supplier in this regard. It is the voice that continuously thinks along as you make the decisions. What that means in practice: your iteration cycle becomes so short that errors are visible early and good ideas arrive quickly.
In practice, this meant: GitHub Copilot actively participated in Visual Studio Code. It wrote code proposals. It helped me to create an architecture. And it provided alternatives when I got stuck.
But it didn't send. That's what I did. I checked every proposal. Tested immediately. Thrown away what was technically correct but did not fit. Gathered feedback from fellow Fusion Developers. Wrote again. Each iteration was small enough to validate quickly. Human-in-the-loop is not a subordinate clause. It's the core of how vibecoding works.
What fusion development means for Power Platform consultants
Fusion development is the combination of low-code and full-code development within one solution and — increasingly — within one person. Low-code forms the foundation; full-code is used where it actually adds value. Thanks to AI, fusion development is no longer the preserve of developers with a full-stack background.
I built a custom PCF component as a Single Page Application. That component now provides full user interaction and forms the heart of the solution, seamlessly integrated into the Power Platform architecture.
A year ago, I couldn't have done this. Or don't dare to try.
The wider shift: role profiles are changing
Here's what fusion development changes in practice for Power Platform consultants:
A consultant who always stopped at the edge of low-code can now access full-code if the solution requires it. Not because he has suddenly become a full-stack developer, but because AI shortens the learning curve and iteration process so drastically that the boundary between profiles is less absolute.
This has consequences for how teams are put together, how projects are priced, and what you can expect from a consultant. At Blis Digital, we are increasingly seeing this in practice. In our own way of working and with the organizations we work with.
We are learning from each other; parts of the Power Pages DevKit were used for this project. Do you want to know more about this? Read the blogs by Thijs Geurts, Fusion Developer at Blis Digital: Microsoft Power Pages: a powerful platform with a lousy toolset
What this meant for the project
A working product was quickly found. Not a design document that has been rewritten for months, but a testable solution that could be validated immediately. The feedback loop became shorter. The risk is lower. The client's trust grew because something quickly arrived.
And it raises a question that we take seriously: if you can build so fast, how do you ensure the quality of what's running in production? That conversation, about AI Assurance and quality assurance of AI-supported solutions, is becoming increasingly relevant. More about that in a future blog.
FAQs
Can a Power Platform consultant with no programming experience really build full-code with AI?
Yes, provided that the iteration cycle is small enough and the consultant sharply validates what AI generates. Vibecoding makes this possible, but does not replace technical insight. You have to understand what you're building, even if you don't write it entirely yourself.
What is the minimum required to vibecode as a non-developer?
Technical baggage helps, but it is not the threshold. What you really need is a critical look at what AI generates. You have to be able to recognize mistakes without being able to write them yourself. That sounds contradictory, but it isn't. It's like an editor who isn't an author but knows when a text is wrong.
When do you choose fusion development over a dedicated developer?
If the solution is for the most part low-code and full-code only provides added value in specific areas. And if the consultant has sufficient technical insight to assess what AI generates. If the full-code component is extensive or mission critical, a dedicated developer or pair engineering approach is wiser.
