Background and the reason for this blog

My name is Christiaan and I’m a software developer who is going to build a garden office with the help of AI. As a developer, AI has become almost impossible to imagine doing without, and it keeps getting more widely useful. During a ‘how’s life’ chat with a former colleague, it came up that he had planned his entire house move with AI. At that point I kept getting stuck on the planning of my garden office, and the penny dropped. That same evening I put my whole plan into a (git) project and carried on working on the garden office with AI.

So I had already started on the garden office, and everything up to and including the foundation was done entirely without AI.

The blog

The blog came about because not many people write up projects like this from start to finish, let alone attempt one with the help of AI. I think that with my background I can help people improve their own renovation/project management with AI. At the same time my background is absolutely not ideal for a DIY building project, so I’m going to make mistakes that others can learn from — or maybe knowledge and help will turn up from the most unexpected corners!

I write the blog entirely myself in Dutch and let AI do the translations for me. My writing style and grammar are far from good, so brace yourself. I also see it as practice in writing more narratively.

My background

Am I suited to a building project? I don’t think so. I grew up in a family full of trades. My father is an electrician and he has lots of brothers who have also mastered a trade. My practical CV consists of a summer tagging along with an uncle in ship joinery as a 15-year-old lad, two years as a trainee structural steel draughtsman after secondary school, and some ‘bodging and brazing’ — in other words, welding — at a carnival club.

After my management degree and some internship experience I moved into software development, so that’s where most of my experience lies.

AI

Since 2025, for quite a few software developers AI has grown from simple prompts in ChatGPT into agentic development with coding agents. In other words: where we used to ask a question about a piece of code, we now tell AI to build something with the whole project in its memory. From a bicycle bell to an entire bicycle in one go.

Agentic development is a term that really refers to a degree of control. If I simply ask AI to build a fitness app, it might do so in three different languages and for any old phone. Total randomness. Agentic development provides the instructions, tools, rules and limits to build what I want, the way I want it. In the end I’m responsible for the code, and if everything breaks I have to be able to fix it without AI too.

I use the word AI because it’s the accepted term. Ultimately it’s a Large Language Model (LLM) that calculates every word and every bit of code, and it is far from actual intelligence. To stay with the bicycle metaphor: it’s a bike that still needs stabilisers. At the time of writing (September 2026) I find the coding agents have been improving only bit by bit for nine months, and I notice little difference in my work. While I’m still paying for a Claude subscription now, I’ll probably switch to a local model soon. Local models always lag behind, but lagging behind no longer seems to matter much.

You can find more about the set-up, how I use it and the lessons learned in the posts from the foundation phase onwards.