I spent six weeks working intensively with AI to produce a fast, consolidated view of my two blogs which is hosted on my root domain uncountable.uk.
This is purely an experiment – I’m not ready to give up my wordpress blogs just yet, but I have been impressed with how much functionality I could achieve without writing a single line of code myself.
It doesn’t look all that great at the moment (July 2025), but I haven’t spent any time on the visual appearance. Instead, I wanted to create a functional view of my content that was as fast as it can possibly be.
If you browse to uncountable.uk you could, if you wish, scroll down all 2000+ individual pieces of content in one go (1,300 photos and 700 posts). It’s blistering fast. You can scroll as fast as you like on a phone, click on any item and see the full post displayed instantly inline.
You can also search equally fast. Try typing “Balsam” in the search box and you’ll see many articles I have written about that. This is particularly impressive because the entire site is static – there is no server processing search or any other request.
The searching is purely done on the client side – all 2,000 pieces of content are indexed and accessible instantly.
It’s fair to say that I would never have been able to produce this level of functionality on my own, no matter how long I allocated. I would have lost interest soon after a couple of weeks of minimal progress.
However, working with AI was enormously engaging, prolific and enjoyable. I was playing the role of product manager, project manager and architect. AI was happy to all the heavy lifting – writing code, documentation and tests according to the plans I gave it.
My own background is not a professional software engineer, however I have been writing code since 1981, and know enough about IT architecture, business requirements and software deployment to be able to adequately supervise it’s output.
The key observations that made this project successful for me are:
- heavy use of version control – feature development in a separate branch, and frequent commits between AI chunks of activity. Have AI write comprehensive commit messages for you.
- Having comprehensive test suites (generated by AI) right from the start. This ensures new functions don’t break old ones. Currently there are 740 automated tests in my application.
- Comprehensive documentation on how the application works. I’m not reading every line of code, but I do know how all the different parts interact. My documentation can be viewed online as part of the app.
- Have AI write a comprehensive plan of what it’s going to do for each feature enhancement. Store the plan as a markdown file in the project and get the AI to update it frequently during feature development.
- Documented standards for your project development. Mine covered the what good documentation looked like, how to write tests and software architecture principles. Ask the AI to check what it has done against these principles.
- Be vigilant. Many times AI would just go off in a convoluted loop, breaking out of your standards and writing spaghetti code. Roll back and start again if you see this happening.
- Keep an application change log and versioning. Only merge back to main when all tests pass, the application builds and documentation updates are complete.
- Use the premium models for anything even moderately complicated. My preference was Claud4. You can have unlimited requests on the free models, but in practice they could only accomplish very small, tightly specified tasks.
For me, this process was enjoyable because I feel I am in control of the code, it’s high quality and it does exactly what I want. Furthermore, if I want to extend it, I know how to specify in detail how and where the extension should be done.
The environment I used was VS Code with the Github copilot extension in Agent mode with a pro subscription of $10/month (with an additional budget of $10 premium requests beyond that). So far, we are 2/3 through July and I’ve used 90% of my free premium requests, so I’ll be dipping into the additional ten bucks budget for the rest of the month.
That last point is truly staggering. For twenty bucks a month, I have employed an extremely competent programmer who is happy to do whatever I ask, whenever I ask.
Whether you like it or not, the world has changed. Software creation is never going to be the same again.