I ran an experiment after reading a couple of chapters from Alan Knox's book, AI Builders.
The Experiment: To keep track of the decisions the AI and I make in a log for a simple application.
To set up my project, I used the 'prompting for options' approach that Alan talks about to establish a strong foundation for my app.
I made around 20 decisions regarding scope, my data model, overall design, etc.. This gave me a great starting point to start building.
Once it was time to start implementing, I let off the reigns and let the AI use my 20 decisions to build the four screens out.
The Result: The AI made about 54 decisions without me.
I got a result that looked good overall. The app checked off the boxes in my original plan, but there were things implemented that I didn't agree to.
- 19 of those were substantive (search function, day-grouping, dark mode, undo delete, tag management, etc.)
- ~35 were smaller implementations (FAB style, field order, placeholder text, notification copy, data-layer storage details, etc.)
Alan brings up an excellent point that in building with AI, there is a significant number of decisions being made by the AI that we have no idea about. And these silent decisions slowly start to build technical debt.
Imagine if I hadn't made the first 20 decisions for the AI to go off of. Who knows what result I would've gotten.
The more informed decisions you make about your app, the better.
After the Initial Experiment
After seeing how many decisions get made without me knowing, I now take the 'decisions then log it' approach. I prompt for options, then I weigh the tradeoffs, then I decide, then I log that decision.
While building Today, I Learned, I also built Decision Explorer AI. It's a Claude skill that keeps track of all your decisions, their tradeoffs, and more.
I found this tool very valuable when I was adding new features because the tool references my existing decisions and reasonings to help me make the best choice for my app.
The image below shows a list of the decisions I accepted/made before building, while building the app, and after submitting the app for review to the App Store.
This image is the result of those decisions. I made a total of 103 decisions for version 1.0.
To begin tracking your app decisions, you can find the free Decision Explorer AI template on GitHub.


Top comments (0)