I started creating an application with the help of an AI Agent a few weeks ago. I told it what I wanted the app to do and it created code. I coded novel pieces of the application myself since there were no on-line examples available for what I wanted (this is the reason for creating the application in the first place).

Eventually I decided I should have some tests. This is a bit ironic since I am a professional software tester. I realized at that point adding features by adding tests was a powerful approach. The AI writes the test and runs the test. When it fails it updates the code to fix the failed test.

Sadly, some new features caused regression issues because my test suite was not very complete and because the AI changed the novel parts I so carefully crafted. Now I'm going back to create tests for everything that was implemented before the test suite was created. We'll see how it goes.

My app is here for the curious: https://github.com/erikshelley/complete-family-tree-viewer/