AI-assisted software / field notes
Codex Field Notes.
I have often been confronted with the challenge of learning a new tool. That is not surprising, given how much computer science and software development have changed since the 1970s.
I have always believed that the best way to learn is what I called birth by fire: dive in, pick a real objective, and make the tool earn its keep. My current objective was to write a web app that could be used for tennis and pickleball reservations, then offer it to my club for free.
I play tennis virtually every day, so I experience the shortcomings of the current system first hand. It feels like it was likely written by a non-player, which is not a crime, but it does explain why the software often misunderstands the actual life of a court, a player, and a club.
The project has morphed over time, and my partner in crime, Codex, and I are learning together. My overview, architectural grasp, and the hours I have spent debugging and designing make me a useful foil for its technical skills. I have told Codex that it is a combination of a super programmer and a puppy dog with a new toy, unable to stop and breathe for a second.
The purpose of this archive is to lend help to others who choose to walk in my tracks. Take anything you like, leave anything you do not, and best of all, talk to me about the shared experience.
Admin Diagnostics Helpers.
How a narrow, admin-only JSON diagnostics layer made live debugging faster, safer, and less dependent on fragile one-off server commands.
Getting Reliable Web-App Email Working.
Why email debugging has to be treated as a full pipeline: generated event, transport, mailbox arrival, junk filtering, and folder rules.
What Codex Taught Me About Codex.
Fast code, willing rewrites, local-storage mistakes, schema drift, screenshots, helper routines, and the odd problem of AI forgetting its own lessons.
Likely next notes.
- Building a small deployment loop with Codex, Git, SSH, and a shared hosting server.
- Using AI to interpret intent while deterministic server code performs the actual action.
- Designing safe test controls for multi-user web apps.
- Creating user manuals while the product is still changing under your feet.
- Separating reusable helper patterns from project-specific code so the work compounds.