As I’ve written about before, LLMs are an amazing tool for programming. In addition to IDE plugins / autocomplete integrations (Github CoPilot, Qodo) I’m finding utility in a pattern of ‘running conversation in ChatGPT in dedicated app.’ I can ask anything there, snippets and one-offs and what-ifs. The full -o1 model is really good.

Anyway, this is the current state – side project (open source here) is an iOS app in SwiftUI that will automatically capture context using Bluetooth iBeacons with the goal of generating a log that can become contact engineer billing records. The idea:

1. Enter region OFFICE at 8:00AM.
2. Leave region OFFICE at 6:00PM.
3. Enter region CAR at 2:15PM at GPS location LAT, LONG.

The GPS idea is for possibly trying to compute miles driven; may work or not. The regions will be marked using iBeacons. Cheap, supported by the operating system, easy.

I don’t really know SwiftUI, so I just asked ChatGPT for code and started hacking. It works and continues to work! LLMs are amazing for learning this way. I can ask random changes, explanations, fixing errors; anything! Here’s todays version of the app:

Damn.

If you’ve not tried coding with an LLM, go try it. Treat it like a freshly graduated student – you can ask it anything, and it’ll know the most amazing ways to write code and solve problems. It’ll be wrong yet confident sometimes, but as a way of getting unstuck and having fun hacking, I’m having a ball.