I am working on a tiny-scale Google Drive clone, and adding pgsql support. I wanted some use cases for small database apps that people could make in a low/no-code environment.

Use cases could help delineate what a minimal feature set could be, and also a maximal capability expectation. I didn’t want this to be MS Access, for example.

Thaura generated some use cases, but it was SO ENTHUSIASTIC in its writing, it was annoying.

I asked it to analyze its work from different directions, and produce some reports. Over time, it appeared that a lot of this writing was coining new jargon.

So I asked it to reduce the enthusiasm, and look for existing jargon, and rewrite its work to reduce the text. It did exactly that.

Basically, I was telling it to use jargon like a library of code, to reduce code size. It acted like compression.

Then I asked it to look for common patterns in the use cases, and make a pattern catalog, coining new terms, and then replacing the repetitive parts with references to the pattern catalog.

The use cases, by this time, had gotten WEIRD:

Across all the use cases, parts had references to ETL terms, some parts to software engineering terms, some parts to the local pattern catalog, some parts to a jargon glossary.

It still kind of made sense to me, because I’d read the originals, and also threw in a lot of my own stories into the chat. A newcomer to all this writing would not understand.

Work, however, didn’t get confused at all, when I told it to make a plan to implement one use case, it did so.

Wow.

It’s good at PHP

I started to write the app to run on node.js, but the LLM was throwing so many errors when it generated JavaScript code, and it said these were due to escaping problems related JSON.

So I abandoned JavaScript and switched to PHP (which the original app was coded in). It seemed smoother.

Then, because it was in PHP, I asked it to develop an extension system similar to WordPress plugins. It was based on loading code, and hooking into events.

Plugins create a clearer boundary between code modules than using PHP classes or JavaScript modules.

The main problem is, it’s PHP, which is like a janky Java these days.

Elm

So I went looking for better languages. I wanted it to be Elm, but it didn’t seem to be the one for the LLM, but I may be wrong.

I’m really interested in Elm, because it’s a functional language, and, when I asked Thaura if it was just syntactic sugar over something like a Scheme language, it scolded me.

No, is said, it’s more like Haskell.

So, that’s good: like Haskell, Elm programs need to be complete. The program flow is based around resolving functions, so little of the code runs in sequence. The compilation step catches many of the errors.

Though Elm is hard to read, it’s one of those languages where things make sense once you ascend the learning curve. Haskell was like that for me. Elm programs seem to be short, because SPAs tend to be simple.

Even more exciting: elm-run.dev says a CLI and Server Elm exists and is in the pipeline!!!