Replying to Avatar Silberengel

I make sketches on paper. Use case diagrams and state cycles and stuff. Really high-level. And then I type out user stories and then Gherkin scenarios.

And then I give the Gherkin to AI and have it code it and then I spend a week crawling through the code and straightening it out and torturing nostr:npub1wqfzz2p880wq0tumuae9lfwyhs8uz35xd0kr34zrvrwyh3kvrzuskcqsyn with code reviews because he's like Svelte AI. 😂

The end result, from the application logic, is very good, but the coding phase is a shitshow. Unless it's PHP. Only language I can just type out.

Most of the logic is already in the diagrams.

yeah, i find when i ask junie to write code to change things it often just piles on more crap instead of changing stuff. very often i wind up rewriting parts of the code and if it fails to actually debug shit and make tests pass, after 3 times i just usually end up writing the code that was needed by hand. and then, of course, the tests pass because it was written the right way.

they can sometimes save a lot of work but sometimes make more work. learning how to avoid those potholes they fall into is a key part of making the work more efficient.

Reply to this note

Please Login to reply.

Discussion

Half of it is frequent commits, so that you can quickly revert and start a new chat. Never spend more than 5 minutes on one solution.

yeah, absolutely, when i am writing the code, i might pack more into one commit, but generally best to avoid big commits. the bugs >_<... wasted quite a bit of time yesterday on big commits with new bugs or regressions.