There I said it, get out your pitchforks or cartoonishly large bags of money and let me have it.
I was, and to a large extent still am, an AI-skeptic. However, it has proven to be very useful numerous times and has helped my learning. Instead of needing to go scouring through dense documentation or finding a GeeksForGeeks page for proper syntax, I can just ask chatGPT and it will spit out not only the information I was looking for, but also relevant follow ups for further down the line. It severely softens and speeds up the learning curve.
Now don't think I was trying to be sneaky and cheat, the professor actively encouraged us to take advantage of LLMs.
Our senior year requires a project to help a company or organization in our community by building a piece of software for them. I, along with my 7 compatriots, are writing a web app for a volunteer organization.
Our team leader made a good decision and chose to use a pre-packaged starter kit. I just wish I had been more involved in the start of the term and voiced my preferences, because the whole stack is Typescript.
I am already not privy to Javascript, but that's more so to do with what it has done to the web. I don't have enough real experience with it or its statically typed counterpart to have a full-throated opinion on it. Nevertheless, working in this codebase feels completely alien, I don't know what I should or shouldn't touch or really how any of it all works.
What I should have done a month ago, when development started in earnest, was to sit down and spend 30 minutes familiarising myself with the language and the stack we are using. What I actually did was tell chatGPT what repo we were basing off of and asking it how to do what I wanted to get accomplished. In my defence, this was during a study date and I didn't want to tell my girlfriend that I needed to put my earbuds in and ignore her for a while. Undermining my defence, I still haven't taken the time to learn it.
I've just been asking the bot a question, fully intending to write the code myself, only to find that it provided a solution in its response. Of course the code still has issues, mostly around where things are placed in the repo, but after enough prompting, I get code that is working and I get to move my tasks to completed on the Jira board. And so has the rest of my group.
That brings us to tonight.
In my off time at work for the past few days I've been cracking away at getting BetterAuth working. I had the back-end portion set up, now it was just time to tie it in with the existing front end. I pasted the bot's code in the files it told me to and tried to get it up and running, but the dev environment kept on complaining about not being able to find a file. I went back and forth for half an hour asking it what I should do with the error message, each time it suggesting to make the file that the interpreter couldn't find. I knew this was wrong, after all this was one of the files I hadn't touched.
In my attempts to try and pinpoint the issue, I was peeking through the files and found all these things for connecting to the database. I looked at our repo history and my back-end partner hadn't touched them either.
So I went ahead and finally looked at the documentation.
I don't care to go into the specifics, but if I had continued doing what I was doing, We all would have to be fighting against the scaffolding already set up for us from the starter kit.
I knew chatGPT has a tendency to "forget" specifics about what you are trying to accomplish once you're a few prompts deep, but with it's instructions it's almost like it straight up ignored what I had told it about the codebase.
I just feel like a fool for not listening to my own advice, to try to fully understand what the code the bots generate is doing before ever using it.
I regularly check my email, If I don't respond quickly, send me a poke:
jasco.website@pm.me