Good news/Bad news in the AI department for me right now. (I suppose some might choose to interpret it as Bad news/Bad news department, but don’t listen to them.)
I’ve decided to start development of my app over for the fifth time.
I imagine actual coders out there are chortling over this news (if they actually cared in the first place, I suppose). Each time I think I’ve got the app figured out, and that I understand how to make AI work for me and have everything function this time. And each time after I’m into the development phase a ways, I discover that I was actually wrong.
What went wrong this time? I’ve come to the stunning realization that AI can’t be trusted to do a good job with coding all on its own. Last time, I thought the solution to this was to make sure I had a ton of documentation written ahead of time that the AI would be able to follow, so I spent a bunch of time developing that. After trying to implement it for a few weeks, I’ve come to understand that all it takes is for AI to “forget” about part of that documentation one time, and it starts to go awry. Pretty soon it’s duplicating efforts and then trying to merge those efforts. It’s a steady spiral into who-knows-where.
The funny thing is that I’d been asking AI all along how my app was doing, and it constantly assured me I was doing a great job, so I felt pretty confident. (Red flag, I know.) Even this last time, I uploaded all of my code to Claude and asked it to give me a grade. It said I was at a B+, mainly because I hadn’t completed the app yet.
I was suspicious, however. It had become so problematic to make progress. Was that just because that’s normally what happened at this stage, or was it because something else was wrong? AI is a notorious suck up, so I had to think of a way to get it to give me its honest feedback. In the end, I uploaded the exact same code, but this time I told it that I’d been assigned to continue development on this app that had been made by someone else, and I was trying to decide if I needed to start over.
Lo and behold, Claude was suddenly very concerned about the direction this app was taking. There were all sorts of red flags. It noted different coding styles, duplicate files, things that conflicted with each other, and more. My best bet, it assured me, was to start over completely.
On the one hand, I understand why people have made AI such a suck up. Everyone likes to be told how smart they are and how great everything’s going. On the other, the fact that it’s like this makes it incredibly less useful than it ought to be. There needs to be some sort of a happy middle, where it breaks the news that actually, things aren’t that great. We’re not all geniuses. But I guess we’re not at that point. In the meantime, just don’t tell AI if something was your idea.
So what am I going to do differently the fifth time around? Well, the good news (See? It was coming.) is that since I’ve been working on this project so much, I’ve gained a much better understanding of exactly what I want the app to do. How I want it to behave. What the database should store for information. How to access that information. I’ve been learning a ton about how an app like this should be built, and I’m to the point where I understand (I think?) how I’d put it together myself if I knew how to code.
With that in mind, I’m going to take a much more active part in the coding process. I’m not just going to blithely assume AI did something right just because it told me it did. I’m going to look at what it came up with, evaluate it each time, and make sure it’s what I want it to be. Yes, this means it will be more difficult. And yes, there’s a chance that what I’m really doing is just taking the long way to teaching myself how to code. But I’m still enjoying it, so what’s the harm?
Will it work, or will I need to start over yet again?
Stay tuned . . .