I'm no coder but I managed to generate an app with 2000 lines of code that does exactly what I want.
That's very good, particularly since, if the code isn't modular, updating anything in a 2k lines of code project would likely hit the daily limit on most well-known free models out there. You seem to have developed a process that works for you. That's the important part!
I built an app using google sheets and app scripts for my offline business. It allows me to simplify and speed up my workflow. Something that I wanted to do for about 10 years and tried several time and always hit a barrier somewhere that prevented me from achieving it. Thanks to AI, I've managed to build something that works:-). I mainly use gemini for doing it. I also found an open source version or cursor that allows me to jump from one model to the next once the free calls are over :-).
I mainly use gemini for doing it.
I found Claude best for writing and updating full blocks of code (their interface is even better when you paste code it includes it much like an attachment, rather than a wall of text in the prompt). But I don't think it's best at step-by-step guiding you through changes. I pretty much used all of them to escape the daily limit.
I also found an open source version or cursor that allows me to jump from one model to the next once the free calls are over :-).
But you still need to provide context when you switch from model to model, right? Or they resubmit the prompt to the next model automatically?
I also like Claude but the free calls are quickly gone :-).
But you still need to provide context when you switch from model to model, right? Or they resubmit the prompt to the next model automatically?
What I do is to actually save the PRD and step-by-step as txt files within the project. Then I call them as context resource when I start with a new model.
I also like Claude but the free calls are quickly gone :-).
You know you can have multiple sessions, right? When a session ends, you can open a clean one. Eventually you run out of free prompts, but not as quickly. You need to be strategic about it. Usually I break things down and follow different threads in different models.
What I do is to actually save the PRD and step-by-step as txt files within the project. Then I call them as context resource when I start with a new model.
That's a good idea, but won't work if the project gets bigger, because you'll run out of context (I ran into an issue that I couldn't complete 1 prompt before I got to the daily limit - very tough to get out of that situation: I used partial responses and then asked more targeted questions to other models or different sessions while removing the parts that I fixed from the input). You need to break it down into pieces and only work onto the ones that don't interact with each other when the project gets bigger.