UI state connected directly to the Backend/DB? Maybe even some sort of sync engine?

Still staying motivated and seeing the chance, in other words, keeping the vision up. Only way to get the edge at some point. Good luck.

Hm, I feel you. Take a break and talk about it (which you are apparently doing). When possible ask for help.

Reason why “let me rewrite this” is so common. While it can be a valid option, it usually tries to avoid the challenge of mapping the domain to software and maintaining it in a coherent way while keeping the thought processes and knowledge aligned, as they are hard to keep simple.

An association that I have in my mind and prob the fact that I’m more interested in the reasons for this. Didn’t want to hijack nor derail.

Sometimes “time to market” is just more important than a feature working the intended way.

Might be I’m wrong, in the end the difference between the status quo and a revolution is a tiny change. I remember writing my bachelor thesis about CNNs (state of the art were SSDs and YOLO v2) for number plate detection and now we are here.

What I perceive is that the most recent improvements with LLMs while being impressive, fail to fix the core issues. I think the next increment will be AIs incorporated into tools that solve specific tasks and specialists utilizing those.

That said I’m yet to see an LLM go beyond the mere basics of a complex project. They have apparent limitations in complex situations like being a yes-sayer, ignoring complexity, expecting input in a specific way etc. It can even fail to help with chores, when it isn’t a certain type of a task.

LLMs are strong in greenfield projects particularly if you don’t need a human to be able to read the intermediate layer (think code) that a LLM generates which again creates the final output.

I do use LLMs on daily basis and I just recently introduced my brother who has barely any coding knowledge into this space and he managed to build a website for a construction company he works for using Claude and Next.js in ~1 week with just a tiny bit of my help.

Might be I’m wrong, in the end the difference between the status quo and a revolution is a tiny change. I remember writing my bachelor thesis about CNNs (state of the art were SSDs and YOLO v2) for number plate detection and now we are here.

What I perceive is that the most recent improvements with LLMs while being impressive, fail to fix the core issues. I think the next increment will be AIs incorporated into tools that solve specific tasks and specialists utilizing those.

That said I’m yet to see an LLM go beyond the mere basics of a complex project. They have apparent limitations in complex situations like being a yes-sayer, ignoring complexity, expecting input in a specific way etc. It can even fail to help with chores, when it isn’t a certain type of a task.

LLMs are strong in greenfield projects particularly if you don’t need a human to be able to read the intermediate layer (think code) that a LLM generates which again creates the final output.

I do use LLMs on daily basis and I just recently introduced my brother who has barely any coding knowledge into this space and he managed to build a website for a construction company he works for using Claude and Next.js in ~1 week with just a tiny bit of my help.