When I started building Chat4Us, it quickly became my favorite topic. I loved discussing this project—which evolved from a simple concept to a working prototype—with fellow tech enthusiasts. I was always eager for feedback.
I’d explain that Chat4Us isn’t an AI model itself. Instead, it’s a software layer that monitors and moderates chat flows. Its key role is to provide human interaction when LLMs hit a wall with complex cases.
After one explanation, I received a blunt reply: I had no future as a developer. The person claimed AI can now build complete software from simple prompts.
A couple of years ago, that comment would have shaken me. So, I tested it. I tried to build an entire app from scratch using a long series of AI prompts. The result? AI only succeeded on small, isolated tasks. It consistently failed to add new features to its own creations.
This may happen someday, but I’m skeptical. My doubt only grew when I heard the CEO of a leading software company state that just 30% of their code is AI-generated.
Today, my approach is a partnership. I use AI exclusively for generating general-purpose, reusable code. However, I still write all the specific application logic myself.
Why? I find it too time-consuming to explain every project-specific detail in a prompt. Furthermore, I genuinely enjoy coding. Writing that core logic myself gives me a real sense of accomplishment.
You can see this division of labor in the Chat4Us-Creator source code on GitHub. For instance, most code in the ‘Util’ package is AI-generated. Conversely, the rest of the codebase is written by my own hand.
Currently, I use DeepSeek as my co-developer buddy. Its code integrates seamlessly into my projects. It has become an indispensable partner, offering valuable perspectives and clarifying parts of my project that need a fresh look.
Ultimately, AI isn’t the solo developer—it’s the ultimate assistant that empowers the human expert.
