My first ai agent using n8n
It took me 8.5 hours to build it, with the help of ChatGPT-5.2. From setting up the local environment to finally managing to produce something very close to what I want.
I actually broke down when fine-tuning the JavaScript code and cried for FASTER help from ChatGPT, especially when our conversation became very buuuulky. Even though I knew it was trying really hard to help me, I still couldn’t help but drop some tears…
I would really recommend anyone who wants to build with LLMs (large language models) to watch the YouTube video how LLMs work by Andrej Karpathy. Because after understanding how they work, you won’t be angry with ChatGPT anymore, but with yourself.
After watching Andrej Karpathy’s 3 hour video (I’m only halfway through), I realized that the anger I felt when building my agent was actually not directed at it. The reason why I cried was probably because, unconsciously, I was blaming myself for my own lack of domain knowledge and understanding of LLMs.
At this stage, LLMs are already very powerful, but using them well isn’t about being taught. I don’t see my relationship with ChatGPT as teacher and student, or boss and employee. Instead, it feels more like working in a transdisciplinary team. I’m the team leader, and ChatGPT is an expert teammate. We’re working toward the same goal, and my responsibility is to understand its strengths well enough to bring out the best results from the team.
So before, when I didn’t understand something after ChatGPT tried to teach me, I could still blame it a bit. But now, I really feel like I can’t blame it anymore.
Oh my god. 天呐。