Vibe Coding Week 1: Building Web Apps on Intuition
I’ve been on a fascinating journey with AI-assisted coding lately. It started, as I’m sure it has for many of you, with a simple copy-and-paste from a chat window. But it’s evolved into something I’m calling “vibe coding” – a way to build things purely on intuition and conversation, even with languages I’ve never used before.
Getting along with AI: From Copy-Paste to Conversation
I started using ChatGPT to help me with coding tasks in late 2023/early 2024. At first, it was all about getting snippets of code to solve specific problems. I would ask for a function, copy it, and paste it into my project. As little as such small bits of help were, they really surprised me with how much they could speed up my workflow and inspired me to think more about a lot of questions. Are they helping me learn, or are they making me lazy? Are they helping me become a more productive developer, or are they making me a less capable one? It really was a mix of both, I think. On one hand, I was very excited about my improved productivity and the ability to get more done in less time, hence having more time actually to do the thinking and creative work. On the other hand, I did feel the fear of losing some of my coding skills and becoming too reliant on AI. It was a bit of back and forth back then, and eventually, I made some agreements with myself regarding how I would use AI tools in coding:
I would allow myself to discuss with AI only after doing the initial brainstorming by myself, to identify challenges and evaluate the viability of my idea, because I want to make sure of my motivations, goals, and requirements that the results should meet. Without those, I can see one easily get lost in the constant dopamine hits seeing code being poured onto the screen, hence a sense of satisfaction about achievement, without knowing something probably already went wrong without being noticed.
I would always read through code generated by AI to make sure they are logically correct, and always check important results immediately to catch errors at an early stage. In the early days, I made the mistake of not checking the code and results, but just kept prompting the AI, assuming everything was going in the right direction. Yes, the speed at which AI generates code is very fast, and so is the speed at which errors snowball. Once that happens, it becomes a bigger headache than either coding manually from scratch or debugging my own code.
Be aware that an AI is an AI all the time. I think the boundary between AI - something that is supposed to be “human-like” but is inherently non-human, and a real human, is what causes most controversies. It is such a simple yet complex question. It is simple because, apparently, we and AI models are nothing alike in our physical and chemical nature. But it is also complex; the “human-likeness” can actually interact with and affect real humans’ feelings and perceptions. Isn’t that such a fascinating question? Ultimately, it boils down to the philosophical questions that have been explored in numerous science fiction works for decades: What is consciousness? What is intelligence? Ok, back to my point! AI chat
Discovering “Vibe Coding”
(Describe how you started using Gemini and Cursor. What did it feel like to code in a new language this way? How did it change your process from just asking for snippets to having a coding partner?)
The Creations
(Showcase the two applications you built. You can add links to them or even embed screenshots.)
Application 1:
Application 2:
What’s Next?
(Share your thoughts on the future. What do you want to try next with this method? What are the possibilities?)