My Thoughts: Through this project, I learned that AI can be very useful in architecture, but only if I use it carefully and not just accept everything it gives me. For me, one of the best things AI can do is gather and organize information quickly. This was clear in the Perplexity part, where I used AI to understand the Port Dickson site better. It helped me look at climate, wind, humidity, sun exposure, sea breeze, rainfall, and passive cooling. This made the site analysis stronger because I could connect the information to actual design decisions, like improving shading, using cross ventilation, opening the building toward the sea, and making the courtyard work as a ventilation core.
I also found that AI is good because it is fast. It helped me test many ideas in a short time. I could explore concept ideas, diagrams, environmental strategies, sustainability ideas, and user experience diagrams much faster than doing everything from zero. This was helpful because I could see different possibilities before deciding which direction was stronger. But I also learned that speed is not always the answer. In architecture, I still need time to slow down, look carefully, think properly, and build confidence in my own work. If I move too fast, I might choose something only because it looks nice, even if it does not really fit the project.
For concept development, AI was very good when it was explaining ideas through words. ChatGPT helped me develop the idea of “Stillness in Motion” and connect Alexander Calder’s work to balance, lightness, movement, suspension, shadow, and reflection. It helped me understand how Calder’s stabiles could become an architectural idea, not just a shape. But when I asked AI to generate images, I noticed a big problem. The images can look very clean and impressive, sometimes even better than a normal architecture render at first glance. But when I looked closely, the details were often lacking. Sometimes what AI said in words and what it generated in the image were not the same. It could talk about lightness and openness, but the image would still look heavy, solid, or fortress-like.
Using three AI tools was useful because I could compare different opinions. Perplexity helped with research, ChatGPT helped with concept, Claude gave critical feedback, and Gemini acted like a third judge. This helped me make better decisions because I was not depending on only one AI answer. Claude and Gemini helped me realize that the project needed stronger structural logic. Lightness could not just be a word in the concept. It had to appear through the structure, roof gesture, courtyard, shading, material, and the balance between solid and void.
At the same time, using different AI tools was not always easy. It was hard to keep track of three different AI chats. Sometimes information got repeated, and sometimes important points got lost along the way. I had to keep summarizing the process and deciding what was actually useful. So AI made the process faster, but it also created more information that I had to control.
Another limitation was editing with AI. I found that it is very hard to edit an AI image exactly the way I want. Usually, the first image AI generates is the strongest or cleanest one. After asking for many edits, the result sometimes becomes worse or moves away from what I wanted. This showed me that AI is good for testing ideas quickly, but it is not always good for precise architectural control.
Overall, AI helped me with research, concept development, visual testing, comparison, and reflection. However, it did not replace my judgement. I still had to decide what to keep, what to reject, and what to change. The most important lesson I learned is that AI should support the design process, not control it. The final direction still needs to come from my understanding of the site, the concept, the user, the structure, and what I believe is right for the project.
Note: Reflection based on my AI workflow using Perplexity, ChatGPT, Claude, Gemini, and Canva AI, and my revised Studio Retreat for Alexander Calder project.