New Year, New Site: AI Coding in 2025

An introduction to my new personal project that I built with AI, and what I learned.

portfolioai codingnocodeweb developmentartificial intelligencetechnology

New Year, New Site

I’ve tried and failed to build a website I love enough to share with the world probably four or five times until this past week. I’ve tried multiple methods: the old e-book code-along, the 2 hour Youtube videos to help you understand a basic template, and I’ve even tried from official documentation as if I know the details of Javascript/HTML/CSS. I realized I have no excuses anymore now that we’re in 2025; in the last year I have come across new tools and gained enough terminal/code editor experience to overcome my fears and create something fun to house my thoughts, musings, contact info, and photos while also giving myself a fun and practical project to build and iterate on.

This really started when I was stumbled across some really cool sites on reddit.com/r/webdev or through AI-related avenues. Sites like https://www.lysonober.com/ or https://www.ericwu.me/ inspired me to get my s*&t together and shit down until I learn enough to get a site running that feels good enough to host my resume on.

Throughout my search, I’ve also been learning more and more about AI tools and their coding capabilities. Claude and ChatGPT have been used in simple applications in the past, but the true unlocks that helped me accelerate my process were the combination of Bolt and Cursor AI or Windframe (a cursor competitor that seems to just understand what I am trying to do a touch better in some cases). These tools helped me visualize and understand every mistake made along the way and continuously improve every detail, from the hover functionality to the filter nav-bar, light and dark mode implementations, blog page formatting, and the subtle-but-necessary animations on the draggable bento grid.

I learned a lot about the tooling to build things with generative AI and different LLMs, but I also got a better understanding of different models and their real-world purposes. Claude is the best at coding, but it’s expensive if you’re using the API to generate content. I’ve had to stop my workflow when I am in the groove because I run out of AI prompts from creating, refactoring, or debugging my code. OpenAI has more models for more use cases, but ChatGPT 4o isn’t great at coding or coming up with creative solutions to my problems, ChatGPT o1 is slow and only OK at code, and the rest just have more niche use cases like automating workflows (which I am starting to learn more about and am excited to some results for). Deepseek and Llama models are super cheap and faster than OpenAI in some cases, but results vary wildly. I think if I had to do thousands of different prompts, I’d use Deepseek v3 and R1 for architecting and planning my code. Alas, I only need to build this simple site and can work with Claude instead. I think these findings tell me that Anthropic is the clear #1 in coding, and I am stoked for their next model to blow us away (and hopefully be cheaper to run).

This blog is just the start to kick off this site — I would love to share some more of my nerdy tinkering and some real world results and implementations that might help someone out there. This has been super fun to play with, so I hope this can inspire you to dust off the idea you’ve been sitting on for a while. Between Bolt, Cursor/Windframe/Claude, and the endless implementations of AI getting better by the day, I think we’re all gonna have pretty sweet web applications in our portfolio very soon.