Building a Website With a Custom Tech Stack for Agentic Engineering

The tech stack I chose for my website is built for quick iteration with AI tooling. I use OpenCode in my personal projects and it has really sped up my development. But something we are finding out is that the AI we use is only as good as the tools that are attached to them. The recent developments in the Open Weight models also make it very economical to develop with these tools. I use OpenRouter to connect to my OpenCode instance so I can experiment with and try different models because they are each really good at a specific task. My go to model for feature development is GLM-5 which was released this past week and is about 10 to 20 times cheaper than the roughly equivalent Claude Sonnet 4.5.

Developing with AI has made creating software more fun for me and helped me learn faster. I get less bogged down with the specific details of the syntax of a language or framework and spend more time building software that is meaningful and useful. I've never been one to entirely specialize. AI tooling and coding with LLMs make me a much more effective jack of all trades. For anyone looking to specialize in a particular framework or language, I am not sure if using AI tooling in the software development cycle will reap benefits. I got to learn Java and Spring Boot well when I was getting consistent feedback from the compiler, debugger, and lsp directly. Handing those lower levels of feedback off to the AI means I am not as familiar with the nitty gritty details of syntax but am more familiar with architectural decisions and tradeoffs.

In other personal projects and ones I have taken on at work, I get to consider the implications of different auth flows, how and where secrets should be stored, and the end user experience of the software I am writing. Solving for these kinds of higher level problems brings me much more joy when developing, even when considering the unpleasant challenge of auth in its many forms.

For the technical challenge of this website, I chose Bun, Eleventy, Tailwind CSS, and HTMX as the basis for my app. Bun gives a very fast JS runtime for my local development and faster builds when deploying the app. With how slow npm is these days, I am hopeful that more enterprises start to look for other solutions because after installing all my dependencies with bun I never want to go back. Eleventy and Tailwind CSS are specific choices for AI tooling. I can write my blog posts and songs in markdown which seems to be most if not all LLMs' best form of text output. Tailwind CSS is also friendly to AI since the classes that come with the library describe what they do. Finding the right classes has always been a needle in a haystack problem but all the best LLMs are great at that as well. In short, Tailwind CSS seems to be the best CSS library for use in Agentic Engineering. My choice for using HTMX is largely due to future proofing my website in case I ever want to have basic state management for a comments section or something of the sort. Right now the only benefit it gives is the ability to serve HTML fragments from the server. It is a very small performance boost but good nonetheless.

All of these technologies are still new to me. I have more learning to do with all of them and their specific quirks but it has been a very pleasant experience so far. I was able to get the kind of newspaper feel I was going for in just a couple days. And as I have made the website I have learned a lot about the technologies along the way. This is my favorite part about using these tools. My understanding of the specific technologies and their strengths and weaknesses is deeper than it was before because I can chat with the AI agent about specific technical challenges or roadblocks I am facing. Water cooler type conversations help me understand these topics much better. And the side benefit is I can ship code that is functional faster.