I Asked AI to Build a Blog. Things Got Weird.
I'm on paternity leave after seven years of increasingly management-heavy roles, and I decided to get my hands dirty again. Not with some grand plan or research agenda, just genuine curiosity about what AI development tools can actually do when you let them loose on a real project.
I started with some basic questions: What's the trade-off between future-proof design and fast start? How does that work when you say NO to ClickOps and want strong documentation? Nothing profound, just the kind of decisions you face on any project.
The premise was simple: build a blog platform using AI assistance and see what happens. No expectations, no hypothesis to prove. Just an experienced engineer taking a step back from meetings and strategy documents to write some code again.
Three weeks later, I'm debugging Terraform modules I didn't write, arguing with an AI about whether we need Skaffold for local development, and somehow coordinating between multiple AI tools that each have strong opinions about architecture.
What I Thought I Was Building
A simple cloud-native app. Nothing more.
What I'm Actually Building
An accidental experiment in human-AI collaboration where I'm making a lot of mistakes and occasionally learning something useful.
It turns out that when you approach AI tools with genuine curiosity rather than skepticism or blind enthusiasm, you discover something interesting. These tools don't just automate coding tasks. They change how decisions get made, how you learn, and how you think about technical problems.
Claude Code suggested Kubernetes for my blog. My first instinct was to push back, but then I started thinking about why. Not because I need container orchestration for a simple blog, but because working with modern deployment patterns might teach me something about how AI approaches infrastructure decisions. So I said yes.
Now I have a multi-environment Terraform setup with proper modules, GitHub Actions pipelines powered by Dagger, and a monorepo structure that's probably unnecessary but somehow makes sense. We also put way more effort into dev environment setup and documentation than the actual application itself.
The Learning Process
After years of reviewing other people's code and making high-level technical decisions, I forgot how much you learn by actually implementing things. But learning with AI assistance is different from traditional hands-on development.
When AI generates a complete Terraform module, you're not learning Terraform syntax through repetition. You're learning infrastructure patterns through analysis. You read the generated code, understand the approach, modify it for your needs, and gradually build intuition about what good infrastructure looks like.
It's faster than learning from scratch, but it requires a different kind of engagement. You need to actively question the AI's choices, understand the trade-offs, and develop judgment about when to accept or reject suggestions.
What's Actually Weird
The weird part is how quickly you go from "simple blog" to "enterprise architecture" when AI removes the friction.
Normally, if Claude Code suggested setting up Kubernetes with Cilium networking for a blog, I'd laugh and tell it to suggest something reasonable. But when it can generate all the configs, explain every decision, and have it running in an hour, suddenly you think "why not?"
Same with the monorepo structure. Claude Code created such an overdone design that we had to delete most of the code and redesign it. Same thing happened with the Terraform - so many modules and abstractions that we scrapped half of it and started over. We also put way more effort into dev environment setup and documentation than the actual application itself.
The strangest part is that these failures feel educational. I'm learning what not to do with modern tooling, which is probably more valuable than following a perfect tutorial. When AI can generate complex solutions instantly, you get to fail fast and understand why simpler approaches often work better.
As a trained historian, I'll also be weaving in historical and cultural context throughout this series. How we got to this point where "simple" requires enterprise architecture, why certain patterns emerged, and what these human-AI collaboration dynamics remind me of from other technological transitions. The story of software development is also the story of how humans adapt to new tools, and there are patterns worth exploring.
What's Coming
Next: The characters that emerged. Then: Technical decisions and why. After that: AI in popular culture and how software delivery changed over 15 years.
About the Author
Yair Etziony is an engineering manager and DevOps veteran with 30 years of experience in systems administration and infrastructure. Based in Berlin, he's served as Engineering Manager at AMBOSS, founded Polar Squad Berlin, and is a CD Foundation Ambassador. He's also an electronic music producer and holds a background in history and philosophy.
Top comments (0)