Building My Personal Website After 16 Years in IT: A 100% AI Vibe-Coding Journey

Published July 27, 2026

Published on July 27, 2026 • Updated on August 27, 2026

By Denny Eka Saputro (16 years in IT)

Building My Personal Website After 16 Years in IT: A 100% AI Vibe-Coding Journey 1

Hello world, and welcome to my first personal website! I've been working in the IT industry for 16 years now. Typically, developers launch a personal portfolio within their first 3 to 5 years of coding, once they've accumulated a solid list of projects, skills, and experience. By that standard, building my first portfolio site now might seem long overdue. But hey, it's never too late to try something new.

In my recent work, I got exposed to AI, specifically Agentic AI. I saw it as a powerful tool to build full applications without having to write all the code manually. That gave me an ambitious idea: create a full portfolio website using 100% AI. Luckily, AI tools today have become powerful enough to generate production-standard software.

The beginning: vibe coding with Kiro

My journey started with Kiro, an agentic AI tool from Amazon. I used ChatGPT to craft a detailed prompt so Kiro would understand all the technical requirements for my site: a Go backend, a React frontend, and deployment on the AWS free tier. With that prompt ready, my vibe-coding adventure began.

The prompt covered all my needs, and I was pleased with the starting point. I fed it into Kiro and began vibe coding using Kiro's free-tier tokens. It was a fascinating experience. It felt like managing a brilliant junior developer who possesses high technical skills but little business domain knowledge. As a programmer, I had to shift my role completely, from writing code to acting as a business analyst and client, guiding the AI on what to build rather than how to write it.

However, it wasn't all smooth sailing. I frequently ran out of tokens because Kiro would misunderstand requirements, introduce bugs, or struggle with creative work like generating icons. It constantly required manual correction. Eventually, after using up free tokens across multiple email accounts, the project ground to a halt and was left in the corner.

The Claude era and token limits

Then I discovered Claude. I was introduced to it on a work project that had a whole configuration behind it: Hermes, Ollama, Obsidian, Slack, and Jira integration. Exploring it was a lot of fun, and it dramatically boosted my productivity. Yet, just like with Kiro, token limits quickly dampened the excitement.

It is often hard to predict how many tokens an AI will consume, what the standard baseline is, or why a simple request suddenly eats through your allowance. Even estimating task complexity costs tokens. Every day, our team's tokens were drained and needed refilling. Fun fact: subscribing to a service doesn't always cover API usage, which is double damage for your wallet.

Claude Code takes over

Despite the token struggles, this website was revived and refined by Claude Code. It easily handled the heavy lifting that Kiro couldn't. Beyond fixing issues, it provided valuable suggestions and added features based on our vibe-coding sessions. Models like Opus transformed this abandoned project into a production-ready codebase.

But history repeated itself. Right at the finish line, the final deployment step, I ran out of tokens once again. With just a few steps left, I was stuck waiting for token replenishment.

Crossing the finish line with Google Antigravity

Determined to finish, I looked for an immediate solution. With the finish line in sight, I took a leap of faith and enlisted help from another AI: Google Antigravity. It took over the deployment process, referencing Claude's last execution plan. Naturally, there were a few hiccups requiring quick hot fixes along the way. But in the end, we made it. My personal website is live at desaputro.com.

After 16 years in IT, having AI agents collaborate to help me launch my personal site was an incredibly satisfying milestone. Here is the complete tech stack behind it:

  • Backend: Go with the Gin framework

  • Frontend: React 18 and TypeScript, built with Vite

  • Data: DynamoDB

  • Hosting: AWS Lambda behind CloudFront, with static assets served from S3

  • Infrastructure: Terraform, with Cloudflare handling DNS

I didn't write that code by hand. Everything was accomplished through vibe coding.

What's next?

Now that the site is published, I have several new ideas and feature improvements planned after spending a few days browsing my own site. Some of them are already real: the content management system I'm using to write this very post came out of that list, and more has followed since.

Will I involve AI again? Absolutely. It has been an amazing journey. Today, AI tools are already fantastic assistants, and I believe they will become indispensable companions in the future.