I usually write backend stuff in Go, so I prefer strict typing, predictable performance, and explicit code. However, I wanted to see if the recent hype around AI-assisted development ("vibecoding") is actually useful for building production-ready apps, or if it's just for quick MVPs.

To test it fairly, I picked a stack I don’t use daily: Node.js (Express 5) and Vanilla JS. The result is a simple toolkit: toolkitch.ru

My main goal was privacy. Most web formatters and encoders send your data to their servers. For this project, everything sensitive (JWT decoding, hashes, ciphers) runs strictly client-side in the user's browser.

The Setup:

Backend/Frontend: Node.js + Express 5, Bootstrap 5.

Security: Production headers via Helmet, strict CSP, and CORS.

DevOps: GitHub Actions, Docker Compose, Traefik with automatic SSL.

I used KodaCode as my AI assistant. My takeaway: it didn't replace my engineering choices, but it handled the grunt work well—scaffolding routes, writing Docker configs, and generating boilerplate Bootstrap UI. It allowed me to focus on architecture and data privacy.

The UI is currently in Russian, but the developer tools are standard.

If you use AI in your daily engineering tasks, how do you handle prompt management for complex DevOps or security configurations? Feedback on the site is also welcome.