How I Built a Web Application at 57 With AI Assistance — A First-Hand Account
My name is Jovica. I'm an electrical technician by training — graduated in 1987, specializing in automation and computing. For the past twenty years, I've worked as a production manager in the textile industry, leading teams of anywhere from ten to a hundred people. The last time I seriously wrote code was over forty years ago, in Basic, on machines that looked like spacecraft technology at the time.
Then I decided to build a web application.
Why Now?
It wasn't some dramatic decision. I noticed a concrete problem — investors building or renovating properties in Serbia have no single place to track the procedure, required documentation, and institutions they need to deal with. The bureaucracy is complicated, deadlines get missed, documents get requested at the last minute.
I thought: this could be solved with an app. And then I thought: but I don't know how to build apps.
And then I remembered that Claude exists.
What Is "Vibe Coding"?
I didn't know the term myself until I came across it. In short — instead of spending years learning a programming language, you work with an AI assistant that understands what you're trying to achieve and helps you build it. You make decisions about architecture, features, and user experience. The AI writes the code.
Sounds simple. It isn't always.
What We Built
The project is called Planer Izgradnje (Construction Planner). In roughly two weeks, working a few hours per day, we built:
- An interactive checklist with 6 construction phases and all required documents
- An institution database with links (land registry, e-permit portal, utilities...)
- A cost calculator
- An AI assistant that answers questions about construction procedures in Serbia
- Document upload functionality for each procedural step
- An admin panel for content management
- Full user authentication
Tech stack: React, Node.js, Express, MongoDB, Anthropic Claude API, Cloudinary, Clerk. Deployed on Vercel and Render. Code is on GitHub.
Yes, it's a real application. Yes, it works. Yes, I surprised myself.
... and another one for fun
Before Construction Planner, I also built AgroAsistent — a web application for Serbian farmers. It provides real-time weather data for any municipality in Serbia, a field activity notebook, parcel management with eAgrar import, a spray dose calculator, and a local resource finder (agricultural pharmacies, vets, markets). The tech stack is the same — React, Vite, Tailwind, Node.js, Express, MongoDB, and Clerk authentication. Deployed on Render, available as a PWA. Code on GitHub: github.com/jovicalesevic/AgroAsistent.
What Went Well
Project architecture. From day one, Claude set up a clear structure — database schemas, API routes, component layout. I didn't have to figure out how to organize the code. As a former manager, this felt natural — I had a plan, I knew what went where.
Speed. Things that would take a junior developer days, we resolved in an hour or two. Mongoose schemas, Express routes, React components with Tailwind styling — everything came together quickly.
Cursor as a tool. An editor that integrates AI directly into the development environment was key. Instead of copying code from one window to another, the AI wrote directly into the file.
What Was Hard
Supporting services. Vercel, Clerk, MongoDB Atlas, Cloudinary — each of these has its own interface, its own logic, its own failure modes. None of them are intuitive for someone coming from a completely different world. A deployment that should take five minutes could eat up an hour because of one incorrectly set environment variable.
Google Chrome. This sounds ridiculous, but the browser was one of the biggest sources of frustration. A cache that refuses to clear, a Service Worker that blocks new versions of the app, CORS errors that only appear in certain browsers — Chrome seemed to actively work against the project.
Communication errors. Sometimes I copied code from the chat into the wrong file. Sometimes I was in the wrong folder in the terminal. Small mistakes that an experienced developer wouldn't even notice, but ones that could cost me half an hour.
What I Learned
That age is not a barrier. That experience from a completely different field — managing people, planning, problem-solving — translates directly.
That the most important part of any application is that it solves a real problem. Code can be generated. The idea has to come from a person.
That supporting services are often a bigger challenge than the code itself.
And that the satisfaction when an application finally works on your phone, in a real browser, with real data — is greater than I expected.
Final Thoughts
Planer Izgradnje isn't a perfect application. There's room for improvement. But it's functional, deployed, and solves a real problem.
If you're considering a similar project — regardless of age or prior experience — my advice is: start. Define the problem. Talk to an AI about architecture before you write a single line of code. And be prepared for Chrome to be your greatest enemy.
Code on GitHub:(https://github.com/jovicalesevic/planer-izgradnje)
(https://github.com/jovicalesevic/AgroAsistent)
Live app's:(https://planer-izgradnje.vercel.app)
(https://agroasistent-client.onrender.com)
Joca Lesevic — Automation and Computing Technician , former production manager, aspiring vibe coding engineer.
Top comments (0)