Over the last few months, a new term has been circulating in developer communities: “vibe coding.”
Depending on who you ask, it's either:
- the future of software development
- pure hype that will disappear in a year
- or the beginning of the end for developer jobs
Instead of debating it endlessly, I decided to run a simple experiment.
For one month, I would build and launch a real full-stack application using AI agents as development assistants.
The goal wasn’t to let AI build everything for me. Instead, I wanted to see:
- How much faster development becomes with AI assistance
- What parts of software engineering still require human thinking
- Whether this workflow actually makes building products more enjoyable
The result of that experiment was Ninco, a personal finance dashboard that I built and launched in about 30 days.
The Experiment
The rules were simple:
- Build a real product from scratch
- Ship something usable
- Use AI agents throughout the development process
- Keep infrastructure costs minimal
- Finish within 1 month
The goal wasn’t to prove that AI can replace developers.
Instead, I wanted to test whether AI could remove friction from development and allow me to focus more on architecture and product decisions.
The Stack
The project ended up using a modern full-stack setup.
AI Development Environment
- Google Antigravity
- AI models: Gemini 3.1, Claude Opus, Claude Sonnet 4.6
Frontend
- Next.js
- Shadcn UI
- TanStack Query
- Clerk
Backend
- Fastify
- Prisma ORM
- PostgreSQL
Infrastructure
- Docker
- GitHub
- Monorepo: pnpm workspaces
This stack allowed me to move quickly while keeping the system structured and scalable.
How AI Actually Helped
AI was incredibly helpful in several areas.
1. Boilerplate Generation
Modern development includes a lot of repetitive work:
- CRUD endpoints
- validation schemas
- API handlers
- form structures
AI agents dramatically sped up writing these pieces.
What normally takes hours could often be generated and refined in minutes.
2. Debugging
Instead of spending long periods digging through documentation or StackOverflow, AI helped quickly identify common mistakes, especially when integrating new libraries.
3. Rapid UI Iteration
I used AI to experiment with UI layouts, component variations, and styling tweaks much faster than I normally would. Sometimes, challenging the AI to create complete interfaces without many details, and the results really surprised me.
This made the UI development process feel much more fluid.
4. Documentation and Refactoring
AI also helped generate documentation and suggest refactors for clearer code structure. If you ask the AI "How to improve this component using the best programming practices for a better performance/security?" or even cite a specific methodology or concept that you want to be applied, the outcomes will be quite satisfactory.
Where AI Didn’t Help Much
Despite all the advantages, some areas still required human engineering decisions.
Architecture
Designing the system structure, deciding how the frontend communicates with the backend, and organizing the monorepo still required careful planning.
Many times, the AI generated very large files with too many functions, reduced performance, or struggled with dependency versions. However, the result is largely a product of how well you use AI. As a beginner in this workflow, I surely committed mistakes with my inputs, lack of contexts or misuse of AI Skills. I truly believe that with some more experience and technical knowledge, the AI can solve most of the architectural issues.
Database Design
Every application requires accurate data modeling.
Defining schemas, relationships, and constraints still required careful manual thinking. I left the first draft to the AI, but of course it won't know how you want to scale or very specific fields that each table should have unless you explicitly tell it, so I had to review and edit some parts.
Product Decisions
AI can help build features, but deciding what features should exist in the first place is still a human job. You can get some ideas asking Chat GPT, but the final decision is yours.
Understanding users and designing workflows remains a core product skill.
The Result: Ninco
After about a month, the experiment resulted in a complete full-stack application called Ninco.
Ninco is a modern personal finance dashboard designed to make money tracking simple and fast.
Some of the core features include:
- Tracking income and expenses
- Register transactions via AI chat
- Multiple financial accounts
- Custom categories with colors and icons
- Advanced transaction filtering
- Interactive financial charts
- Data export (CSV, JSON, PDF)
The focus of the product is clarity and speed.
Instead of complex financial workflows, the goal is to make tracking finances feel simple and intuitive.
Screenshots and Demo
Ninco demo | Arcade
app.arcade.software
You can visit the app on: https://ninco.app
Lessons from the Experiment
After building a real product with AI assistance, a few things became very clear.
AI drastically speeds up repetitive coding
Boilerplate tasks that normally take hours can now be generated in minutes.
Architecture still matters
AI can generate code, but good system design still comes from human decisions.
AI works best as a collaborator
The most productive workflow wasn’t replacing development with AI.
It was pair-programming with it.
Shipping products is easier than ever
With modern frameworks, cloud infrastructure, and AI tools, a single developer can build surprisingly complex systems in a short time.
Final Thoughts
This experiment changed how I think about software development.
AI didn’t replace engineering work, but it amplified productivity and removed a lot of friction from the development process.
More importantly, it made building the project much more fun.
The result of this experiment is Ninco, which is now live and evolving. Check it out!
I’d Love Your Feedback
I'm curious to hear what other developers think.
- Do you use AI in your development workflow?
- Do you think “vibe coding” is a real shift in software engineering?
- What features would you expect from a personal finance dashboard?
Feel free to share your thoughts or feedback — both about the experiment and the product.
Top comments (0)