Two years ago, I started building a website from scratch and quickly discovered that artificial intelligence could be an excellent coding partner. This story shows how I gradually learned to collaborate with AI to code faster, better, and with more control.
Two years ago, I started building a React website from scratch for my husband. Frontend, backend, database, tests — the full package. At first, I did everything myself, but over time, artificial intelligence became an increasingly integral part of my workflow.
From Copy-Paste to Collaboration
In the beginning, AI worked more like a code notebook — I would copy fragments, add context, and ask for fixes. Over time, autocomplete tools (TabNine) appeared, which actually sped up the work.
And then — Claude. That was the breakthrough. Suddenly, a site that previously took me weeks to tinker with could be built over coffee. Claude didn't just write code; it understood the project structure. Thanks to it, the website creation process became smooth and creative.
What Working with AI Taught Me
Break Down Large Tasks
When I asked AI to cover the entire codebase with tests — it failed. Only when I broke the project into small pieces did it start working.
Now I always create a plan: first, I write down the functionality, AI breaks it into steps, and then Claude implements them one by one.
Check Every Test
AI assumes everything works. I now know it doesn't — it often skips errors or "skips" tests to reach a green result.
Out of 350 tests, it could leave me with 80 passing, the rest disabled. That's why I always review the logic and results.
Care for Architecture and State
AI won't understand your architecture if you don't show it. I always provide context — how to connect to the database, how to mock APIs, where to store data.
Watch Out for CSS
AI can write layouts brilliantly, but it doesn't see the visual result. When you ask for "full width," it sometimes breaks the entire page. You need to iterate and test visually.
Security First
Every API key or environment access should be kept outside the repository. Never give AI direct access to production data or files.
My Conclusion
AI isn't magic — it's a partner that needs context and supervision. It works best when you think like an architect, not an operator. This way, you can have both control and speed.
Today, AI isn't a curiosity for me, but an everyday work tool. The best projects emerge when humans and AI truly collaborate.
