Coding with Cursor AI

In my previous post I mentioned that I was coding a project entirely with AI, but I was not using Cursor at that time. Just Zed with Ollama.

I decided to try Cursor only with claude-3.5-sonnet and it was a GREAT experience. Here are my thoughts.

Good things:

• Interprets the given context really well. I added a markdown file to the Composer with info and a JSON sample. It was able to create an initial implementation of how to fetch and visualize the data with its attributes. I was even able to specificy main keys to be shown on a table alongside other additional and optional information.
• Generated code was vast. I ask for a .tsx page plus some react components with some specifications of how to parse and visualize the code and it generated a lot of code. I was using a Tanstack Table with Shadcn UI.
• Everything is fast. cmd + k or cmd + l (chat) or cmd + i and just tab tab tab was REALLY fast. Just fast!

Caveats:

• It rarely failed implementing what I wanted… but when it did, I had to read a lot of code.
• …also I didn’t know the code. It did what I wanted, but it was coded by somebody else so I wasn’t fully aware of every line of code. It was like doing a feature fix on your first day on a new job with.

It definitely intrigues me how does this work behind the scenes…