Initial thoughts on using Cursor

I do not know how to write Vite plugins. Thanks to Cursor, I've made one in about 5 minutes.

Cursor is an "AI-powered code editor". There was some initial hype about it, but I decided to give it a go yesterday.

Articles in this blog are written in Markdown, and are stored in the articles directory. As part of rebuilding this blog, I wanted to experiment with pre-loading all the articles in the Remix server on startup, in order to speed up page load time.

My initial approach was to write a function that would load all files in the articles directory via import.meta.glob. Unfortunately, this caused the site to hit a startup CPU limit on deployment.

So instead, I decided to compile all the articles into a single file at build time, and import them in the server. This is theoretically a much faster process, as it avoids the overhead of dynamic imports at runtime. This basically meant I needed to write a Vite plugin.

I have no experience writing Vite plugins. I know how to use Vite plugins, and I know how they work in principle. But I've never had reason to write one, and so I've never taken the time to learn how. But that didn't matter. I just asked Cursor to write a Vite plugin for me, and it did.

Worked on the first attempt too, which was a pleasant surprise.

As a long-term user of Jetbrains products, I was fairly skeptical about LLMs for programming. Claude could generate code reasonably well, but lacking the context of the project meant I had to do a lot more heavy lifting to make Claude-generated code work. This wasn't the case with Cursor.

I will daily-drive Cursor in the coming weeks before making any final judgements, but I am cautiously optimistic about the future of AI-driven IDEs.