Trying out the paid Bug Finder in Cursor

Wondering what the new Bug Finder in Cursor provides? Then this article is exactly what you're looking for. I decided to take one for the team and spend $1.85 to try out the Bug Finder.

(BTW: If you have a ChatGPT subscription, I've written how to get o1 to do a full review of your code changes in a way similar to Cursor's Bug Finder, but without having to pay every single time, here is the article).

Let's get started with the screen you most likely saw: The Bug Finder shows all the files that are about to be analyzed with the ability to remove some of them to bring down cost. For this test, I had done a minor improvement to the Chrome extension of Galaxy Brain, my read-it-later meets archiving app. When on an article already archived previously, I wanted to change the popup to instead display a link to the article in the application. This implied using creating a simple internal API, and adding some HTML, JS, and CSS in the extension. Reasonably simple stuff. Before pushing my code, I would normally review it but I thought I would skip that step and just ask Cursor!

Once you pay, you need to wait quite a bit for the results to be ready. Considering the cost, they must be running quite a few LLM requests, which indeed takes a while.

Once finished, you get a list of potential bugs with their likelihood:

I was actually surprised to notice the first two were legitimate. The first was because of a broken background color that I was in the process of refactoring but didn't finish. The second bug was about the headers in the fetch function being defined twice. Oops. Cursor even provides a button to directly ask the composer to fix the problem. Pretty neat. The two other bugs were misses, but Cursor was also doubtful.

Overall, the Bug Finder was useful. As I mentioned, I didn't review the code before, but it's possible I would have missed them.

The Bug Finder is a very interesting feature. In its current format, it's not really usable as there's just too much friction involved in having to decide every single time if you want to pay or not. Cost feels quite high considering this was a small commit. However, it does give a glimpse of the features we can expect in the near future. It makes absolute sense to have your code reviewed by an LLM before it gets deployed.

On a side note, I look forward to having services like Cursor being more comfortable with asking for higher price tags for their subscriptions. $200 for ChatGPT Pro makes little sense considering a chat is quite low bandwidth and low context. But Cursor has all the context it needs, it's just a matter of the LLMs being good enough. Can't wait to see how AI Code Editors continue to improve our productivity.