Welcome to my personal blog! I use it to share what I'm currently learning or thinking about, usually on topics related to technology, business, and health.
Trying out Cursor's BugBot PR reviewer
If you're on this article, it's because you've seen BugBot in Cursor's interface or heard about it somewhere, and are curious if it could be useful to you. So let's jump right in!
Once you've activated BugBot on your repositories, it will automatically start reviewing them for you.
Here are some examples of responses you can expect:
Example 1
Example 2
Example 3
Were those reviews useful? Well, here is out BugBot did:
Example 1 was a fake PR. It came from Codex (which I've reviewed here) and I had discarded it because I realized the approach I had asked wouldn't make sense due to some duplication and consistency challenges. The only way to catch that would have been to take a step back to understand how the application as a whole was working. Codex didn't do this and failed to catch the problem. I wouldn't necessarily blame it however as doing so would have also required quite a few more tokens. From what I've read online, most prefer to keep costs low.
Example 2 was from a real change I've asked Codex to do and what about to try out on my end. Cursor BugBot did catch something here! Not a huge problem but absolutely something I would have noticed and fixed myself.
Example 3 was code that I think is ready for production. Good to see we're on the same page!
As you can see in the first two examples, you can click on link to send the bug information to Cursor's agent:
Is BugBot useful?
Automatic AI reviews of PRs are common sense. Just as we have CI to run the tests, we'll reach a point where every commit is analysed by LLMs. A few months ago, I even posted how I use LLMs to review my git diffs.
The problem with those systems is usually twofold: 1. False positives: What are supposedly bugs but aren't. Those aren't too annoying when only the coder sees them, but if they start polluting the discussion feed of PRs, it's a different matter. Currently BugBot only posts texts and doesn't seem to do any edits. Ideally, they would allow the owner of the PR to discard reports to keep things organised. 2. Cost: For now BugBot is free to use and doesn't share the cost of each run. We'll have to wait to know for sure but considering how competitive things are for AI companies, I suspect the cost will be very reasonable.
Finally, it's also important to note that we do not know which model performs the review. This matters because, just like I do, you can easily write a script that pastes your diffs and some context about your repo to o3 or Gemini 2.5 Pro. If BugBot uses a cheaper model, I'm gaining some automation and losing some intelligence, which is not a trade-off I would want right now.
💬 Comments