We had similar experience! We tried CodeGuru Reviewer on one of our Python codebases (around 71 KLOC) and abandoned it after a week.
The feedback was indeed similar to what standard linters and static analyzers can provide with more noise. I have no doubt they can improve their recommendations as they get more feedback and analyze more codebases, but we were not interested to pay to contribute.
We have had much more success with https://deepsource.io/ , which found real bugs and we generally agree with their recommendations. We are currently evaluating whether or not we will include this service as part of our process.
Deepsource is rule-based, but my understanding is that they are starting to use ML to rank recommendations.
The feedback was indeed similar to what standard linters and static analyzers can provide with more noise. I have no doubt they can improve their recommendations as they get more feedback and analyze more codebases, but we were not interested to pay to contribute.
We have had much more success with https://deepsource.io/ , which found real bugs and we generally agree with their recommendations. We are currently evaluating whether or not we will include this service as part of our process.
Deepsource is rule-based, but my understanding is that they are starting to use ML to rank recommendations.