I'm playing around with the AI reviews, comparing ...
# how-do-ib
Ben Nizette
04/20/2026, 2:09 AMI'm playing around with the AI reviews, comparing against a few other more specialised review options. Graphite is super-conservative. Across a corpus of 26 PRs I've selected, Claude/Greptile/CodeRabbit gave ~50 comments; Qodo and Copilot gave around 120, the original human reviews were around 100 comments. Graphite gave 3. And they were all good comments, but missed heaps of other things that would make it usable day to day. Does anyone else have similar stories? Does anyone perhaps have a custom rule that works well for them, along the lines of "be more thorough, report issues of "Medium" / P2 priority and up"?
✅ 1
p
Pylon
04/20/2026, 2:10 AM(If applicable) To help us more clearly understand what you're trying to do, please provide additional details about:
- Screenshots / debug logs / list of commands you've tried to run so far
- Your workflow
- Current state
Pylon
04/20/2026, 1:46 PMhey ben, thanks for doing that apples-to-apples pass. we intentionally skew toward fewer, higher-signal findings so teams aren’t flooded with low-confidence noise. that means you’ll often land closer to “a handful of solid comments” than “dozens per PR,” compared with tools that optimize for volume. we're also hearing a lot that the right "sweet spot" varies by team, and that's something we're tracking on the product side.
👍 1
b
Ben Nizette
04/20/2026, 9:59 PMRighto @Jocelyn Beaudoin noted, thanks
2 Views