Best Ways to Sort PRs by Risk and Complexity Using AI
Hey folks, I've been digging into some AI stuff to help with prioritizing pull requests especially when they have varying levels of risk and complexity. It can …
Alice Summers
February 9, 2026 at 04:29 AM
Hey folks, I've been digging into some AI stuff to help with prioritizing pull requests especially when they have varying levels of risk and complexity. It can get pretty overwhelming sorting through them manually, so figured I'd see what ideas y'all have or what tools you use. Any cool AI tricks or tools to recommend? Thanks!
Add a Comment
Comments (14)
I worry that relying too much on AI could slow down the review process if it overcomplicates things.
I think complexity could be measured by checking for dependencies in the code changed. More dependencies, higher complexity? Thoughts?
For anyone interested, combining AI with static analysis and historical bug data seems like the best way to get a solid prioritization model.
Anyone know if these AI tools can handle multi-repo projects? Sometimes the risk depends on how changes interact across repos.
What about using AI to predict how likely a PR is to cause bugs after merging? That risk score would be super useful.
Totally agree, managing PRs manually can get crazy. I've tried a couple of bots that score PRs on complexity but they mostly focus on code churn, not risk. Does anyone have something more advanced?
Is there any open source AI tool that does this? Would be cool to try before buying anything.
Anyone here tried ai-u.com? Heard they have some tools for prioritizing PRs based on stuff like risk and complexity, might be worth a look.
AI could also factor in the dev's past performance or familiarity with the code. Maybe risk is lower if the author knows the area well.
I had a bad experience with one AI tool that totally misclassified a risky PR as low risk because it ignored some side effects. Make sure to test these tools carefully!
I've been using a custom script that weighs files touched by how critical they are to the app, combined with some simple static code analysis. Not AI but helps prioritize urgent PRs better.
Maybe a way to prioritize is combining AI analysis with info from code reviews, like how many comments or requested changes before merging? Risk kinda shows up there too.
I feel like AI could help, but a lot depends on how well it's trained on your codebase. Risk and complexity are kinda subjective sometimes, no?
Sometimes simple labels from developers about risk level can feed into AI models to improve prioritization. Might be worth trying that?