Responses from users suggest that AI tools are helpful in catching obvious bugs, especially algorithmic ones, but struggle with domain-related bugs due to lack of domain knowledge. Some users see value in human code reviews for knowledge sharing and collaboration, while others believe AI can manage many aspects of code reviews, such as identifying best practices and evaluating alternative solutions. However, they also note that AI struggles with predicting how new code may interact with complex business domain logic.
Key takeaways:
- AI-based code review tools can help catch obvious bugs and discrepancies between comment strings and actual code, but they struggle with domain-specific issues due to lack of domain knowledge and experience.
- Code reviews have value beyond just catching bugs - they are also a platform for knowledge sharing and collaboration, which AI tools cannot replicate.
- While human reviewers are necessary for understanding tribal knowledge or undocumented plans, AI can manage other aspects of code review such as identifying best practices, spotting missed bugs, and evaluating alternative solutions.
- AI review tools can identify bad patterns or practices in code or popular SDKs, but they struggle with predicting how new code may interact with business domain logic in highly distributed tech stacks.