◇ Could this help me?
Yes, with an important caveat. Adding an AI reviewer to catch issues before merge and prompting it for architecture and correctness over style is a genuinely useful augment, and keeping a human as the final approver (as this video does) is the right posture. The caveat: treat AI review as a complement, never a replacement, for human review and real security testing. An LLM reviewer misses and hallucinates issues, so a passing AI status check must not become false confidence on security-critical code (SQL injection and the like still need dedicated SAST/DAST and dependency scanning). Adopt it as one safety net among several, not the safety net. This is a general workflow practice, not tied to one tool.
Help me add an AI code-review step to my pull request pipeline. First, research the options for an automated reviewer (a published GitHub app such as CodeRabbit or Sourcery, or a custom GitHub Action that calls an LLM): what each is, how it works, who maintains it, and whether it is reputable and safe, and present those findings before installing anything. Then draft a review prompt that focuses on business logic, correctness, SQL-injection vectors, unhandled edge cases, and N+1 queries, ignores style and linting, and flags anything touching payments or data deletions. Finally, outline how to add a required GitHub status check that blocks merges on critical findings while treating warnings as informational. Do not install or enable any tool or status check until I approve, and remind me that AI review complements but does not replace human review and dedicated security testing.