Repository navigation
Review Process: AI-assisted code reviews #1148
Description
Is your feature request related to a problem?
The current PR review process relies heavily on two human reviewers, leading to complacency and skimming. While PR creation has accelerated with AI, the review process remains a human bottleneck, causing delays.
Describe the solution you'd like
- One reviewer should act as an AI/agentic code reviewer.
- The other reviewer performs a functional review: testing the branch, using Postman collections or the UI, and verifying acceptance criteria without a second code read.
- Implement Akhilesh's suggestion to generate a sequence diagram/implementation plan alongside the PR to highlight design issues without reading the code.
Original issue
Context
We currently assign two human reviewers, but the psychology is "Claude wrote it, it's probably fine" — reviewers skim and approve. PR creation is ~10x faster with AI, but review is still the human bottleneck.
Proposal
- One reviewer becomes an AI/agentic code reviewer.
- The remaining human does a functional review: check out the branch, get the Postman collection from the author (or exercise the UI), run the cases, and verify acceptance criteria / functional requirements — not a second code read. Side benefit: genuine knowledge transfer.
- Akhilesh's addition: generate a sequence diagram / implementation-plan artifact alongside the PR. Many design problems (e.g. three LLM judge calls that could be one) are visible at the high level without reading code; drop into code only where the high-level looks wrong.
Activity
[-]Revamp PR review process: agentic code review + human functional review[/-][+]Review Process: AI-assisted code reviews[/+]