Theme 1 – Cost vs. Quality of AI Review Models
Many discuss whether cheaper models like Luna offer enough value compared to pricier, more capable models such as Astra/Opus, weighing cost per review against bugs found and false‑positive rates.
- “My aim right now is ~$1 per review (must have passing builds first), because it catches enough little things that my time just reading and replying costs more.” – verdverm
- “$0.10 extra per pr review is nothing. What software company is willing to accept worse reviews and less bugs found to save 10 cents?” – StevenWaterman
Theme 2 – Need for Human Oversight and Managing AI Noise
Commenters warn against blindly trusting AI‑generated reviews, noting hallucinations, false positives, and the risk of complacency; they stress that humans must validate or triage AI output.
- “I am hesitant to trust a stupid model to do code review because I become complacent and when it suggests a small change that seems reasonable … I might not notice that it just did a stupid until much later.” – SwellJoe
- “You wouldn't ask an agent to review a PR then just copy/paste the output into the PR, would you?” – nonethewiser
Theme 3 – Structured AI‑Assisted Workflows (Subagents, Orchestration, Iterative Loops)
Users describe sophisticated pipelines where different models act as planners, implementers, and reviewers, often using agent multiplexers, Docker isolation, or snapshot‑based iteration to keep the process reliable.
- “Almost all of my work is now plan, generate, review, plan, generate, review, commit, push.” – jacobgold
- “Generally using Claude Code with Fable 5.1 (high) to plan and implement (Opus 5 (medium) as the implementer subagents), and using Codex with Astra high to review the plan and review the implementers' output.” – aschobel