4-LLM review panel on 3 PRs (Sonnet, Opus 4.7, Opus 4.8, Codex/gpt-5.5).
Biggest finding wasn't 4.7 vs 4.8.
It was that Codex caught 2 HIGH-severity bugs that ALL three Claudes missed.
Cross-family code review is additive, not redundant.
@henrytdowling I have found the retrospective to be the most powerful piece. Sprint meetings, in a way. More ad hoc though, currently. When needed, all planning comes from multiple agents iterating until the path forward is solid from multiple perspectives.
@arne__ness It's not about it moving. It's the efficience of complicated text layouts. Text layout is one of the most pain in the ass programming tasks.