Hacker News Reader

Comment by felipeerias | Hacker News Reader

I use Claude Code with a MCP that lets it communicate with Codex and tell it to “iterate until both of you are happy”.

The agents then go for several rounds criticising each other plans and implementations, catching big and small issues on each other’s work. The end result is not perfect, but it is a lot better than what I can get from relying on only one model.

View on HN

I do the same with Gemini. I have a skill that teaches any AI to use the antigravity cli (agy) to communicate with Gemini and ask to review or check decisions

There's some real research-backed evidence that this approach elevates output quality, and it was even discussed a few months back: https://news.ycombinator.com/item?id=47169627

Anthropic confirmed it in their recent report https://news.ycombinator.com/item?id=49316271

> We expect that agents coordinating in the wild will act in higher variance ways than we see here, because they’ll have different backgrounds and therefore different contexts. They also, presumably, won’t all be Claudes.

I do use a second Claude (both Opus) to review the first one's output - it almost always finds worthwhile (often critical) observations. They have different 'perspectives', one wants the job done, one wants to find problems, is how I had it explained.

Claude chat itself called it 'Claude-on-Claude action', which I found cute.

btw, there's a dedicated plugin for similar setup https://github.com/openai/codex-plugin-cc

I do a similar thing manually and you have to watch it very carefully to avoid over engineering, ironically you need judgement here not vibe, otherwise (codex especially) will ratchet the project into some complicated enterprise grade over engineered monstrosity by eg coming up with minor visual-only concurrency bugs that would need you to be Facebook scale.

Reply on news.ycombinator.com