Using Opus to review Opus is not futile but they need Astra as a stern older brother figure to mellow them out. I think Opus gets overly too excited by discoveries it forgets to review if the findings are true
RSI will be a more moved goal post than AGI. When we reach our idea of RSI right now, it will look diminutive in comparison to our concept of what RSI is then
2,000 to 10,000 agents is not that much if they are not concurrent. I do pretty routine work in a RLM and if I count each subagent call as an “agent” I can get to 500+ quick in a few hours