Jev + Opus 5.5 做研究 agent 的决策层
Opus 负责研究并生成候选，Jev 在循环里做重复决策，宿主验证后再继续，扫描 X、GitHub、PDF 与一手来源，减少上下文浪费。
jev + opus 5.5 is a ridiculous combo for research instead of throwing a massive prompt at one model and hoping it checks everything correctly i let opus build the option set and jev make the repeated decisions inside the loop for this demo, the agent can scan: → X posts → GitHub repos → PDFs / lectures → primary sources for verification the flow is simple: opus researches → builds candidates → jev picks → host verifies → loop continues so the expensive model spends time on reasoning, not on making the same tiny decisions over and over again less context waste. fewer unnecessary model calls. cleaner verification. this is what an actual decision layer inside an ai agent looks like ↓
这条是引用着另一条写的：x.com/de1lymoon/status/2103841654423556397