Jarred prompted Claude to try again, and it spent a day and a half coordinating about 60 Claude subagents, which this time went much deeper
真正的变量可能不是模型更聪明，而是搜索预算：650 个想法全废之后，靠 60 个子智能体、一天半、3100 万输出 token 才捞出结果。这更像是算力换命中率，意味着可复制性取决于谁付得起这笔搜索成本，而不是提示词写得多好。