一篇发表在Hacker News上的文章通过比喻阐述了Agent AI的两种核心工作模式[1]。温室模式适用于目标不明确的探索性工作，此时AI生成多个可能的结果方案，由使用者事后评估选择[1]；透镜模式则适用于目标明确的任务，AI集中所有努力直指既定目标[1]。两种模式都构成了合法的生产力工具，区别在于应用场景的不同[1]。
文章强调，真正的高级技能在于识别何时采用哪种模式，以及在两种模式之间灵活切换的能力[1]。领域专业知识在两种模式中均发挥关键作用：在温室模式中用于筛选值得保留的方案，在透镜模式中用于识别偏离既定目标的输出[1]。
A framework has been presented distinguishing two fundamental approaches to how agentic AI systems operate in productive contexts [1]. The "greenhouse" mode applies to exploratory work where the desired outcome remains undefined; in this mode, AI generates multiple potential results that are subsequently evaluated and filtered for merit [1]. Conversely, the "lens" mode targets work with clearly established objectives, channeling all effort toward achieving a predefined goal [1].
The framework emphasizes that both modes serve as legitimate productivity tools, with the critical skill lying in recognizing which mode suits a given task and transitioning fluidly between them [1]. Domain expertise—functioning as an "oracle"—plays a pivotal role in both approaches: it serves to identify and retain promising options in greenhouse mode, while in lens mode it helps detect when outputs deviate from the intended target [1].