Every new control layer shifts the human role from doing the work to designing the feedback loop. Observe, compare, adjust, repeat. Same pattern since 1780.
A 250-year pattern
While reading OpenAI’s harness engineering post, George (@odysseus0z) spotted the same shape three times in 250 years:
- Watt’s centrifugal governor (1780s). The worker moves from manually adjusting the valve to designing the governor that adjusts the valve.
- Kubernetes. The engineer moves from restarting services to writing the declarative spec the system reconciles toward.
- AI agent harness. The engineer moves from writing code to designing the environments, feedback loops, and constraints that let agents write code.
Norbert Wiener named this in 1948: cybernetics, from Greek kubernētēs, steersman. The structure underneath is always the same — observe actual state, compare with desired state, adjust, loop. Each time, humans shift from rower to steersman. They move up one abstraction layer instead of disappearing.
Why it shows up everywhere
The pattern doesn’t care what domain you’re in. Mechanical engineering, software infrastructure, AI workflows, management, biological homeostasis. Wherever a feedback loop can replace direct manual control, the shape is the same.
The missing link: Maltz (1960)
Maxwell Maltz is the part of the cybernetics chain almost no one names. He took Wiener’s 1948 framework and applied it to human psychology — 60+ years before anyone said “AI harness.” The brain, in his telling, is a goal-seeking servo-mechanism that auto-corrects via feedback.
The steersman role maps onto it cleanly: set the target (self-image, goal), trust the automatic process, don’t micromanage the servo’s execution.
Maltz predates Kubernetes by 54 years and AI harness engineering by 66.
The timeline:
- Wiener (1948) — theory
- Maltz (1960) — human psychology
- Kubernetes (2014) — infrastructure
- AI harness (2026) — agent engineering
The steersman role isn’t new. We keep rediscovering it. When the shift shows up in your field, will you recognize it?
每多一层控制，人的角色就从“做这件事”挪到“设计反馈环”。感知、比较、调整、循环。1780 年以来，一直是同一个模式。
250 年的同一个模式
George (@odysseus0z) 读 OpenAI 那篇 harness engineering 时，看出同一种形状在 250 年里出现了三次：
- 瓦特离心调速器（1780s）。 工人从手动调阀门，挪到设计那个会自动调阀门的机构。
- Kubernetes。 工程师从重启服务，挪到写一份声明式 spec，让系统自己朝它收敛。
- AI agent harness。 工程师从写代码，挪到设计让 agent 写代码的环境、反馈环、约束。
Norbert Wiener 在 1948 年给这个东西起了名字：控制论（cybernetics），源自希腊语 kubernētēs，意思是舵手。底下的结构每次都一样：感知实际状态、跟期望状态比较、调整、循环。每一次，人都从划桨者变成舵手。人没有消失，只是往上挪了一个抽象层。
为什么哪儿都能见到它
这模式不挑领域。机械工程、软件基础设施、AI 工作流、管理、生物稳态。只要反馈环能替掉直接手动控制，形状就都一样了。
漏掉的一环：Maltz (1960)
Maxwell Maltz 是控制论这条线里大家很少点到的一环。他比 AI harness engineering 早 60 多年，就把 Wiener 那套 1948 年的框架挪到了人类心理学上：大脑是一台伺服机构，盯着目标，靠反馈自动修正。
舵手的角色直接套上去：设定目标（自我形象或目标），信任自动过程，别去微管伺服怎么执行。
Maltz 比 Kubernetes 早 54 年，比 AI harness engineering 早 66 年。
时间线：
- Wiener（1948）— 理论
- Maltz（1960）— 人类心理学
- Kubernetes（2014）— 基础设施
- AI harness（2026）— agent 工程
舵手这个角色早就不是新东西了。我们换个领域，就重新发现它一次。等它轮到你这行的时候，你认得出来吗？
Comments留言