The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI
The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI
约束效应:编排设计如何决定企业智能体 AI 的 Token 经济学
Abstract: Agentic AI development today runs on token maxing: buying capability with tokens — longer reasoning traces, more turns, wider tool payloads, bigger replayed contexts — so tokens per task grow faster than task value. Falling per-token prices mask the pattern; total spend rises anyway.
摘要: 当今的智能体 AI 开发正处于“Token 最大化”的陷阱中:通过消耗更多 Token 来换取能力——更长的推理轨迹、更多的交互轮次、更广泛的工具负载、更大的重放上下文——导致每个任务的 Token 消耗增长速度超过了任务本身的价值。单价的下降掩盖了这一模式;但总支出却在不断上升。
We argue the decisive lever against token maxing is the harness: the orchestration layer that assembles context, exposes tools, sequences turns, delegates work, and carries enterprise observability and governance.
我们认为,对抗“Token 最大化”的决定性杠杆是“约束层”(Harness):即负责组装上下文、暴露工具、排序交互轮次、委派工作,并承载企业级可观测性和治理能力的编排层。
We isolate it with a controlled swap: 22 locked evaluation tasks, six foundation models (Claude Sonnet 4.6, Gemini 3.1, Gemini Flash 3.5, Qwen 3.6, GLM 5.1, Palmyra X6), changing only the orchestration layer — a frozen conventional production loop versus the Writer Agent Harness.
我们通过受控替换实验验证了这一点:在 22 个固定的评估任务中,使用六种基础模型(Claude Sonnet 4.6、Gemini 3.1、Gemini Flash 3.5、Qwen 3.6、GLM 5.1、Palmyra X6),仅改变编排层——对比传统的生产循环与 Writer Agent Harness。
Holding models constant, the harness cuts blended cost per task 41% ($0.21->$0.12), median wall-clock 44% (48s->27s), and tokens per task 38% (14.2k->8.8k), with task-completion quality at parity (0.78->0.81, directional at this sample size).
在模型保持不变的情况下,该约束层将每个任务的混合成本降低了 41%(从 $0.21 降至 $0.12),中位运行时间缩短了 44%(从 48 秒降至 27 秒),每个任务的 Token 消耗减少了 38%(从 14.2k 降至 8.8k),同时任务完成质量保持相当(从 0.78 升至 0.81,在此样本量下呈上升趋势)。
Efficiency is model-invariant — every model gets cheaper (33-61%) — while quality gains are capability-dependent: a model’s gain correlates almost perfectly with its baseline strength (r=0.99, n=6), a phenomenon we term harness leverage.
效率提升与模型无关——每个模型的成本都降低了(33-61%)——而质量增益则取决于模型自身的能力:模型的增益与其基准强度几乎完全相关(r=0.99, n=6),我们将这种现象称为“约束杠杆”(harness leverage)。
Quality per dollar rises 82%; task-completions per million tokens rise from 54.9 to 92.0. On this workload the orchestration layer moved cost per task more than the full spread of the model menu did.
每美元的质量产出提升了 82%;每百万 Token 的任务完成数从 54.9 提升至 92.0。在此工作负载下,编排层对任务成本的影响力超过了模型选择本身带来的差异。
We formalize token economics at the orchestration layer (including effective input price under prompt caching), detail the six mechanism families behind the effect — cache-shape discipline to failure-spend governance — compare six widely used agent systems on the same axes, and argue the harness is the one component whose efficiency multiplies across every model an organization runs — present and future.
我们对编排层的 Token 经济学进行了形式化定义(包括提示词缓存下的有效输入价格),详细阐述了该效应背后的六大机制系列——从缓存形态规范到故障支出治理,并在同一维度下对比了六种广泛使用的智能体系统。我们认为,约束层是唯一能够将其效率优势放大到组织所运行的每一个模型(无论当前还是未来)上的核心组件。