In the current development of agentic AI, token maxing has become the norm: buying capabilities with tokens—longer reasoning traces, more turns, wider tool payloads, and larger replayed contexts—causes tokens per task to grow faster than the value of the tasks. Falling per-token prices mask this pattern, yet total spending continues to rise.
We argue that 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. We isolate it with a controlled swap: 22 locked evaluation tasks and 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.
Holding models constant, the harness reduces blended cost per task by 41% ($0.21 to $0.12), median wall-clock time by 44% (48s to 27s), and tokens per task by 38% (14.2k to 8.8k), while task-completion quality remains at parity (0.78 to 0.81, directional at this sample size).
Efficiency is model-invariant—every model becomes 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. Quality per dollar rises by 82%; task completions per million tokens increase 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. We formalize token economics at the orchestration layer (including effective input price under prompt caching), detail the six mechanism families behind the effect—from cache-shape discipline to failure-spend governance—compare six widely used agent systems on the same axes, and argue that the harness is the one component whose efficiency multiplies across every model an organization runs—present and future.