The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI

Short summary

This paper argues that the orchestration harness—not the model—is the decisive lever against token maxing in enterprise agentic AI. In a controlled swap across six foundation models and 22 tasks, the Writer Agent Harness cut blended cost per task 41%, wall-clock time 44%, and tokens per task 38% while maintaining quality parity. The effect is model-invariant: every model gets cheaper, and quality gains correlate with baseline model strength (r=0.99).

• •Orchestration layer cuts cost per task 41% and tokens 38% across six models
• •Quality per dollar rises 82%; effect is model-invariant
• •Harness moved cost more than the full spread of the model menu did

Generated with AI, which can make mistakes.