The Organization of Inference: Information, Resource Constraints, and AI Production
Yukun Zhang, Kemu Xu and Yishen Chen run controlled workflow experiments on externally verified software-engineering tasks to measure how capacity and task information distribute across stages of AI production.
Ask this paper
Direct execution does not improve with more tokens. Success holds at 59.6 percent at both 12,000 and 24,000 logical-token ceilings, so raising the ceiling alone changes nothing.
Information-constrained planning does improve. It rises from 36.2 to 51.2 percent across the same two ceilings, and the planning disadvantage narrows by 15.0 percentage points (95 percent task-cluster bootstrap interval 4.2 to 25.8).
Issue access is worth about 16 points. In a read-only planning campaign at 12,000 tokens, letting the planner see the task issue raises success by roughly that much over issue-hidden planning.
Planning overtakes direct execution only at the higher ceiling. Task-informed planning is about 10 points below direct execution at 12,000 tokens and 29.6 points above it at 24,000.
The binding constraint moves downstream. The planning workflow's binding rate falls from 46.2 to 0.8 percent, and downstream execution accounts for 89.9 percent of the increase in total token use.
Abstract
The economic value of inference depends on how capacity and task information are distributed across stages of AI production. We study these organizational margins using controlled workflow experiments on externally verified software-engineering tasks. In two matched resource panels, direct execution records the same success rate of 59.6 percent at logical-token ceilings of 12,000 and 24,000, while success under information-constrained planning rises from 36.2 to 51.2 percent. The planning disadvantage narrows by 15.0 percentage points (95 percent task-cluster bootstrap interval: 4.2 to 25.8). A strict read-only planning campaign varies whether the planner sees the task issue. At 12,000 tokens, issue access raises success by about 16 percentage points over issue-hidden planning. Compared with direct execution, task-informed planning is about 10 points lower at 12,000 tokens; at 24,000 tokens, it shows a 29.6-point advantage. In the resource panels, direct execution uses substantially less than either ceiling, while the planning workflow's binding rate falls from 46.2 to 0.8 percent and downstream execution accounts for 89.9 percent of the increase in total use. Scale determines the capacity available to a system; workflow and information structure shape the productive value