Demirer, Horton, Immorlica, Lucier, Shahidi (2026)

View Paper

Two-stage cost-minimization: steps → tasks (AI strategy), then tasks → jobs (job design).

Setup

Stage 1AI Strategy: Steps → Tasks

Each step is executed as Manual, Automated, or Augmented.

A Manual step becomes a manual task; an Augmented step, alone or preceded by Automated steps, forms one AI chain task (length ≥ 1).

Manual step
Automated step
Augmented step
Manual task
AI chain task
Job hand-off

Stage 2Job Design: Tasks → Jobs

Given the task sequence from Stage 1, tasks are partitioned into jobs (one worker per job).
Each job boundary incurs a hand-off cost determined by the final step of that job.

Manual task
AI chain task
Job
Job hand-off

Each step has a skill cost and a time cost for each mode of human (cMi, tMi) or AI-assisted (cAi, tAi) execution. These costs for AI-automated execution are zero. qi and tHi denote the AI's success probability and the (potential) hand-off time cost between workers.

The solver minimizes the firm's cost minimization problem, denoted in Equation 3 in the paper.

Step cMi ($) tMi cAi ($) tAi qi tHi Mode

Derived Tasks

Each Manual step becomes a manual task with its (cM, tM); consecutive Automated steps followed by an Augmented step form one AI chain task whose skill and time costs are the augmented step's (cA, tA) and whose success probability q is the product of the constituent steps' q values. tH is inherited from the task's last step.

Task c ($) t q tH Type

Geometric View of the Optimal Design

Each step is a rectangle: width = time, height = skill. Each job's bounding box has area = wage bill.
Hand-off rectangles appear between non-final jobs. Automated cells use a diagonal-stripe fill — they contribute zero direct cost.

Each task is a rectangle: width = time, height = skill. Each job's bounding box has area = wage bill.
Hand-off rectangles appear between non-final jobs.

Each job is a rectangle: width = time, height = skill. Each job's bounding box has area = wage bill.

Production Cost Breakdown

Per-job wage bills under the cost-minimizing design. The total cost equals the sum of the wage bills across all jobs.

Total cost:
Job Steps Tasks Σ cb ($) Σ tb (time) tH(Jj) Wage Bill ($)
Wage Bill of job Jj = (Σb∈Jj cb) · (tH(Jj) + Σb∈Jj tb).
Total cost = Σj Wage Billj.