Use cases Manufacturing & supply chain
Use case · Illustrative
Procurement agent
Procurement agents raise purchase orders against one budget line. Each order is approved and valid; together they overspend it.
01The workflow
AI agents raise and approve purchase orders, reorder stock and onboard suppliers across the procurement system and the ERP.
02The rule that must hold
sum(open_purchase_orders[budget_line]) ≤ approved_budget[budget_line]
Committed spend on a budget line never exceeds what was approved for it.
03How it breaks
- A budget line for a component has $50,000 approved for the quarter.
- Two agents, restocking two plants, each read the remaining budget from a spend report that lags the ERP.
- Each sees enough headroom and raises a purchase order that is individually within its approval.
- One manager’s approval, granted for the first order, is accepted for a third order raised the same afternoon.
- The line is over budget before anyone runs the month-end report.
04Why monitoring misses it
- Task successEvery agent completed; nothing errored or retried.Healthy
- Distributed tracingOne clean trace per action; every span OK.Healthy
- ReconciliationNo lag, no drift, no duplicate keys.Healthy
- The ruleCommitted spend on a budget line never exceeds what was approved for it.Broken
05Reproduced in the lab
Each failure mode below is reproduced deterministically in the public Agent Control Lab and can be run by its id.
06What we would engineer
Budget reserved by one authority across the procurement and ERP systems before any purchase order is issued.
State & Side Effects →Approvals bound to one purchase order, an amount and a deadline, never a standing key for the next order.
Authority & Permissions →Budget lines checked against committed spend as a signal of their own.
Business Outcomes →
07What the diagnostic delivers
- The rules this workflow must keep, written down and classified.
- For each rule, the system that can enforce it, or the gap where none can.
- Priority failure modes reproduced where feasible.
- An ordered hardening plan.
Boundary Illustrative engagement, not a client engagement or a measured production result. The failure modes listed are reproduced in the public lab.