Production schedule repair reads a downtime event, a lot hold, a yield drop or a component slip, re-checks the schedule against material and capacity, proposes the re-sequence that protects committed dates, and releases only the orders whose material, capacity and routing are ready.
Tuesday 14:10, plant 1210. The manufacturing execution system reports the placement machine on line 2 down, and the maintenance lead’s estimate is six hours. The schedule repair agent, a role built from the blueprint, reads the event, the nine open production orders on line 2, three of them with Northbay Logistics commit dates this week, the free capacity on line 3 and the component position from the shortage queue.
It proposes to move the three committed orders to line 3 after a 40-minute changeover and to push two internal replenishment orders to Thursday. Every customer date holds; line 3 needs 1.5 hours of overtime. Version 1 of the rules lets the agent move orders inside the week on its own decision, and a changeover above 30 minutes on another line goes to the line scheduler, who approves it at 14:22 in the production scheduling SmartApp.
SAP S/4HANA returns the new operation dates, the dispatch list on the shop floor updates, and the 15:00 shift handover pack shows the change, the reason and the approver. The overtime waits for the production manager; the maintenance notification already carries its number.
72% of supply chain organisations report AI capabilities in advanced planning and scheduling (The Hackett Group, 14 May 2026) 8, and Gartner expects 60% of supply chain disruptions to be resolved by systems on their own by 2031 (Gartner, 18 May 2026) 3. This is the Produce half of the chain, and the suites shipped for it in 2026.
SAP’s production planning and operations agent, generally available in the first quarter of 2026, automates the material, capacity and scheduling checks before a production order is released, recommends workarounds and releases orders when ready; its manufacturing assistant, generally available in the second quarter, connects shop-floor signals to operational context for faster disruption response (SAP, October 2025 and 14 May 2026) 24 23 25.
Oracle’s production readiness and shift operations workspaces move manufacturing from manual checklists to proactive corrections, with agents that progress routine work within guardrails and surface exceptions for human judgement (Oracle, 9 April 2026 and 29 June 2026) 30 31. Blue Yonder links supply planning to the factory floor in both directions and optimises material availability, capacity, changeovers and throughput together (Blue Yonder, 19 May 2026) 34, and Siemens with Accenture describe agents that analyse late part shipments and support automated rescheduling in minutes (Siemens, product page) 54.
The pain is the largest number in the chain: unplanned downtime costs the Fortune Global 500 about $1.4 trillion a year, 11% of revenue, and 27 hours a month per plant (Siemens, via AEMT, 1 August 2024) 62. Named customer results for scheduling agents are still thin; the C3 AI refinery case reports 2 million constraints and 16 weeks to pre-production (C3 AI, customer story, undated) 53. A production scheduling SmartApp is already in production use at 200 users production · anonymised; the screen exists, the role is a design.
The scheduler runs finite scheduling each morning, in the planning tool or a spreadsheet Gantt, and issues the dispatch list. A breakdown, a hold or a yield loss reaches the scheduler by phone, e-mail or the shift handover, often hours after the event.
The scheduler lists the affected orders, checks component availability and capacity on the alternate lines by hand, drags orders in the Gantt, and calls the material planner about components, customer service about commit dates and the production manager about overtime.
Orders are re-released (a release with missing parts fails in the ERP), the ERP and the execution system are updated, and the change is written into the handover pack. Two to eight hours pass from the event to a confirmed new schedule, an estimate; the rest waits for the daily production meeting.
The afternoon by hand: two to eight hours from the event to a confirmed schedule, an estimate. Illustrative times.
Schedule repair is the Produce end of the chain and the use case with the most roles on the blueprint behind it: eighteen, from finite scheduling to the shift handover pack. It would read the execution system’s events beside the ERP’s orders and capacity, and the production scheduling SmartApp that is already in use would be its approval screen.
Twelve minutes from the event to the approval, as designed; the outcome check follows when the machine returns. Showcase estate; the role is yet to be built.
The market runs re-sequencing at draft-and-approve, with execute within rules for readiness checks and order release: of the vendors cited above, one surfaces the next step to operators, one releases orders once its checks pass, one progresses routine work inside guardrails. The blueprint designs finite scheduling, change impact analysis and dispatch at A3, release authorisation and the hot-lot policy at A2. That is one rung above the market on the re-sequence, and the limits in the scene make it safe: inside the week on its own, across a committed date with a person.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Protect every committed customer date at plant 1210 when the floor changes, at the lowest cost in changeovers and overtime | The production scheduler | New version |
| Data scope | Production orders and operations, work centres and calendars, routings and the changeover matrix, execution and maintenance events, holds and yield; plant 1210 | A. Novak, data lead | New version |
| Rules | The schedule repair policy and its planning table, version 1 | The scheduler with the Planner Agent | New version, simulated first |
| Trigger | An execution or maintenance event, a component slip, a hot-lot request, an adherence breach, and the shift-start and daily runs | The production scheduler | At once |
| Actions permitted | Reschedule an order · move an operation to an alternate work centre · release a ready order · set the hot-lot flag · firm planned orders inside the fence · create a maintenance notification | The production scheduler | New version |
| Limits | Moves inside the current week · changeover under 30 minutes on another line · overtime and commit date changes always to a person · 50 actions a shift · confidence floor 0.80 | The scheduler and the production manager | New version |
| Escalation | A changeover above the limit to the line scheduler · overtime to the production manager · a commit date at risk to J. Ridley · a held lot to the quality engineer | The production scheduler | At once |
| Level | A3 Delegated inside the week; A2 across a committed date or a changeover above the limit | J. Ridley, director | Up on evidence, down in one step |
| Feedback | Every approved, edited and rejected re-sequence is recorded and used; repair estimates are compared with actual repair times | Always on | Fixed |
Every row is a setting or a versioned rule; the level and the limits are the ones a manager signs.
Lima’s agent blueprint is a design of 960 agent roles across five end-to-end chains. It is the map of what Juno can be configured to do, and the first agents are built from it one at a time, the way the PO cancellation agent was.
Additional capability in Plan-to-Produce lists the rest of the map for this chain.
| Measure |
|---|
| Schedule adherence and on-time delivery |
| Minutes from disruption signal to approved reschedule (baseline: hours by hand) |
| Unplanned downtime hours per month (baseline 27 hours, Siemens) |
| Orders released with complete material and capacity |
| Work in progress, cycle time and changeovers |
| Share of reschedules executed within rules, approved, edited or rejected |
Baseline minutes from disruption signal to approved reschedule first, because it is the measure the by-hand path is worst at, hours against twelve minutes in the scene, and the audit trail gives it for every event from the first day. Schedule adherence and on-time delivery are the outcome measures and take a month to move. Orders released with complete material and capacity is the readiness measure, and a failed release in the ERP is its baseline.
Because the limits split it. Inside the week, with every committed date held and the changeover under the limit, the cost of a wrong move is small; across a committed date a person decides, which is A2 for those cases.
The estimate is on the record as a fact the plan assumed. When the machine stays down past it, the trigger fires again and the agent re-plans from the new estimate; the outcome check compares estimates with actual repair times.
It is the approval screen. The SmartApp is in production use today; the agent would put its proposals in that queue, with the reason, the dates it protects and a clock that passes an untouched item to the next desk.
The floor kept its dates, the parameters are current and the plan is right. At month end all of it becomes numbers that finance has to close. Does the same machinery serve the ledger?
Reconciliations, journal anomalies, accruals and variance commentary: the four use cases finance is deploying now, with the evidence.
Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.
