Maintenance-Planning AI Agent (Critical Equipment)
Turn condition data and inspection findings into a work list the shutdown window can hold, with the parts dates checked — while the superintendent, the reliability engineer and the statutory position keep every approval.
Take the shutdown window, the asset list and the open defect notifications from the maintenance system.
02
Read condition-monitoring output, inspection findings and the parts position for each critical asset.
Reason
03
Turn condition data and inspection findings into candidate jobs against the asset's own standard job plans.
04
Assemble the work pack — task steps, the isolations the job needs, trades, lifting resources and permits to request.
05
Check each job's long-lead parts against the promised arrival date and the window it is planned into.
Decide
06
Flag jobs the window cannot hold, parts that will not land in time, and statutory work approaching its date.
07
Route what it cannot resolve to the planner, the reliability engineer or the superintendent who owns the window.
Out
08
Present the draft work list, the packs and the parts exposure — as proposals, with the evidence beside them.
09
Retain every reading and finding used, the job plan revision, the rationale and every planner amendment.
→Product statement
The agent plans, assembles and checks. Approving the window, deferring statutory work, issuing an isolation and return to service stay with the named person.
Example workflow
One window, end to end
AgentHuman
1Window openedA fixed window on the shutdown calendar, its duration, the asset and the crews available
2Condition and defects readVibration, oil, thermography and wear data, inspection findings and open defect notifications
3Candidate jobs builtEach finding matched to the asset's standard job plan, its duration, trades and long-lead parts
4Window fitted, parts checkedJobs sequenced against the window, statutory work placed first, parts read against promised dates
No human action required
Stages 1 to 4 run without a person in the loop — reading, matching and the parts check finish before anyone is asked to read anything. Nothing is approved in that stretch.
5DecisionSplits on plan confidence and whether statutory or registered plant is involved
Routine job, parts on site
Reaches the planner as a complete draft.
Statutory work or a parts risk
Goes to the reliability engineer before it goes further.
Planner or maintenance superintendent
Reads the draft work list, the packs, the parts dates and the flagged jobs, then decides what the window will hold.
Accept plan · Amend · Request review
Plan accepted — handed back▼
6Plan routed for approvalHanded to the planner as a proposal; no window, deferral, isolation or return to service is approved by the agent
7Outcome evaluatedWhat was added or dropped at the pre-shutdown review, what overran, and where the parts actually were
Amendments
Scope a planner adds or drops is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Approving a shutdown window or a change to it.
Deferring or cancelling a statutory inspection.
Extending a mandated overhaul or a life-limited interval.
Authorising work on a winder, hoist or registered plant.
Automation boundaryAgent acts unaided
✓Turn condition data and inspection findings into candidate jobs.
✓Assemble the work pack from the standard job plan.
✓Check long-lead parts against the window's dates.
✓Show which jobs fall out of scope when the window moves or shortens.
Write actions run only inside the approval boundaries agreed during implementation. A statutory inspection is never one.
Issuing an isolation certificate or a permit to work.
Certifying a component fit for return to service.
Changing a maintenance strategy, task or interval.
Placing or expediting a long-lead parts order.
Example output
One job in the window, annotated
Everything the agent proposes is attached to the job and the shutdown window it came from.
Draft plan · one job, one windowIllustrative example
Job as raised
Raised from
Long-lead part
Proposed
Confidence
Window
SAG mill pinion change-out
Vibration alarm and inspection
On order
Plan into next window
86%
Not approved by the agent
As receivedThe job as the inspection raised it, what triggered it and where the part stands — nothing on this side is rewritten.
Evidence usedVibration route historyBearing inspection findingJob plan, current revision
Why the parts flagThe pinion's promised date falls inside the window, not before it.
ActionAccept planAmendRequest review
What the score decidesConfidence decides how hard the planner reads the pack, not what the window may hold.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every job competing for a windowCondition alarm, inspection finding or defect
03Planning & packing
Work to the asset's own plan
Use the site's standard job plans, task intervals, statutory dates and the parts position current for that window.
01Approved path
Stop the plan drifting from the plant
The work list is built from the condition and defect position as it stands this week, so the planner argues about scope rather than rebuilding the list.
02Human review
Put the parts risk in front of a person
Jobs whose long-lead parts will not land before the window reach the reliability engineer while the scope can still be changed.
04Build an evidence trail
Retain the readings and findings used, the job plan revision, the parts dates, the confidence and every planner amendment — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Work managementSAP PM · Pronto · Maximo Work orders · notifications
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the shutdown plan
Each control wraps the one inside it. A draft plan clears every layer before a planner reads it, and the approval sits outside all six.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeReturn planning to the maintenance team if evaluations or production signals degrade.Roll back
L5Change controlRecord the readings used, the job plan revision, the proposal, the named approver and every amendment.Record
L4Named-person gateWindow approval, deferral, isolation and return to service stay with the named position.Gate
L3Statutory holdStatutory inspections and mandated overhauls are read from the calendar and placed before discretionary work; any the agent cannot date is raised.Hold
L2Job plan matchSteps, trades and durations are checked against the standard job plan; jobs without one are flagged.Flag
L1Source and readingEach job carries the reading, finding or interval it was raised from.Cite
Model corePlan proposed — candidate jobs, work packs, the parts position and confidence
L1 – L2Decide whether a proposed job may stand
L3Decides what is placed before discretionary work
L4 – L5Keep approval with a person and the record intact
L6Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole plan — not only the work list a planner reads.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the draft plan the planner opens
Depth of coverage ▼
E1Final-output evaluationDid the proposed work list fit the window as it was actually worked?
E2Step-level evaluationDid it read the current job plan, task interval and inspection finding?
E3Tool evaluationDid it resolve the correct asset, work order and parts reservation?
E4Statutory recallWas every statutory inspection and mandated overhaul placed and dated?
E5Slice evaluationHow does plan quality change across specific job cohorts?
E6Business outcomeHow much scope was added or dropped, and what overran?
Floor — the work the crew actually completes in the window
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycleDirection of processing →
01 · Retrieval2 modes
MP-01
Pre-rebuild data read
Trend data from before an overhaul is treated as current.
MP-02
Statutory date not surfaced
A due inspection is not read from the calendar.
Stage gathersCondition data, findings, job plans and the parts position
02 · Planning2 modes
MP-03
Sensor fault becomes a job
A faulty transducer raises work that takes window time.
MP-04
Real defect scored as noise
A finding is dismissed and never reaches the work list.
Stage proposesCandidate jobs, durations, resources and the sequence
03 · Verification1 mode
MP-05
Isolation missing from pack
The pack omits an isolation the job actually needs.
Stage checksStatutory dates, parts arrival and window capacity
04 · Handoff / write1 mode
MP-06
Parts date read as firm
A promised date is presented as a confirmed one.
Stage presentsThe draft plan the planner reads and the pack it carries
05 · Change / Version1 mode
MP-07
Silent job-plan regression
A job-plan change moves durations with no record.
Stage tracksModel, prompt, job-plan and task-interval changes
Sev-1 · a crew or a statutory duty is exposedSev-2 · a wrong plan reaches the plannerSev-3 · planning degrades, more is amended
Plan quality tracks how much history the asset has
Plan quality is a function of history: if the site has done the job many times the estimate holds; if the machine came back from a rebuild last month, or the part is still on order, it does not. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Rebuilt assets, no history
4.2%
2.8×
Review
Newly instrumented assets
3.3%
2.2×
Review
Long-lead parts on order
2.6%
1.7×
Watch
Routine conveyor change-outs
1.1%
0.7×
Normal
Bar: amended-or-dropped job rate, lift vs. routine change-out baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
One overrun changes the job plan, not just the schedule
A job that overran is not closed by fixing one window. The job plan, the alarm threshold or the parts rule behind it is changed under change control.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Added-or-dropped scope or overruns move in a cohort.
02Diagnose
Either the reading was wrong or the duration was — the window says which.
03Improve
The job plan or parts rule goes through change control, with an approver.
04Verify
Re-planned against held-out windows from the affected cohort.
05Learn
That window is kept as a regression case and the rule enters the planning standard.
Learn → DetectThe return edge. A change to a task interval is a strategy change, made by a named approver.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, assets and calendars, planning and packing logic, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Workflow discovery and automation-boundary definition.
02Asset register and criticality assessment.
03Statutory inspection and overhaul calendar.
04Condition-monitoring and defect data access.
05Standard job plans and duration estimates.
06Work-pack assembly and isolation listing.
07Long-lead parts and supplier date checks.
08Window fitting and scope-change costing.
09Evaluation suite, statutory recall and regression cases.
10Planner review and amendment capture.
11Work-management integration and write-back.
12Observability, deployment and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
Capability✓ in scope · — not at this tierPilotOne asset class, one siteProductionProduction maintenance integrationAdvancedMulti-site / multi-commodity
Introduced at Pilot
Candidate jobs from condition and defects✓✓✓
Work-pack assembly and isolation listing✓✓✓
Long-lead parts and supplier date checks✓✓✓
Statutory and overhaul dates placed first✓✓✓
Window fitting and scope-change costing✓✓✓
Approval stays with the named person✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Planner review workflow and write-back—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multi-site and multi-commodity calendars——✓
Multiple maintenance and planning systems——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on condition-data sources, maintenance-system and parts integrations, the number of sites, asset classes and shutdown calendars, approval controls and deployment requirements.
Separate from buildBuild pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bringWe build with it
01Your shutdown calendar and the windows already committed→Statutory inspection and overhaul calendarWeek 1
02The asset register and how you rank criticality→Asset register and criticality assessmentWeek 1
03Access to the maintenance system, condition data and stock→Condition-monitoring and defect data accessWeek 2
04Your standard job plans, durations and trade requirements→Standard job plans and duration estimatesWeek 2
05Supplier lead times and how a promised date gets confirmed→Long-lead parts and supplier date checksWeek 3
06Real shutdowns that overran, and the scope you dropped→Evaluation suite, statutory recall and regression casesWeek 4
07Named planners and a reliability engineer to review plans→Planner review workflow, then pilot plans and production validationWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the evaluation work and the first plans a planner reviews.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Shutdown calendar, asset criticality and the automation boundaryW2Maintenance system, condition data and the parts position wired inW3Job-plan matching and work-pack assembly on the first asset classW4Evaluation suite, statutory recall and scope-change costingW5Work-order write-back, supervised plans and targeted correctionsW6A window planned and worked, scope changes counted, then handover
Reading the bandThe statutory calendar is tested against last year's actual dates in week 4. A plan drawn before that test would be a plan drawn on an untested calendar.
At the end of W6Draft plans have been through a window the site actually worked and what was added or dropped has been counted, then Agent Care takes over monitoring.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Mining AI agent
Build a maintenance-planning agent around your shutdown calendar.
Show us your shutdown calendar, your standard job plans and how a defect reaches a work order today. What comes back is one window planned to your own job plans, with each long-lead part marked against a promised date rather than a confirmed one.
