Manufacturing AI agent · Intralogistics
Intralogistics & Robot-Fleet Orchestration AI Agent
Assign and sequence material-movement tasks across your fleet, watch congestion, charge and blocked routes, and escalate when the plan cannot be met — inside the zones and speed limits your engineers set.
- Typical delivery
- 4–6 weeks
- Scope
- Task plan only
- Safety stop
- On the vehicle
- After launch
- Agent Care
What this agent does
Plans the moves, never drives the truck
In
- 01
Take movement demand from the MES, the WMS, a line call or a replenishment rule — what moves, from where, by when.
- 02
Read fleet state from the master control — position, load, charge, errors and the routes it has already blocked.
Reason
- 03
Match each move to the vehicles rated for that load, permitted in that zone, and holding the charge to finish it.
- 04
Sequence and prioritise against due times, the cells already waiting, and the congestion the plan is itself creating.
- 05
Plan charging windows inside your charge policy, so the fleet is not flat at the hour the plan needs it most.
Decide
- 06
Detect chokepoints, blocked routes, stranded loads and cells the priority scheme has quietly left waiting.
- 07
Escalate to the area supervisor when the plan cannot be met, instead of dropping the move and reporting nothing.
Out
- 08
Hand an ordered task list to the master control, which is what actually dispatches the vehicles.
- 09
Retain the demand, the fleet state read, the assignment, the escalation and every supervisor override.
Product statement
The agent allocates and sequences tasks and nothing else. Zones, speed limits, routes and safety parameters are engineering changes with a named approver, and the vehicle's own safety system stops it whatever the plan says.
Example workflow
One move, end to end
- 1Move requestedFrom the MES, the WMS, a kanban call, a line signal or a replenishment rule
- 2Fleet state readPosition, load, charge, current order, faults and the routes the master control has blocked
- 3Candidates matchedWhich vehicles are rated for the load, permitted in the zones on the way, and charged enough to finish
- 4Plan sequencedPriority, due time, the cells already waiting, chokepoint load and the charging the rest of the shift needs
No human action required
Stages 1 to 4 run without a person in the loop — reading demand, reading fleet state and sequencing all happen before anything is handed over. A move the fleet cannot meet ends that stretch on the spot.
Goes to the master control as an ordered task list.
Stops with the reason and the cells at risk.
Area supervisor
Reads what cannot be met, decides what waits, what a manual truck takes, and what goes further up.
Approve plan · Re-prioritise · Escalate
Approved — handed back
- 6Task list handed overWritten to the master control as tasks — it dispatches the vehicles under its own traffic rules
- 7Outcome evaluatedCompletions, re-assignments, deadlocks, floor interventions, fairness across cells and charge at handover
Overrides
What the supervisor re-prioritises is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Changing a zone, a route or a speed limit.
- Altering any safety parameter on a vehicle.
- Approving a new or re-surveyed area map.
- Adding or reconfiguring a vehicle in the fleet.
Automation boundary
Agent acts unaided
- ✓Match a move to vehicles rated and permitted to carry it.
- ✓Sequence and prioritise the queue against due times and waiting cells.
- ✓Plan charging windows inside the charge policy your engineers set.
- ✓Escalate what the fleet cannot meet, with the reason and the cells at risk.
The agent writes tasks and priorities to the master control. Safety systems, zones and speed limits are out of reach.
- Overriding a protective stop or a blocked route.
- Returning a vehicle to service after a fault.
- Opening automatic running into a manual-traffic area.
- Halting or restarting the fleet.
Example output
One move, annotated
Everything the agent proposes is attached to the demand it came from and the fleet state it was planned against.
Task assignment · single move
Illustrative example- Move requested
- Full pallet, weld cell 3 to despatch
- Called by
- MES replenishment rule
- Due by
- 14:20
- Assigned to
- Tugger 4, east aisle route
- Confidence
- 91%
- Zones and speed
- Unchanged — set on the vehicle
As received
The demand as the MES called it, and the fleet state the master control reported.
Evidence used
- Live fleet state
- Charge to finish the move
- East aisle queue depth
Why this vehicle
Tugger 4 is the only free vehicle rated for that load, that zone and the distance.
What the score decides
Confidence decides whether a supervisor sees the plan first, never a vehicle's speed.
Action
Approve planRe-prioritiseEscalateValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every movement demand
MES, WMS, kanban call or replenishment rule
Plan inside the layout you already run
Use the plant's own zone map, vehicle capabilities, priority rules and charge policy — not a generic traffic model.
Keep the queue moving without a radio call
Routine moves are matched, sequenced and handed over continuously, instead of being called one at a time by whoever is nearest a screen.
Say a cell will run dry before it does
Moves the fleet cannot meet reach the area supervisor with the reason and the cells at risk, while there is still time to send a manual truck.
Build an evidence trail
Retain the demand, the fleet state read, the assignment, the sequence, the escalation and the supervisor's override — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent, and the fleet's own control layer stays in charge of dispatch. Which of them are in scope is decided in discovery.
Fleet & traffic control
- VDA 5050
- Vendor fleet managers
- Traffic and zone configuration
Plant systems
- SAP EWM
- Oracle
- Opcenter MES
- In-house MES
- Replenishment rules
Vehicles
- AGV tuggers
- Forklift AGVs
- AMRs
- Conveyor and lift handshakes
Agent
Intralogistics & fleet orchestration
- Reads fleet state
- Assigns and sequences
- Escalates the gaps
Floor signals
- Kanban calls
- Andon signals
- Charger telemetry
- Dock status
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and your fleet
Each control wraps the one inside it. A plan clears every layer before it is handed over, and the vehicle's own safety system sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Keep the agent off safety and off the wrong truck
- L3
- Decides when a person is told instead
- L4 – L5
- Keep priority with a person, trail intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole plan — not only the moves that completed.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the task list the master control receives
- E1Final-output evaluationDid the assigned move complete, on the vehicle it was given to?
- E2Step-level evaluationDid it read live fleet state, vehicle capability and zone permissions correctly?
- E3Tool evaluationDid the task reach the master control as planned, once and only once?
- E4Intervention rateHow often did a deadlock need a person to walk over and clear it?
- E5Slice evaluationHow does the plan hold across areas, shifts and traffic conditions?
- E6Business outcomeWere cells fed on time, and was the fleet charged at shift change?
Floor — the material that reached the cell in time
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle. None of them moves a vehicle — the master control dispatches, and the vehicle's own safety-rated detection, speed monitoring and protective stop are what stop it, whatever the plan asked for. When a plan goes wrong the area supervisor halts the fleet from the plant's own fleet stop, vehicles hold where they are with their loads, and the task record shows every move in flight, what was on it and which cell was waiting for it.
Agent lifecycle
Direction of processing →
01 · Retrieval
2 modes
- RF-01
Stale area map
Planned on a layout or an aisle the floor has moved past.
- RF-02
Manual traffic unmodelled
An aisle the plan treats as clear fills with trucks and people.
Stage reads
Demand, fleet state, capability and zone permissions
02 · Reasoning
2 modes
- RF-03
Move the vehicle cannot make
Assigned a load, height or zone the truck is not rated for.
- RF-04
Chokepoint the plan did not model
Vehicles queue and deadlock where the model saw an open aisle.
Stage plans
The assignment, the sequence, the charge window
03 · Hand-over
1 mode
- RF-05
Duplicate move, stranded load
The same move dispatched twice, or a load left mid-route.
Stage hands
The ordered task list, written to the master control
04 · Output
1 mode
- RF-06
One cell starved by priority
The scheme feeds one area well and leaves another waiting.
Stage returns
The plan and the escalation the supervisor reads
05 · Change / Version
1 mode
- RF-07
Charge plan drifts off shift
A rule or calendar change leaves the fleet flat at handover.
Stage tracks
Model, priority-rule, map and fleet-configuration changes
- Sev-1 · a person walks into vehicle traffic
- Sev-2 · the move fails, material sits
- Sev-3 · one area waits, totals look fine
Affected slices
The plan is only as good as the traffic it assumed
Two aisles in the same plant are not the same problem. A dedicated route with no foot traffic runs close to plan; a shared aisle on a busy shift, with manual trucks and people in it, does not. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Mixed pedestrian and manual traffic | 6.4% | 3.8× | Review | |
| Single-aisle chokepoints | 4.9% | 2.9× | Review | |
| Shift change and breaks | 3.1% | 1.8× | Watch | |
| Dedicated aisles, settled area | 1.2% | 0.7× | Normal |
Bar: intervention-rate lift vs. dedicated-aisle baseline · scale 0–4.0× · tick marks 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
A jam someone walked in to clear becomes a planning case
Floor interventions, deadlocks and moves that could not be met come back as replays against the same aisle, the same shift and the same traffic.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Re-assignments, deadlocks and floor interventions cluster on one aisle or one shift.
- 02
Diagnose
Traced to the map read, a vehicle capability, the priority weights, or traffic nobody feeds the planner.
- 03
Improve
Priority rules, the capability table or the charge policy change — anything touching a zone, a route or a speed limit goes to your engineers instead.
- 04
Verify
Replayed against recorded demand and fleet state from that area, including the shift it jammed on.
- 05
Learn
The jam is kept as a replay case, and the aisle goes into the traffic review with your safety officer.
Learn → Detect
The return edge. A change to a zone, a route or a speed limit is an engineering change — assessed and approved by a named person before a vehicle runs on it, and never made by the agent.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, fleet and layout access, planning logic, evaluation, replay against recorded shifts, then supervised running and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and boundary definition.Week 1
- 02Fleet, master-control and interface assessment.Week 1, Week 2
- 03Zone map, routes and speed limits as built.Week 1, Week 2
- 04Vehicle capability and load-type catalogue.Week 2
- 05Demand intake from MES, WMS and line calls.Week 2, Week 3
- 06Fleet-state read and freshness checks.Week 3
- 07Assignment, sequencing and priority rules.Week 3, Week 4
- 08Charge policy and shift-boundary planning.Week 4
- 09Escalation routing and supervisor workflow.Week 4, Week 5
- 10Evaluation suite and recorded-shift replay.Week 4, Week 5
- 11Congestion and deadlock replay harness.Week 5, Week 6
- 12Observability, deployment and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final 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 tier | PilotOne area, one fleet | ProductionProduction fleet integration | AdvancedMulti-area / multi-vendor fleets |
|---|---|---|---|
| Introduced at Pilot | |||
| Task assignment matched to vehicle capability | ✓ | ✓ | ✓ |
| Read-only fleet-state monitoring | ✓ | ✓ | ✓ |
| Escalation when a plan cannot be met | ✓ | ✓ | ✓ |
| Deadlock and congestion replay in evaluation | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Sequencing and priority rules | — | ✓ | ✓ |
| Charge-window planning | — | ✓ | ✓ |
| Master-control task hand-over | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-area and multi-vendor fleets | — | — | ✓ |
| Enterprise controls and cross-area reporting | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on fleet size and vendors in scope, the master-control interface, areas and shifts covered, priority and escalation rules and deployment requirements.
- Separate from build
- Build 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 bring
We build with it
01Your as-built layout, zone map and route plan
→Zone map, routes and speed limits as built
Week 102The fleet, and what each vehicle is rated to carry
→Vehicle capability and load-type catalogue
Week 203Master-control access and the task interface it exposes
→Fleet, master-control and interface assessment
Week 204How a move gets called today, from MES, WMS or the floor
→Demand intake and fleet-state freshness checks
Week 305Your priority rules, and which cells must never wait
→Assignment, sequencing and priority rules
Week 306Recorded shifts, including the ones that jammed
→Evaluation suite, recorded-shift replay and deadlock harness
Week 407Named area supervisors and your engineering-change approver
→Escalation workflow, then supervised running on your floor
Weeks 5–6
Nothing else is required
Deployment, 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 replay harness and the first shifts your supervisors run the plan on.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Workflow discovery, the boundary and your zone map as built
W2Fleet and master-control access, and the vehicle capability catalogue
W3Demand intake, assignment, sequencing and charge-window planning
W4Evaluation suite, escalation testing and slices by area and shift
W5Replay against your own recorded shifts, then supervised running
W6Your supervisors run the plan under your own engineering-change process
- Reading the band
- The feasibility gate and escalation are tested in week 4, before a task list reaches your master control in week 5.
- At the end of W6
- The plan has run on your floor under your supervisors, with zones, routes and speed limits exactly as your engineers left them, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Manufacturing AI agent
Build a fleet-orchestration agent around your intralogistics.
Show us your layout, your fleet and how a move gets called today. We'll replay one of your own busy shifts against your zone map and show you where the plan would have needed a person.
- Typical delivery
- 4–6 weeks
- Scope
- Task plan only, always
- Safety stop
- On the vehicle, always
- After launch
- Agent Care available
Nestack Agents · Intralogistics & fleet orchestration
AGT-MFG-08 · Agent Care available after launch