Engineering-Change & Configuration-Integrity AI Agent

What this agent does

Checks across the systems, not inside one

A change is raised, and the parts, documents, routings and open orders it touches are assembled with it.

A packet arrives thin, and the missing categorisation, justification and implementation plan are named.

Reason

A change waits on one over-subscribed reviewer, and the queue is chased there, not across the board.

A change is released in PLM, and the ERP change master, the MES routing and the work instruction are read back.

An effectivity date passes, and the revision the cell is building is compared with the one the order carries.

Decide

An effectivity type differs between systems — date here, serial there — and the mismatch is raised, not averaged.

An open order cites the superseded drawing, and the supplier side is in scope — SAE EIA-649C (2019) puts it there.

Out

A divergence is found, and it is routed with both records behind it to the named change authority.

Execute write actions only inside the approval boundaries agreed during implementation.

→ Product statement

The agent reports where the change stands; ISO 10007:2017 leaves the verification with a dispositioning authority, and a named person dispositions the stock.

Example workflow

One change order, raised to landed

AgentHuman

1Change request raisedPLM change request, supplier request, deviation or customer-driven change

2Impact assembledAffected parts, documents, routings, work instructions, open orders and stock on hand

3Systems comparedRevision, effectivity type and date per system, gaps and confidence

4Controls appliedRevision-string checks, effectivity-type comparison, open-order checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is released at any of them — the agent is reconciling, and the authority's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the change authority to release.

Low confidence

Adds a configuration-lead read first.

Authority release

The change is held with its status per system, its open gaps and the confidence.

Approve · Hold · Send to board review

Approved — released for implementation▼

6Downstream systems updatedOnly where write access and approval policy allow it

7Outcome evaluatedTime to landed, systems behind at effectivity, and divergence found after build

Holds

Every change the authority holds is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Deciding a change may go ahead.

Dispositioning stock built to the old revision.

Waiving a customer's approval of a change.

Setting the effectivity date a change takes.

Automation boundaryAgent acts unaided

✓Assemble the change packet from the systems of record.

✓Carry each revision forward with the system it was read in for the named owner.

✓Compare revision, effectivity type and date across systems.

✓Mark the gaps the authority should weigh, and hold the change.

Any write happens inside the boundaries agreed at implementation, never ahead of release.

Judging whether a trial run is needed.

Closing a change while re-approval is open.

Granting or extending a deviation.

Changes to effectivity, routing or approval rules.

Example output

One change order, annotated

Everything the agent reports is attached to the system it was read in.

Configuration-status output · single changeIllustrative example

Change

Finding

Revision live

System of record

Confidence

Effectivity

Housing ECO · one plant

The work instruction still serves the revision this order supersedes

Rev A

MES work instruction

84%

Date here, serial in ERP

As receivedRead from PLM, ERP and MES as they stand — nothing on this side is inferred by the agent.

Evidence used Released change order ERP change master MES revision in force

Why this is flaggedThe revision is released and this system is still serving the old one.

ActionApproveHoldSend to board review

What the score decidesBelow the configured threshold the change picks up a configuration-lead read first.

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 changeFrom the change order

03Reconciling

Compare beyond the PLM

Draw on the released order, the baseline the board agreed, and the systems the PLM does not reach.

01Approved path

Released is not the same as landed

Routine changes arrive with their status per system already assembled.

02Human review

Send review to the divergence

Open gaps and low-confidence comparisons are marked, so the authority's read starts where systems disagree.

04Build an evidence trail

The change, the system it was checked in and the authority who released it stay on the order.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

PLM and change controlTeamcenter · Windchill
3DEXPERIENCE · Arena

ERP and item masterSAP S/4HANA · Oracle
Infor · Dynamics 365

MES and shop floorSiemens Opcenter · Plex
Tulip · Apriso · DELMIA

Agent

Change & configuration integrity

Reads each system
Compares the revision
Holds for release

Supplier portal and qualityAriba · Coupa · SupplyOn
PPAP and drawing packages

Observability & evaluationOpenTelemetry · 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 the line

Each control holds the one it surrounds, six in all. What no control holds is in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modePull reconciliation back to reporting-only when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, effectivity-rule and connector changes.Track

L4TraceabilityRecord each system read, the revision found, the gaps and the release time.Record

L3Authority releaseHold changes for the named authority; it governs release, not whether the release is right.Gate

L2Policy guardrailsTest comparisons against configured effectivity and baseline rules; a failure returns the change.Restrict

L1Confidence thresholdsRoute low-confidence comparisons to a configuration read before the authority sees them.Require review

Model coreStatus produced — revision and effectivity per system, open gaps and confidence

L1 – L2Test whether a status may stand

L3Puts the release in an authority's hands

L4 – L5Keep the change and the system behind it

L6Reverts to status reporting when signals degrade

How Nestack evaluates it

Evaluate the reconciliation across systems — not only the change inside the PLM.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the status the board sees

Depth of coverage ▼

E1Final-output evaluationDid each system's revision match what the order released?

E2Step-level evaluationDid the agent read the right systems, baseline and effectivity rules?

E3Tool evaluationDid it read and write the correct change and the correct system?

E4Confidence calibrationDo low-confidence comparisons actually attract more holds?

E5Slice evaluationHow does performance change across specific system pairs?

E6Business outcomeHow many divergences were found only after parts had been built to the old revision?

Floor — the revision the cell builds

Failure modes

Where each failure originates in the agent

Seven failure modes, set where each one begins.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

NG-03

Connector-limited view

Systems outside the connector set go unread.

Stage gathersChange orders, item masters, routings and portals

02 · Reasoning2 modes

NG-04

Effectivity type ignored

A date rule is read as equal to a serial rule.

NG-06

Closed before it landed

Workflow closure is taken as implementation.

Stage proposesRevision, effectivity and gaps, with confidence

03 · Tool / write2 modes

NG-02

Old stock left standing

Superseded stock sits with no named disposition.

NG-05

Supplier kept on the old

Open orders keep citing the superseded drawing.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

NG-01

False agreement reported

Systems are called matched where they differ.

Stage returnsThe status the authority releases the change from

05 · Change / Version1 mode

NG-07

Silent scope regression

A rule change narrows which systems are compared.

Stage tracksModel, prompt, effectivity rules and connectors

Sev-1 · release recorded outside the boundary Sev-2 · the cell builds a superseded revision Sev-3 · a system is unreadable, change holds

See the complete Manufacturing failure-mode catalogue →

Affected slices

One system can be behind while the rest agree

A company-wide agreement rate can look settled while one or two systems carry most of the divergence. Nestack reports the divergence rate by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
MES work instructions	6.7%	3.5×	Review
Supplier portal and open POs	4.6%	2.4×	Review
Inspection and control plans	3.3%	1.7×	Watch
ERP item master and BOM	1.8%	0.9×	Normal

Bar: divergence-rate lift vs. ERP-item-master baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

Each cycle ends in a new case

A cycle closes when the change that never landed is a regression case the next release must pass. That suite is what the next order tracked is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Divergence rate rises in one system.

02Diagnose

The cell that built to a revision the change board retired in April is read back until one cause is left.

03Improve

The fix gets a number, and the change orders that raised it are held against it.

04Verify

No release while a touched effectivity case is outstanding; it reruns until green.

05Learn

The case stays for good, and the systems it spans are recorded against it.

Learn → DetectThe return edge. The next order is tracked against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, cross-system reconciliation, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Change workflow discovery and boundaries and boundary definition.

02PLM, ERP, MES and portal assessment.

03Effectivity, baseline and route mapping and rule mapping.

04Change-order ingest and system mapping.

05Cross-system comparison and binding.

06Confidence scoring and gap routing.

07Authority release workflow.

08ERP, MES and portal integration.

09Effectivity and revision cases.

10Guardrails and release controls.

11Change-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 change board, one plant ProductionProduction PLM, ERP and MES AdvancedMultiple plants / systems

Introduced at Pilot

Reconciliation across your systems✓✓✓

Authority release✓✓✓

Revision-agreement baseline✓✓✓

Introduced at Production

Reporting by system—✓✓

Release workflow in your systems—✓✓

Approved write-back—✓✓

PLM and ERP integration—✓✓

Introduced at Advanced

Multi-customer approval routes——✓

Multi-stage board approvals——✓

High change volume——✓

Multi-system configuration controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, 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 change process and who sits on the board → Change-order ingestion and system mappingWeek 1

02Representative change orders, closed ones included → Reconciliation baseline, revision and effectivity bindingWeek 2

03Your effectivity rules and customer approval routes → Effectivity, baseline and customer-route mappingWeek 1

04Access to relevant APIs, feeds or exports → PLM, ERP, MES and portal assessment, then integration setupWeek 2

05Changes you would not want released → Divergence cases and failure-mode testingWeek 4

06What no release may go out without → Confidence scoring, gap routing, guardrails and release controlsWeek 3

07A named change authority to release changes → Authority release workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The weeks are drawn from the change board, so the fifth carries evaluation and the pilot together.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Change workflow discovery, route mapping and the automation boundary W2System integration and the reconciliation baseline W3Cross-system comparison, confidence logic and release controls W4Evaluation suite, effectivity checks and failure-mode testing W5Portal and MES integration, pilot changes and targeted corrections W6One change board run under the configuration lead, then Agent Care handover

Reading the bandA bar runs across the weeks its work is named in and no further. The week 5 overlap is real.

At the end of W6Validation closes on live orders, and Agent Care picks up monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Manufacturing AI agent

Build a configuration-integrity agent around your change board.

Show us one released change and where it should have landed. Ask the planner who cut last week's orders which revision his routing gave him — that answer is the one we keep current.

Discuss this agent →

Nestack Agents · Engineering change & configuration integrityAGT-MFG-19 · Agent Care available after launch