Main site

Energy AI agent · Asset maintenance

Grid & Asset-Maintenance AI Agent

Read the inspection record, detect against the criteria your engineers configured, bind every finding to the frame it came from, and hold each one for the named engineer who dispositions it.

Typical delivery
4–6 weeks
Deployment
Your stack
Named engineer
Pre-disposition
After launch
Agent Care

What this agent does

A detection is not a condition assessment

In

  1. 01

    Ingest inspection imagery, LiDAR passes and patrol records from supported asset and inspection systems.

  2. 02

    Normalise frame, pass and asset identifiers, and carry each capture forward with the record it came from.

Reason

  1. 03

    Detect against the criteria configured for the asset class, and score each finding.

  2. 04

    Apply the utility's own severity bands, clearance rules and asset-class thresholds.

  3. 05

    Bind every detection to the frame, the structure and the inspection it came from.

Decide

  1. 06

    Mark what the imagery did not cover — the angle not flown, the span obscured, the pass cut short.

  2. 07

    Route every detection to the named engineer who dispositions it.

Out

  1. 08

    Retain the imagery, the detection, the engineer's disposition and the outcome against the asset.

  2. 09

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

Product statement

The agent proposes detections; a named engineer dispositions each one, and the maintenance programme — not the detection — is what the auditor asks to see.

Example workflow

One detection, frame to disposition

  1. 1Inspection record receivedAerial pass, ground patrol, thermal scan or LiDAR survey
  2. 2Asset context assembledStructure, circuit, asset class, prior findings and capture conditions, each with its source
  3. 3Detection proposedAsset, defect class, frame and confidence
  4. 4Controls appliedCriteria checks, coverage checks, prior-finding checks and confidence threshold

No human action required

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

5DecisionBranches at the confidence threshold
High confidence

Goes to the named engineer to disposition.

Low confidence

Adds a second-reader read first.

Engineer disposition

The detection is held with its frame, its coverage gaps and the confidence.

Confirm · Reclassify · Send to second reader

Dispositioned — sent to the programme

  1. 6Asset systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedCorrection rates, coverage gaps, field verification and defects surfaced later

Corrections

Every engineer correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Dispositioning a defect, or clearing one from the list.
    • Changing an inspection interval, or the coverage flown.
    • Declaring an asset, a span or a structure healthy.
    • Scheduling, deferring or cancelling maintenance work.

  • Automation boundary

    Agent acts unaided

    • ✓Read the inspection record against the configured criteria.
    • ✓Bind each detection to the frame and the asset it came from.
    • ✓Carry the pass, the imagery and the capture conditions.
    • ✓Mark what the imagery did not cover, and hold for an engineer.

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

    • Closing a maintenance record or an unresolved issue.
    • Recording that an annual inspection was performed.
    • Deciding that a span met its clearance distance.
  • Changes to detection criteria or severity rules.

Example output

One detection, annotated

A detection rests on a frame, and the frame is the whole of what the agent saw.

Detection output · single asset

Illustrative example
Asset
Lattice suspension tower
Detection
Corrosion at a suspension-clamp fitting, tower body clear
Frame
AR-0417
Source of record
Aerial pass on file
Confidence
88%
Held for
Named engineer, not dispositioned

As received

Taken from the inspection pass and the asset register — nothing on this side is written by the agent.

Evidence used

  • Aerial pass frame
  • Asset register entry
  • Prior structure finding

Why it is held

A fitting this small sits at the edge of what imagery resolves — the engineer looks.

What the score decides

Below the configured threshold the detection picks up a second reader before.

Action

ConfirmReclassifySend to second reader

Value

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 detection

From the inspection record

03Detection

Detect from the record

Draw on the criteria your engineers configured: prioritising what to cut is legitimate, prioritising whether to inspect is not.

01Approved path

Detect it, do not clear it

Routine passes arrive already read, and neither the coverage nor the prescribed intervals move.

02Human review

Send review to the risky findings

Coverage gaps and low-confidence detections are marked — vegetation contact has caused outages and fires, so a miss is not a workload problem.

04

Build an evidence trail

The detection, the imagery behind it and the engineer who dispositioned it stay on the asset.

Integrations

Typical integrations

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

  • Asset and work management

    • Maximo
    • SAP PM
    • Copperleaf
    • Cascade

  • Inspection capture

    • Drone and helicopter passes
    • LiDAR
    • thermal
    • ground patrol

  • GIS and network model

    • ArcGIS Utility Network
    • Smallworld
    • circuit models

Agent

Grid asset maintenance

  • Reads the record
  • Proposes detections
  • Holds for disposition

  • Vegetation programme

    • Work plan and cycle records
    • Contractor cut records

  • 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 the work order

The layers nest. What each one misses is named in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull detection back to proposal-only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, criteria and severity-band changes.Track
L4TraceabilityRecord the frame, the detection, the coverage gaps and the disposition time.Record
L3Engineer dispositionHold detections for the named engineer; it governs the disposition, not whether the asset is sound.Gate
L2Policy guardrailsTest detections against configured criteria, asset class and coverage rules; a failure returns the finding.Restrict
L1Confidence thresholdsRoute low-confidence detections to a second reader before the engineer sees them.Require review
Model coreDetection proposed — asset, defect class, frame, coverage gaps and confidence
L1 – L2
Test whether a detection may stand
L3
Puts the disposition in an engineer's hands
L4 – L5
Hold the imagery the detection rested on
L6
Reverts to raw inspection records when signals degrade

How Nestack evaluates it

Evaluate the inspection workflow — not only the finished detection.

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

Surface — the finding the engineer reads

  1. E1Final-output evaluationDid the detection match what the frame actually shows?
  2. E2Step-level evaluationDid the agent use the right pass, asset record and configured criteria?
  3. E3Tool evaluationDid it read and write the correct asset and the correct field?
  4. E4Confidence calibrationDo low-confidence detections actually attract more engineer corrections?
  5. E5Slice evaluationHow does performance change across specific asset classes?
  6. E6Business outcomeHow many detections were corrected, and how many defects surfaced later?

Floor — the outage the utility answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • HB-03

      Frame the pass never took

      The agent cannot see what the imagery did not capture.

    Stage gathers

    Inspection record, asset register and prior findings

  2. 02 · Reasoning

    2 modes

    • HB-04

      Condition read from one frame

      A single view is treated as an assessment of the asset.

    • HB-06

      Criteria run on the wrong class

      Thresholds set for one asset class are applied to another.

    Stage proposes

    Asset, defect class, frame and confidence

  3. 03 · Tool / write

    2 modes

    • HB-02

      Coverage cut by risk score

      Inspection coverage is reduced where the model scores risk low.

    • HB-05

      Detection loses its frame

      A finding reaches the record without the image behind it.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • HB-01

      Small fitting missed

      A bolt, clamp or fitting is not detected in the frame.

    Stage returns

    The detection the engineer dispositions

  5. 05 · Change / Version

    1 mode

    • HB-07

      Silent criteria drift

      A model or criteria change narrows what the agent will flag.

    Stage tracks

    Model, prompt, detection criteria and severity bands

  • Sev-1 · a record changed outside the boundary
  • Sev-2 · a defect the agent did not flag
  • Sev-3 · imagery degrades, detection routes to review
See the complete Energy & Utilities failure-mode catalogue

Affected slices

A fleet-wide figure is not an asset's figure

Small fittings are the least-counted things on the line — hardest to resolve, hardest to label, and they fail without warning. Nestack reports the engineer-correction rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Small fittings and bolts9.7%3.6×Review
Adverse-weather imagery6.5%2.4×Review
Substation thermal scans4.3%1.6×Watch
Steel lattice structures2.4%0.9×Normal

Bar: engineer-correction-rate lift vs. steel-lattice baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A miss closes as a test, not a note

A cycle is done when the miss has become a test the next release has to survive. That suite is what the next detection off the line is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Engineer-correction rate rises in an asset slice.

  2. 02

    Diagnose

    Not the model first — read the frames and the asset records behind the corrections until one criterion is left.

  3. 03

    Improve

    Version-stamp the change and attach the detections that exposed it.

  4. 04

    Verify

    The affected cases run again, and a fail stops the release.

  5. 05

    Learn

    It becomes a standing test, and the inspection rules change with it.

  6. Learn → Detect

    The return edge. The next detection runs 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, detection workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Inspection workflow discovery and automation-boundary definition.Week 1
  2. 02Inspection and asset source assessment.Week 1, Week 2
  3. 03Detection criteria, severity band and asset-class mapping.Week 1, Week 2
  4. 04Imagery ingestion and normalisation.Week 2
  5. 05Detection logic and frame binding.Week 2, Week 3
  6. 06Confidence scoring and coverage routing.Week 3
  7. 07Engineer disposition workflow.Week 3, Week 4
  8. 08Asset- and work-management integration.Week 5
  9. 09Missed-defect regression cases.Week 4, Week 5
  10. 10Guardrails and disposition controls.Week 4
  11. 11Asset-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation 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 tierPilotOne circuit, one asset classProductionProduction asset systemsAdvancedMultiple regions / programmes
Introduced at Pilot
Detection to your record and criteria✓✓✓
Engineer disposition✓✓✓
Detection-quality baseline✓✓✓
Introduced at Production
Reporting by asset class—✓✓
Disposition workflow in your systems—✓✓
Approved write-back—✓✓
Asset-management integration—✓✓
Introduced at Advanced
Multi-region criteria and severity rules——✓
Multi-stage engineering approvals——✓
High inspection volume——✓
Multi-region programme controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls 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.

  1. 01Your asset classes and inspection record structure

    Imagery ingestion and asset mapping

    Week 1

  2. 02Representative inspected assets

    Detection baseline, criteria extraction and frame binding

    Week 2

  3. 03Your detection criteria and severity bands

    Detection criteria, severity band and asset-class mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Inspection and asset assessment, then integration setup

    Week 2

  5. 05Detections you would not want acted on

    Coverage cases and the evaluation suite

    Week 4

  6. 06What must reach an engineer before work is scheduled

    Confidence scoring, coverage routing, guardrails and disposition controls

    Week 3

  7. 07Named engineers to disposition detections

    Engineer disposition workflow, then pilot and production validation

    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, which is why week 5 doubles up rather than padding.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Inspection workflow discovery, criteria mapping and the automation boundary

W2Imagery and asset integration and the detection baseline

W3Detection workflow, confidence logic and disposition controls

W4Evaluation suite, coverage checks and failure-mode testing

W5Asset-system integration, pilot circuits and targeted corrections

W6One inspection cycle read under the asset team, then handover

Reading the band
Each bar sits on the weeks its own work runs, and no others. Week 5 genuinely carries two kinds of work.
At the end of W6
The last checks clear on live inspections and monitoring moves to Agent Care.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Energy AI agent

Build an asset-maintenance agent that stops at the disposition.

Show us your inspection passes and the criteria your engineers already apply. You bring the named engineer who dispositions a finding and the programme it feeds; we set the automation boundary at that person.

Typical delivery
4–6 weeks
Deployment
Your asset stack
Named engineer
Pre-disposition
After launch
Agent Care available

Nestack Agents · Grid asset maintenance

AGT-EN-02 · Agent Care available after launch

Build an asset-maintenance agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your assets

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.