Manufacturing AI agent · Energy
Energy-Optimization & Utilities AI Agent
Meter and disaggregate plant consumption by asset, find idle load, leaks and peak exposure, and propose reductions with the evidence and the process risk named — an engineer decides, and the agent moves no setpoint.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Setpoint change
- Engineer only
- After launch
- Agent Care
What this agent does
Measures and proposes, moves no setpoint
In
- 01
Read interval meters, sub-meters, utility bills and the tags your BMS, SCADA and MES already record.
- 02
Take production output, shift calendars, weather and asset run states over the same interval.
Reason
- 03
Disaggregate consumption to asset, utility and process, and reconcile the parts against the incoming meter.
- 04
Separate load that tracks production from load that runs regardless — idle, off-shift, standby and leak.
- 05
Model consumption against production volume and weather, so a change is read against an adjusted baseline.
Decide
- 06
Name the process parameter a proposal would touch, and route it to change control rather than offer it as a tweak.
- 07
Hold anything that alters pressure, temperature, flow or air change for the engineer who owns that process.
Out
- 08
Present each reduction with its evidence, its estimate, its process risk and what would have to be re-measured.
- 09
Retain the meter data, the baseline model, the adjustments applied and every proposal an engineer turned down.
Product statement
The agent measures, disaggregates and proposes. Changing a setpoint, a schedule or a process parameter stays with the engineer who owns it, under your change control.
Example workflow
One proposal, end to end
- 1Interval data readMeters, sub-meters, BMS and SCADA tags, utility bills and the tariff those bills are struck on
- 2Load disaggregatedConsumption split to asset, utility and process, then reconciled against the incoming meter
- 3Baseline modelledConsumption fitted against production volume, weather and shift pattern over the baseline period
- 4Candidate foundIdle or off-shift load, leak-rate drift, a scheduling gap, or a peak the tariff makes expensive
No human action required
Stages 1 to 4 run without a person in the loop — reading, disaggregating, baselining and finding candidates all finish before an engineer is asked to read anything. A candidate that moves a process parameter ends that stretch on the spot.
Goes to the engineer as a proposal.
Held for change control and the process owner.
Energy or process engineer
Reads the disaggregation, the estimate against the adjusted baseline and the process risk, then decides whether the process can take the change.
Accept · Reject · Send to change control
Accepted — handed back
- 6Proposal put to an engineerThe estimate against the adjusted baseline, the evidence behind it and the process risk named
- 7Saving re-measuredAt 30 and 90 days against the adjusted baseline, alongside what was accepted and what was rejected
Rejections
Why an engineer turned a proposal down is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Changing any setpoint on plant or utility equipment.
- Altering compressed-air pressure, dryer or receiver settings.
- Changing HVAC air change, temperature or humidity limits.
- Adjusting a furnace, oven or curing profile.
Automation boundary
Agent acts unaided
- ✓Meter, sub-meter and disaggregate consumption by asset and process.
- ✓Model an adjusted baseline against production volume, weather and shift.
- ✓Find idle load, off-shift load, leak drift and peak exposure.
- ✓Propose a reduction with its evidence, its estimate and its process risk.
The agent writes proposals and measurements. Moving a setpoint and declaring a saving are not its to write.
- Changing chilled-water, steam or process-water conditions.
- Switching off, idling or restarting any asset.
- Shedding load or acting on a peak event.
- Signing off a saving as achieved.
Example output
One compressed-air proposal, annotated
Everything the agent proposes is attached to the meter data, the baseline and the process risk it was built from.
Energy output · one proposal on one utility
Illustrative example- Asset
- Compressed-air header
- Off-shift draw
- 31% of run load
- Meter mapping
- Walked down
- Proposed
- Isolate zone 4 off-shift
- Confidence
- 82%
- Saving
- Estimated, not yet verified
As received
Sub-meter intervals and the run states SCADA records. The mapping to this header was confirmed on a walk-down.
Evidence used
- Off-shift draw, 12 weeks
- Zone 4 valved already
- No trace heating on it
Why this proposal
Zone 4 runs nothing off-shift and feeds no trace heating. Header pressure is untouched.
What the score decides
Confidence is in the mapping, not the saving — measured after the change, never before.
Action
AcceptRejectSend to change controlValue
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 metered interval
Sub-meters, BMS and SCADA tags, utility bills
Read the plant against its output
Split consumption to asset, utility and process, and measure it against production volume, shift pattern and weather rather than against last month.
Give the engineer the workings
Each proposal arrives with the load it came from, the baseline it was measured against and the parameter it would touch — so the argument is about the process, not the number.
Put the process risk in front of a person
A pressure, a profile or an air-change proposal reaches the engineer who owns that process, with what it would change, before anyone touches a keypad.
Build an evidence trail
Retain the meter data, the baseline model, the adjustments applied, the estimate, the engineer's decision and what the saving did at 30 and 90 days.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Metering & billing
- Interval meters
- sub-meters
- Utility bills
- tariff and demand data
Utility & building controls
- BMS / BAS
- compressed-air controllers
- Chiller and boiler plant
- HVAC schedules
Process & production data
- MES output
- SCADA historians
- Shift calendars
- asset run states
Agent
Energy optimization & utilities
- Disaggregates load
- Adjusts the baseline
- Proposes with risk
Site context
- Weather data
- metering schematics
- Asset registers
- change-control 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 a setpoint
Each control wraps the one inside it. A proposal clears every layer before an engineer sees it, and the change itself sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether a saving is real at all
- L3
- Decides what counts as a process change
- L4 – L5
- Keep the change with an engineer, trail intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole measurement — not only the saving it estimates.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the proposal the engineer opens
- E1Final-output evaluationDid the saving hold at 30 and 90 days against an adjusted baseline?
- E2Step-level evaluationWas the baseline adjusted for production volume and for weather?
- E3Tool evaluationDid it read the right meter, tag and interval for that asset?
- E4Metering-to-asset accuracyDoes the sub-meter measure the asset the agent says it does?
- E5Slice evaluationHow does it hold across asset types, seasons and shared utilities?
- E6Business outcomeWhich proposals were accepted, and why were the rest rejected?
Floor — the saving still there in ninety days
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 setpoint, sheds a load or signs off a saving — the engineer's decision and your own change control are the controls that stop them. If one gets through and a change is already running, the meter data, the baseline model and the adjustments held in the record are what the reversal and the re-measurement are built from.
Agent lifecycle
Direction of processing →
01 · Metering
1 mode
- EO-01
Sub-meter on the wrong asset
A tag that has not matched the panel for years.
Stage gathers
Meter, sub-meter and tag data mapped to assets
02 · Disaggregation
1 mode
- EO-02
Idle load that is deliberate
Trace heating, a sump pump, a standby that protects an asset.
Stage splits
Consumption to asset, utility and process
03 · Baseline
2 modes
- EO-03
A saving production volume made
Output fell that quarter; the measure did nothing.
- EO-04
Weather did it, not the measure
A mild quarter reads as a chiller improvement.
Stage adjusts
Consumption against production and weather
04 · Proposal
2 modes
- EO-05
A setpoint that is a process change
Air pressure, an oven profile, an air-change rate.
- EO-06
Peak action against a commitment
A shed hour is the hour an order had to run.
Stage proposes
The reduction and the risk an engineer reads
05 · Verify / Change
1 mode
- EO-07
A saving that costs equipment life
Short cycling, low flow, a motor off its map.
Stage tracks
What the saving did, and what changed since
- Sev-1 · a process or a commitment is at risk
- Sev-2 · a wrong number reaches the decision
- Sev-3 · the plant pays for it somewhere else
Affected slices
A shared header is nobody's meter
Two proposals of the same size are not the same measurement. A dedicated meter says what one machine drew; a header feeding four lines and a paint shop 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 |
|---|---|---|---|---|
| Shared utilities, no sub-meter | 6.8% | 3.6× | Review | |
| Seasonal heating and cooling | 5.3% | 2.8× | Review | |
| Compressed-air proposals | 3.4% | 1.8× | Watch | |
| Single asset, dedicated meter | 1.5% | 0.8× | Normal |
Bar: unheld-saving-rate lift vs. dedicated-meter baseline · scale 0–4.0× · tick marks 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
A saving that did not hold is the useful one
What an engineer rejected, and what the meter said ninety days later, come back as baseline corrections and limits on what may be proposed.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
A saving stops holding at 30 or 90 days, or one class of proposal keeps being rejected.
- 02
Diagnose
Traced to a mapping, an adjustment the baseline missed, a non-routine change on site, or a process the proposal did not know about.
- 03
Improve
The mapping, the baseline model or the proposal rule is corrected — and any process parameter goes back through your change control with a named approver.
- 04
Verify
Re-measured against the adjusted baseline over comparable volumes and seasons, not against the month before the change.
- 05
Learn
The deliberate load, the unmodelled process or the tariff term becomes a check the next proposal has to clear.
Learn → Detect
The return edge. Re-basing for a new line or a tariff change is itself a change to the record — it is documented before the next saving is measured.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, metering and access, disaggregation and baselines, evaluation, then engineer review and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Energy-workflow discovery and boundaries.Week 1
- 02Metering plan and meter-to-asset walk-down.Week 1, Week 2
- 03Meter, BMS, SCADA and tariff data access.Week 1, Week 2
- 04Interval ingestion and meter reconciliation.Week 2
- 05Load disaggregation by asset and process.Week 2, Week 3
- 06Baseline model — production, weather, shift.Week 2, Week 3
- 07Idle, off-shift and leak-drift detection.Week 3, Week 4
- 08Peak-demand and tariff exposure analysis.Week 3, Week 4
- 09Process-parameter flags, set by your engineers.Week 4
- 10Evaluation suite and replayed past measures.Week 4, Week 5
- 11Engineer review and change-control handoff.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 utility, one area | ProductionPlant-wide metering and baselines | AdvancedMulti-site / multi-utility |
|---|---|---|---|
| Introduced at Pilot | |||
| Load disaggregation for one utility | ✓ | ✓ | ✓ |
| Meter-to-asset mapping and walk-down record | ✓ | ✓ | ✓ |
| Adjusted baseline — production and weather | ✓ | ✓ | ✓ |
| Idle and off-shift load detection | ✓ | ✓ | ✓ |
| Proposals with the process risk named | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Peak-demand and tariff exposure analysis | — | ✓ | ✓ |
| Re-measurement at 30 and 90 days | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Asset-type and seasonal slices | — | — | ✓ |
| Multi-site and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the utilities and areas in scope, the state of your sub-metering, meter, BMS and SCADA integrations, tariff complexity 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 metering schematic and what each sub-meter actually feeds
→Metering plan and meter-to-asset walk-down
Week 102Access to interval data, BMS and SCADA tags and your utility bills
→Meter, BMS, SCADA and tariff data access
Week 203Production output and shift calendars over the same period
→Baseline model — production, weather and shift
Week 304The loads that look wasteful and are not — trace heating, sumps, standby
→Load disaggregation and deliberate-load exclusions
Week 305Your tariff, demand charges and any ratchet or capacity terms
→Peak-demand and tariff exposure analysis
Week 406Which setpoints and profiles are process parameters, and who owns them
→Process-parameter flags, set by your engineers
Week 407Named engineers who accept a proposal and take it into change control
→Engineer review workflow, then the change-control handoff
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 replayed measures and the first proposals your engineers read.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Energy-workflow discovery, metering plan and the walk-down
W2Meter, BMS, SCADA and tariff access, then interval ingestion
W3Disaggregation, baseline model and deliberate-load exclusions
W4Idle, leak and peak detection, and process-parameter flags
W5Replayed past measures and the first proposals engineers read
W6Engineer review, change-control handoff, then Agent Care starts
- Reading the band
- The meter-to-asset walk-down in weeks 1 and 2 is physical work on your site. Nothing disaggregated before it is finished is worth reporting.
- At the end of W6
- The agent has proposed against your own meter data, been read by the engineers who decide, and the route into your change control is agreed — 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 an energy agent around your own meters.
Show us one utility, whatever sub-metering you have on it, and a year of bills, output and shift calendars. We'll disaggregate that load against your own production and weather, tell you which of last year's savings survive an adjusted baseline, and mark every proposal that would be a process change.
- Typical delivery
- 4–6 weeks
- Deployment
- Your plant systems
- Setpoint change
- An engineer, always
- After launch
- Agent Care available
Nestack Agents · Energy optimization & utilities
AGT-MFG-10 · Agent Care available after launch