Agriculture AI agent · Irrigation
Irrigation-Management AI Agent (Scheduling & Allocation)
Plan sets from soil moisture, weather and crop demand, test them against allocation and restrictions, run what the farm pre-authorised and reconcile the water applied against the meter.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Unattended run
- Pre-authorised
- After launch
- Agent Care
What this agent does
Plans the water, does not own it
In
- 01
Take soil-moisture readings, in-field weather, reference ET and the forecast for each field and zone.
- 02
Take crop, growth stage, soil type, system capacity, allocation headroom and any live restriction.
Reason
- 03
Work out crop water demand against what the root zone already holds, field by field and zone by zone.
- 04
Fit that demand to set times, zone order, system capacity, pump and energy limits and the tariff window.
- 05
Test the schedule against allocation headroom, permit terms, re-entry intervals and injection rules.
Decide
- 06
Separate what the pre-authorised envelope already covers from what needs a person, and say which is which.
- 07
Hold or stop a set when a limit, a restriction, a sensor or a source changes after the plan was drawn.
Out
- 08
Run the sets the farm signed for, then reconcile water applied against planned against the meter reading.
- 09
Draft the use record with the readings, the sources, what actually ran and what a person decided.
Product statement
The agent plans, runs inside a signed envelope and reconciles. The water decision and the allocation stay with a named person on the farm.
Example workflow
One week of sets, end to end
- 1Field state readSoil moisture by depth, in-field weather, reference ET, the forecast, crop stage and soil type
- 2Limits readAllocation headroom, permit terms, live restrictions, re-entry intervals and tariff windows
- 3Schedule proposedSet times, zone order and volumes, fitted to system capacity, pump limits and energy cost
- 4Envelope testedWhat the signed zones, volume cap and stop time already cover, and what they do not
No human action required
Stages 1 to 4 run without a person in the loop — reading, planning and limit-testing finish before anyone is asked to look at anything. A limit already reached stops the plan on the spot.
Runs on the approved schedule.
Waits for a person before water moves.
Irrigation manager
Reads the plan, the readings behind it, the headroom left and what the restriction says, then decides whether the water goes on.
Release · Change the plan · Hold the water
Released — handed back
- 6Sets run and recordedOnly inside the signed envelope; anything outside it waits for a person to release it
- 7Outcome evaluatedApplied against planned against the meter, headroom accuracy, sensor health and stop latency
Overrides
A set the manager holds or rewrites is recorded with the reason and counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Setting the season water plan and its allocation.
- Signing the unattended envelope, its volume cap and stop.
- Running water while a restriction or curtailment applies.
- Irrigating a block inside a live re-entry interval.
Automation boundary
Agent acts unaided
- ✓Build the schedule from soil moisture, weather, crop stage and system capacity.
- ✓Test it against allocation headroom, permit terms, restrictions and intervals.
- ✓Run pre-authorised sets inside the signed zones, volume cap and stop time.
- ✓Reconcile applied against planned against the meter and draft the use record.
Run and write actions stay inside the envelope agreed in implementation: named zones, a volume cap, a stop time.
- Injecting fertiliser or effluent into the system.
- Drawing water past the remaining allocation.
- Filing a meter reading or a water-use return.
- Changing headroom sources, envelope rules or stop thresholds.
Example output
One night's set, annotated
Everything the agent proposes is attached to the readings and the limit it was tested against.
Schedule output · single planned set
Illustrative example- Field and zones
- Field 12, zones 3–5
- Root-zone moisture
- Below the refill point
- Planned set
- Tonight 22:00, off-peak
- Schedule result
- Held — headroom short
- Confidence
- 93%
- Water decision
- The farm's, not the agent's
As received
The field state as the probes and the station report it, beside the set the plan had drawn for tonight.
Evidence used
- Probe at two depths
- Permit balance to date
- Meter total, reconciled
Why it stopped here
The permit leaves less than this set would draw, and that is a legal limit.
What the score decides
Confidence decides how hard the manager checks the probe, not what the permit leaves.
Action
ReleaseChange the planHold the waterValue
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 field and zone
From the probes, the station and the forecast
Apply the limits that govern it
Use the allocation balance, the permit terms, any live restriction, the re-entry intervals and the farm's own field record.
Plan the week before the pump starts
Demand, set times, capacity, tariff windows and headroom are worked out while the schedule can still be changed, not after the water is on the ground.
Put the water decision to a person
A short allocation, a restriction that moved or a probe that has stopped agreeing with the field reaches the manager while the set can still be held.
Build an evidence trail
Retain the readings, the ET and forecast used, the headroom at the time, what actually ran, the meter reading and what the manager decided — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Farm records & fields
- Operations Center
- FieldView
- Trimble Ag
- field and crop records
Sensors & weather
- Soil-moisture probes
- in-field stations
- Reference ET
- forecast feeds
Control & pumps
- Pivot controllers
- valve control
- Pump and VFD telemetry
- flow meters
Agent
Irrigation planning & scheduling
- Reads the field
- Plans the sets
- Reconciles the meter
Water & energy accounts
- Allocation balances
- permit terms
- District orders
- tariff windows
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 pump
Each control wraps the one inside it. A set clears every layer before water moves, and the allocation sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide how much water the plan may draw
- L3
- Decides whether a set may open at all
- L4 – L5
- Bound unattended running and keep the record
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the water that reached the field — not only the schedule.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the schedule the irrigator opens
- E1Final-output evaluationDid the water applied match what the plan and the field called for?
- E2Step-level evaluationDid it read the right probes, ET figures, crop stage and headroom?
- E3Tool evaluationDid the valve, pump and meter it addressed actually respond?
- E4Limit-response latencyHow fast did a changed restriction or a short allocation stop a set?
- E5Slice evaluationHow does schedule error change across soils, zones and field types?
- E6Business outcomeHow much water went on off-plan, and what did a person have to undo?
Floor — the water that reaches the field
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycle
Direction of processing →
01 · Sensing / retrieval
2 modes
- IR-01
Probe drifted or badly placed
It reads wet while the root zone is dry.
- IR-02
Curtailment read too late
A restriction issued mid-season is not picked up.
Stage gathers
Soil moisture, weather, meter totals and restriction status
02 · Scheduling
2 modes
- IR-03
Headroom overstated
The plan draws more than the permit leaves.
- IR-04
Skipped on rain that missed
The forecast event does not arrive and the crop goes short.
Stage plans
Set times, zone order and the capacity they fit inside
03 · Limit gate
1 mode
- IR-05
Injection into a live interval
Fertigation starts while a re-entry interval still applies.
Stage holds
Headroom, restrictions and intervals before a set opens
04 · Run / control
1 mode
- IR-06
Set that ran only on paper
A valve or pump did not run but the record says it did.
Stage runs
Only inside the signed zones, volume cap and stop time
05 · Reconciliation
1 mode
- IR-07
Meter never reconciled
Reported use drifts away from the metered total.
Stage reconciles
Applied against planned against the meter total
- Sev-1 · water runs where it may not
- Sev-2 · the field or the record is wrong
- Sev-3 · the plan degrades, more to review
Affected slices
Schedule error concentrates where one probe speaks for the most ground
A schedule-error rate that reads acceptably across the farm can sit almost entirely in the fields where one probe stands in for several soils. 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 |
|---|---|---|---|---|
| Coarse sands, one probe | 5.6% | 3.3× | Review | |
| Fertigation and effluent sets | 4.4% | 2.6× | Review | |
| Tail-end zones, low pressure | 3.2% | 1.9× | Watch | |
| Uniform silt-loam pivots | 1.7% | 1.0× | Normal |
Bar: schedule-error rate lift vs. silt-loam pivot baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Water on the ground cannot be pulled back out
A set that ran wrong is closed out in the field first — hold the programme, tell whoever must be told, correct the record — then in the agent.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Field by field, applied-versus-planned gaps, headroom errors and stop latency are watched.
- 02
Diagnose
A probe, the ET figure, the headroom read, an envelope rule or the controller is named as the cause.
- 03
Improve
An envelope change is a farm decision — signed, dated and version-linked before a set runs on it.
- 04
Verify
The schedule is rebuilt for seasons already irrigated and compared against the meter.
- 05
Learn
The set becomes a fixed test, and what the envelope missed is added to the pre-run checks.
Learn → Detect
The return edge. A wider unattended envelope is a farm decision — it is signed before the next season, not inferred from a season that went well.
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, scheduling logic, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Water rights, allocation and permit-term mapping.Week 1, Week 2
- 03Field, soil, crop and system-capacity inventory.Week 1, Week 2
- 04Sensor, weather and reference-ET source assessment.Week 2
- 05Crop water demand and schedule logic.Week 2, Week 3
- 06Zone order, set times, pump and tariff constraints.Week 3
- 07Headroom, restriction and re-entry gating.Week 3, Week 4
- 08Unattended envelope, volume cap and stop rules.Week 4
- 09Sensor-health and reconciliation evaluation.Week 4, Week 5
- 10Controller, valve and pump integration.Week 5
- 11Meter reconciliation and use-record drafting.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 field group | ProductionProduction controller integration | AdvancedMulti-source / district scale |
|---|---|---|---|
| Introduced at Pilot | |||
| Schedule building from moisture and weather | ✓ | ✓ | ✓ |
| Crop water demand and growth-stage tracking | ✓ | ✓ | ✓ |
| Allocation headroom and permit-term tracking | ✓ | ✓ | ✓ |
| Water decision left with a named person | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Restriction and curtailment watch | — | ✓ | ✓ |
| Pre-authorised unattended running, capped and stopped | — | ✓ | ✓ |
| Meter reconciliation and use-record drafting | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-source allocation and permit portfolios | — | — | ✓ |
| Districts, contractors and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on water sources and permit terms, sensor and controller integrations, field and zone count, reporting requirements, 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.
You bring
We build with it
01Your water rights, allocations and permit terms
→Water rights, allocation and permit-term mapping
Week 102Field boundaries, soils, crops and system capacity
→Field, soil, crop and system-capacity inventory
Week 103Your probes and stations, and where each one actually sits
→Sensor and weather source assessment, and sensor-health checks
Week 204The envelope you are willing to sign for unattended running
→Unattended envelope, volume cap and stop rules
Week 405Seasons that went wrong — short crops, breaches, wasted sets
→Evaluation suite, regression cases and failure-mode testing
Week 406Meter access and the readings you reported last season
→Meter reconciliation and use-record drafting
Week 507A named irrigation manager and an agronomist to review
→Review workflow, then supervised sets and 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. Week 5 carries both the first supervised sets and the meter readings they are checked against.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Water rights, allocations and the automation boundary
W2Sensors, weather sources and the first field group
W3Demand, set times, zone order and pump and tariff limits
W4Limit gating, envelope rules and sensor-health testing
W5Controller integration, meter reconciliation and supervised sets
W6Sets run inside the signed envelope, then Agent Care starts
- Reading the band
- The unattended envelope is written and tested in week 4, before a set runs on its own in week 5. Nothing moves water unsupervised until it has.
- At the end of W6
- Schedules have run alongside your existing practice, been read by the manager who releases water, and reconciled against the meter, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Agriculture AI agent
Build an irrigation agent around your allocation, not around a forecast.
Show us a season of schedules, your allocation and permit terms, and the meter readings you reported against them. We'll set the caps with you — zones, volume, a stop time — and run a past season against them before anything is left unattended.
- Typical delivery
- 4–6 weeks
- Deployment
- Your irrigation and water stack
- Unattended run
- Pre-authorised, capped, stoppable
- After launch
- Agent Care available
Nestack Agents · Irrigation planning & scheduling
AGT-AGR-06 · Agent Care available after launch