Agriculture AI agent · Feedstock CI & 45Z
Feedstock Carbon-Intensity & 45Z Documentation AI Agent
Assemble the field practice records, mass-balance documentation and CI inputs your fuel producer needs, mark what is still missing, and hold every load for your grain originator to release.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Originator-led
- Pre-attestation
- After launch
- Agent Care
What this agent does
Documents the field, not the tax position
In
- 01
A pass is logged, and the agent pulls the field, the product and the rate behind it from the farm record.
- 02
A nutrient budget is filed, and it is read against subpart G of USDA's final rule of 29 June 2026.
Reason
- 03
A soil test is loaded, and its date is checked against the two-year window the report must be current for.
- 04
A tillage pass appears in year three, and the commitment behind an already-delivered report is reopened.
- 05
A load is delivered, and scale tickets are reconciled to the dry-weight basis the mass balance runs on.
Decide
- 06
A pool is built at the elevator, and each farm's attestation is listed as present or still open.
- 07
A load of non-North American origin arrives, and it is marked out of the pool for post-2025 production.
Out
- 08
A report is assembled, and it is stamped with the rule version and GREET release it was built against.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the documentation; a named producer attests, and the fuel producer registered under section 4101 claims the credit.
Example workflow
One load, field to report
- 1Delivery recordedFarm record, agronomy platform, scale ticket or practice log
- 2Practice assembledNutrient budget, tillage passes, cover crop and soil tests, each with its source
- 3Report draftedPractice records, mass balance, open gaps and confidence
- 4Controls appliedNutrient-management checks, soil-test currency, dry-weight reconciliation and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is attested at any of them — the agent is assembling, and the originator's lane opens at the confidence gate.
Goes to the grain originator to release.
Adds an agronomy read first.
Originator release
The report is held with its practice records, its open gaps and the confidence.
Release · Amend · Send to agronomy review
Released — sent to the refiner
- 6Farm records updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedGap closure, verifier queries, rejected fields and post-delivery corrections
Amendments
Every amendment made at release is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Signing the farm producer attestation.
- Determining or petitioning for an emissions rate.
- Making any representation on a producer's tax return.
- Attesting at aggregation under penalty of perjury.
Automation boundary
Agent acts unaided
- ✓Assemble the field practice records behind a delivered load for the named owner.
- ✓Carry each record forward with the source it came from.
- ✓Apply the record formats and currency windows in force.
- ✓Mark the gaps the originator should weigh, and hold the report.
Any write happens inside the boundaries agreed at implementation, never ahead of release.
- Committing a field to no-till in future years.
- Issuing an LCFS verification statement.
- Deciding which programme an acre is enrolled in.
- Changes to practice, format or release rules.
Example output
One delivered load, annotated
Everything the agent assembles is attached to the record it was drawn from.
Feedstock report output · single load
Illustrative example- Load
- Corn, farm delivery
- Practice recorded
- No-till on the delivering field, with a nutrient budget on file
- Dry weight
- 58,400 lb dry
- Source of record
- Farm record on file
- Confidence
- 89%
- Attestation status
- Producer attestation signed
As received
Taken from the farm record and the scale ticket on file — nothing on this side is asserted by the agent.
Records gathered
- Nutrient budget on file
- Soil test within 2 years
- Scale ticket, dry basis
Why this record
It records what the field did; 45Z is still a proposed rule, so nothing here is priced yet.
What the score decides
Below the configured threshold the report picks up an agronomy read first.
Action
ReleaseAmendSend to agronomy reviewValue
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 delivered load
From the farm record
Assemble from the record
Draw on the field records on file and the formats USDA's final rule of 29 June 2026 sets.
You document, they claim
Routine loads arrive with their practice records already gathered.
Send review to the thin fields
Open gaps and low-confidence reports are marked, so the originator's read starts where the record is thin.
Build an evidence trail
The field, the practice recorded on it and the producer who attested stay on the load.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Farm and agronomy records
- John Deere Operations Center
- Climate FieldView
- Granular
Grain and scale systems
- Bushel
- AgriDigital
- Cultura
- Agris
- Greenstone
Sustainability platforms
- Truterra
- Regrow
- Continuum Ag
- Verity Tracking
Agent
Feedstock CI documentation
- Reads the field
- Assembles the report
- Holds for release
Models and registries
- 45ZCF-GREET
- FD-CIC
- CARB LCFS Reporting Tool
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 report
Six layers deep, each inside the one before. What still gets past is listed in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a report may stand
- L3
- Puts the release in the originator's hands
- L4 – L5
- Keep the field and the practice behind it
- L6
- Drops to record collection when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the finished report.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the report the refiner receives
- E1Final-output evaluationDid every practice record trace back to a field record?
- E2Step-level evaluationDid the agent use the right rule version, GREET model and currency windows?
- E3Tool evaluationDid it read and write the correct field and the correct season?
- E4Confidence calibrationDo low-confidence reports actually attract more amendments?
- E5Slice evaluationHow does performance change across specific field types?
- E6Business outcomeHow many reports needed an amendment, or a correction after the load had gone?
Floor — the documentation the originator 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 →
01 · Retrieval
1 mode
- ZF-03
Superseded practice record
Practice read to an interim-rule field definition.
Stage gathers
Field records, tickets, tests and rule versions
02 · Reasoning
2 modes
- ZF-04
Nutrient budget assumed
A field passes on tillage with no budget on file.
- ZF-06
Wet-weight mass balance
Scale weight used where dry weight is required.
Stage proposes
Practice records, mass balance and confidence
03 · Tool / write
2 modes
- ZF-02
Estimate read as a score
A modelled input is passed on as a CI score.
- ZF-05
Stale soil test carried
A test outside the two-year window moves on.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- ZF-01
Attestation gap at handoff
The first aggregation point never signed its attestation.
Stage returns
The report the originator releases to the refiner
05 · Change / Version
1 mode
- ZF-07
Silent version regression
A rule or model change alters what the report accepts.
Stage tracks
Model, prompt, rule versions and GREET release
- Sev-1 · report released outside the boundary
- Sev-2 · a wrong record reaches the refiner
- Sev-3 · source degrades, report goes to review
Affected slices
Whole-farm completeness can hide one field
A farm-wide record-completeness score can look acceptable while a small number of field cohorts carry most of the gaps. Nestack reports the open-gap rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Multi-farm aggregation pools | 7.9% | 3.7× | Review | |
| Rented ground, split operators | 6.3% | 2.9× | Review | |
| Fields under a second programme | 3.9% | 1.8× | Watch | |
| Single-operator owned fields | 1.1% | 0.5× | Normal |
Bar: open-gap-rate lift vs. single-operator-field baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Nothing closes until a test exists
A cycle closes when the unsupported practice claim is a case the next release must pass. That is what the next load is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Open-gap rate rises in a field slice.
- 02
Diagnose
The truckload that arrived without the field behind it is traced back, load by load, until one cause survives.
- 03
Improve
The fix gets a number, and the loads that raised it are held against that number.
- 04
Verify
No release while a touched practice case is outstanding; it reruns until green.
- 05
Learn
The case stays for good, and the model version it was found against is recorded.
Learn → Detect
The return edge. The next load is documented 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, report assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Delivery-load discovery and documentation boundary work.Week 1
- 02Farm-record and grain-system assessment.Week 1, Week 2
- 03Practice-record, format and currency-window mapping.Week 1, Week 2
- 04Field-record ingestion and normalisation.Week 2
- 05Report assembly and record binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Originator release workflow.Week 3, Week 4
- 08Farm and grain-system integration.Week 5
- 09Practice and mass-balance cases.Week 4, Week 5
- 10Guardrails and documentation controls.Week 4
- 11Load-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne farm, one elevator | ProductionProduction farm records | AdvancedMultiple elevators / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your records and rules | ✓ | ✓ | ✓ |
| Originator release | ✓ | ✓ | ✓ |
| Record-completeness baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by field | — | ✓ | ✓ |
| Release workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Farm-record integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-programme rule sets | — | — | ✓ |
| Multi-stage originator approvals | — | — | ✓ |
| High delivery volume | — | — | ✓ |
| Multi-programme documentation controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom 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.
You bring
We build with it
01Your field records and delivery tickets
→Field-record ingestion and practice mapping
Week 102Representative delivered loads
→Assembly baseline, record indexing and source binding
Week 203Your programme rules and the refiner's report format
→Practice, format and currency mapping
Week 104Access to relevant APIs, feeds or exports
→Farm-record and grain-system assessment, then integration setup
Week 205Loads you would not want audited
→Verification cases and failure-mode testing
Week 406What no attestation may assert
→Confidence scoring, gap routing, guardrails and release controls
Week 307A named grain originator to release reports
→Originator release 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
The weeks are drawn from the harvest, so the fifth carries evaluation and the pilot together.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Load discovery, rule mapping and the automation boundary
W2Farm-record integration and the assembly baseline
W3Report assembly, confidence logic and release controls
W4Evaluation suite, currency checks and failure-mode testing
W5Grain-system integration, pilot loads and targeted corrections
W6One harvest delivered under the grain originator, then Agent Care handover
- Reading the band
- A band sits only on the weeks its work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- Validation closes on live loads and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Agriculture AI agent
Build a CI documentation agent around the report your refiner needs.
Show us your field records and who releases to the refiner. Your agronomist signs that the budget fits the field — and DOE has not yet wired the USDA practices into 45ZCF-GREET.
- Typical delivery
- 4–6 weeks
- Deployment
- Your farm records
- Originator release
- Pre-attestation
- After launch
- Agent Care available
Nestack Agents · Feedstock CI & 45Z documentation
AGT-AGR-19 · Agent Care available after launch