Agriculture AI agent · Commodity marketing
Commodity-Marketing Copilot (Position & Scenarios)
Show priced and unpriced bushels against production, contracts and delivery obligations, and do the arithmetic on the scenarios asked for — the decision stays with the grower and their broker or adviser.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Trade decision
- Grower & broker
- After launch
- Agent Care
What this agent does
Reports the position, does not advise
In
- 01
Take the books — production, priced and unpriced bushels, contracts, deliveries, storage and the cash-flow calendar.
- 02
Pull futures settlements, the elevator bids you already work from, and basis for the location and delivery period.
Reason
- 03
Net the position — produced, committed, delivered and unpriced — each bushel counted once against its own contract.
- 04
Carry storage, interest and the delivery window into the comparison so a stored bushel is not read as a free one.
- 05
Read the written marketing plan, the insurance elections and the loan terms as constraints the position sits inside.
Decide
- 06
Separate what the books settle from what they do not, and mark each figure as recorded, estimated or calculated.
- 07
Route anything that would need a price view, a hedge or a recommendation to the grower's broker or licensed adviser.
Out
- 08
Present the position and the scenario arithmetic with every input, source and date it was built from.
- 09
Re-check what moves after the position is drawn — deliveries, contract changes, expiry and first notice — and flag it.
Product statement
The agent counts, sources and calculates. Whether to price, sell or hedge — and any advice about it — stays with the grower and their broker or adviser.
Example workflow
One position, end to end
- 1Position requestedA marketing meeting, a delivery week or a standing position update
- 2Books gatheredProduction and delivery records, contracts by type, storage, insurance elections and the loan calendar
- 3Position nettedProduced, committed, delivered and unpriced bushels reconciled so nothing is counted twice
- 4Market data attachedFutures settlements, the elevator bids in use, and basis for that location and delivery period
No human action required
Stages 1 to 4 run without a person in the loop — the books are read and the position netted before anyone is asked to look at it. A bushel that appears on two contracts stops the run where it is found.
Reaches the grower with the sources attached.
Goes to a person before the figure is used.
Grower, with their broker
Reads the netted position, the contracts behind it and the arithmetic, then decides what to do — or takes it to their adviser.
Accept · Correct the books · Ask the broker
Decided — handed back
- 6Position filedWritten to the marketing record only where access and policy allow; no order is placed and no contract is signed
- 7Outcome evaluatedPosition accuracy against the books, obligation recall, date correctness and arithmetic re-checked independently
Corrections
A figure the grower corrects 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
- Recommending a sale, a purchase or a hedge.
- Offering a price view, forecast or market opinion.
- Placing, changing or cancelling an order.
- Signing, rolling or cancelling a grower contract.
Automation boundary
Agent acts unaided
- ✓Net priced, unpriced and delivered bushels against production and contracts.
- ✓Attach futures settlements, elevator bids and basis for the location and period.
- ✓Carry storage, interest, insurance guarantees and loan dates into the position.
- ✓Do the arithmetic on the scenarios the grower asks for, and show the inputs.
Write actions run only inside the approval boundaries agreed in implementation. Placing an order is not one of them.
- Choosing which contract a delivery is applied to.
- Setting basis, or deciding a roll or delivery point.
- Changing insurance elections or marketing-plan limits.
- Changing data sources, contract rules or thresholds.
Example output
One position, annotated
Everything the agent returns is attached to the record and the contract it was counted from.
Position output · single marketing question
Illustrative example- Asked for
- Old-crop corn position
- Books read
- Production, four contracts
- Delivery window
- March, one elevator
- Position returned
- Same bushels on two contracts
- Confidence
- 93%
- Recommendation
- None — not the agent's to give
As received
The question the grower asked and the books it was built from, as those records stand today.
Evidence used
- Contract and delivery tickets
- Bid, basis and its date
- Production and storage
Why it stopped here
The same bushels sit on a forward and a basis contract, so unpriced is overstated.
What the score decides
Confidence decides how hard the grower re-checks the books, not what the market does.
Action
AcceptCorrect the booksAsk the brokerValue
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 marketing question
From the books, the contracts and the market data
Apply the grower's own plan
Use the written marketing plan, the signed contracts, the insurance elections and the loan calendar that already govern the position.
Have the position ready before the call
Bushels, contracts, basis and dates are netted while there is still time to act, not added up on a notepad during the call with the elevator.
Put the judgement calls to a person
A contract the books do not settle, a date that moved, or anything that would need a price view reaches the grower and their broker while it still matters.
Build an evidence trail
Retain the books read, the contracts counted, the bids and basis used, the dates checked, the arithmetic shown and what the grower 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 & books
- Operations Center
- FieldView
- Grain accounting
- production records
Contracts & elevators
- Cash and forward
- elevator bids
- Basis and HTA contracts
- delivery tickets
Market data
- Futures settlements
- cash bids
- Basis by location
- delivery dates
Agent
Commodity-marketing reporting
- Nets the position
- Attaches the data
- Shows the arithmetic
Finance & programme
- Loan calendar
- insurance elections
- FSA loans
- LDP 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 marketing decision
Each control wraps the one inside it. A position clears every layer before a grower reads it, and the decision to price or hedge sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the figure may stand
- L3
- Decides whether the data is current
- L4 – L5
- Keep the decision with a person and the record intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole position — not only the figure the grower reads.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the position the grower opens
- E1Final-output evaluationDid the position match the books, contract by contract?
- E2Step-level evaluationDid it read the right contract, delivery period and basis location?
- E3Tool evaluationDid it query the right entity, crop year and bid source?
- E4Recommendation-language detectionDid anything returned read as advice, a forecast or a price view?
- E5Slice evaluationHow does position accuracy change across contract types and entities?
- E6Business outcomeHow much of the position did the grower correct before using it?
Floor — the marketing decision the grower makes
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 · Retrieval
2 modes
- MK-01
Contract obligation missed
A scenario runs on bushels already committed.
- MK-02
Basis from the wrong place
A bid for another location or delivery period.
Stage gathers
The books, the contracts, the bids and the dates
02 · Netting
2 modes
- MK-03
Bushels counted twice
Production, contract and delivered bushels overlap.
- MK-04
Date misread
Expiry, first notice or delivery dates read wrong.
Stage nets
Bushels produced, committed and still unpriced
03 · Arithmetic
1 mode
- MK-05
Insurance interaction ignored
A scenario cuts across the guarantee or a claim.
Stage calculates
Only the scenarios the grower has asked for
04 · Output
1 mode
- MK-06
Reads as advice
Wording steers the grower instead of reporting.
Stage returns
The position and the wording the grower reads
05 · Change / Version
1 mode
- MK-07
Silent feed regression
A bid or contract-feed change moves the position.
Stage tracks
Model, prompt, contract-rule and feed changes
- Sev-1 · the wording steers the decision
- Sev-2 · a decision rests on a wrong figure
- Sev-3 · reporting degrades, more to check
Affected slices
Position accuracy is not the same across contract types
An error rate that reads acceptably across a whole book of bushels can sit almost entirely in the contract types with the most moving parts. 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 |
|---|---|---|---|---|
| Hedge-to-arrive and basis contracts | 7.2% | 3.8× | Review | |
| Multiple entities, split ownership | 5.3% | 2.8× | Review | |
| Grain under an FSA loan | 3.6% | 1.9× | Watch | |
| Single-elevator cash sales | 1.3% | 0.7× | Normal |
Bar: position-error rate lift vs. single-elevator baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A wrong position is corrected in the books, not in the market
A marketing decision cannot be unwound at yesterday's price. Once it is made, the work is the correction, the record and the next position.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Position errors, missed obligations and advice-like wording surface one contract type at a time.
- 02
Diagnose
The count is unpicked: the record read, the contract parsed, the bid source, the netting rule.
- 03
Improve
Your compliance lead clears the changed source, netting rule or wording constraint, version by version.
- 04
Verify
Held-out positions are netted again and compared line by line with what the books say.
- 05
Learn
The position is held as a fixed test case, and the wording constraint goes into the runbook.
Learn → Detect
The return edge. A change to what the agent may say about a position is reviewed by your compliance lead or adviser before it ships.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, books and contracts, position logic, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Books, delivery-record and contract-source assessment.Week 1, Week 2
- 03Contract-type and obligation mapping.Week 1, Week 2
- 04Position netting and double-count rules.Week 2
- 05Futures, cash-bid and basis data by location and period.Week 2, Week 3
- 06Storage, interest and cash-flow arithmetic.Week 3
- 07Insurance, programme and loan-covenant constraints.Week 3, Week 4
- 08Scenario arithmetic and the inputs it shows.Week 4
- 09Position-accuracy and recommendation-language evaluation.Week 4, Week 5
- 10Multi-entity and split-ownership handling.Week 5
- 11Farm-system and grain-accounting integration.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 crop, one entity | ProductionProduction books integration | AdvancedMulti-entity / grain and livestock |
|---|---|---|---|
| Introduced at Pilot | |||
| Position netting against the books | ✓ | ✓ | ✓ |
| Contract and delivery obligation tracking | ✓ | ✓ | ✓ |
| Futures, bid and basis data attached | ✓ | ✓ | ✓ |
| Trade decision left with the grower | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Storage, interest and cash-flow arithmetic | — | ✓ | ✓ |
| Insurance and loan-constraint flags | — | ✓ | ✓ |
| Scenario arithmetic on request | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-entity and split-ownership positions | — | — | ✓ |
| Livestock and multi-commodity positions | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on record and contract sources, farm-system integrations, contract types, entity structure, 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 production, storage and delivery records
→Books, delivery-record and contract-source assessment
Week 102Every contract you are holding, by type
→Contract-type and obligation mapping
Week 103The bid, basis and futures sources you work from
→Futures, cash-bid and basis data by location and period
Week 204Your written marketing plan and the limits in it
→The constraints the position is reported inside
Week 305Insurance elections, FSA loans and loan covenants
→Insurance, programme and loan-covenant constraints
Week 306Positions that came out wrong, including the near misses
→Evaluation suite, regression cases and failure-mode testing
Week 407Named growers and a broker or adviser to review
→Review workflow, then supervised positions 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 multi-entity work and the first supervised positions.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Workflow discovery, contract types and the automation boundary
W2Books, delivery records and the first position netted
W3Bid and basis data, storage arithmetic and plan constraints
W4Position-accuracy evaluation and recommendation-language testing
W5Farm-system integration, scenario arithmetic and supervised positions
W6Growers work from live positions, then Agent Care starts
- Reading the band
- Recommendation-language testing runs in week 4, before a position reaches a grower in week 5. Nothing informs a marketing decision until it has.
- At the end of W6
- Positions have run alongside the spreadsheet you keep today and been read by the growers who market from them, 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 a position agent around your marketing book.
Show us a season of contracts, your delivery records and how a position is put together today. We'll rebuild those positions from your own books and show you where the count, the dates or the obligations came out differently.
- Typical delivery
- 4–6 weeks
- Deployment
- Your farm and marketing stack
- Trade decision
- The grower and their broker or adviser
- After launch
- Agent Care available
Nestack Agents · Commodity-marketing reporting
AGT-AGR-04 · Agent Care available after launch