Manufacturing AI agent · Trade & tariff
Trade, Tariff & Country-of-Origin AI Agent
Propose HTS classifications, assemble the origin evidence each legal test needs and track duty exposure as rates move, and hold every part for the named trade-compliance manager who approves it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Manager approval
- Pre-entry
- After launch
- Agent Care
What this agent does
Proposes the classification, not the entry
In
- 01
A part is added to the item master, and its drawings, as-shipped weights and supplier declarations are pulled in.
- 02
A code is proposed, and the reasoning and any CBP ruling it was read against are recorded beside it.
Reason
- 03
A rate moves in the Federal Register, and the parts sitting on that 9903 subheading are flagged against it.
- 04
A part sits near the 85 percent metal test that replaced 95 percent on 8 June 2026, and it is held for a check.
- 05
A derivative is entered, and the metal and non-metal content are worked out as two lines, not one.
Decide
- 06
A steel part arrives, and melt and pour is read from the mill certificate, not from the supplier's address.
- 07
A 232 exclusion is modelled, and the Section 301 forced-labour line it opens is added, not netted out.
Out
- 08
A pack is assembled, and the evidence, the proposal and the amendments are held with it for the named manager.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent proposes the classification; the importer of record, in person or through a licensed customs broker, files it under penalty.
Example workflow
One part, item master to entry
- 1Part record receivedItem master, drawing, supplier declaration or commercial invoice
- 2Evidence assembledWeights, bills of material, mill certificates and prior rulings, each with its named source
- 3Classification proposedHTS code, 9903 subheading, origin tests and confidence
- 4Controls appliedMetal-threshold checks, dual-line rules, melt-and-pour evidence and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is filed at any of them — the agent is proposing, and the manager's lane opens at the confidence gate.
Goes to the trade-compliance manager.
Adds a customs-broker read first.
Manager approval
The proposal is held with its evidence, its open gaps and the confidence.
Approve · Amend · Send to broker review
Approved — released to the filer
- 6Trade systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAmendment rate, ruling outcomes, CBP requests and post-entry corrections
Amendments
Every manager amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Filing the entry or entry summary with CBP.
- Adopting the classification the entry will rest on.
- Transacting customs business for another party.
- Certifying US content or country of origin.
Automation boundary
Agent acts unaided
- ✓Propose an HTS code and the 9903 subheading beside it.
- ✓Carry each origin answer with the legal test it answers.
- ✓Compute metal-content and US-content worksheets from source data.
- ✓Flag the gaps the manager should weigh, and hold the pack for them.
Any write happens inside the boundaries agreed at implementation, never ahead of the filer.
- Requesting a ruling, or relying on another party's.
- Deciding whether to make a prior disclosure.
- Deciding whether to protest a liquidation.
- Changes to rate tables, thresholds or filing boundaries.
Example output
One classification, annotated
Everything the agent proposes is attached to the record it was drawn from.
Classification output · single part
Illustrative example- Part
- Aluminium bracket
- Description
- Cast aluminium bracket, machined and plated
- Metal content
- 88% by weight
- Proposed treatment
- Composed entirely of metal
- Confidence
- 91%
- Origin test
- Smelt and cast, from the mill
As received
Taken from the item master and the mill certificate — nothing on this side is inferred.
Evidence used
- Mill certificate
- As-shipped weights
- Prior CBP ruling
Why this treatment
It clears the metal test as it stands after 8 June 2026 — the line the manager weighs.
What the score decides
Below the configured threshold the proposal picks up a broker read before it.
Action
ApproveAmendSend to broker 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 part
From the item master
Propose against your rules
Draw on the weights and rulings on file, and the subheadings in force on the day.
A code is a statement to CBP
Routine parts arrive with a proposed code and the subheading in force attached.
Send review to the exposed lines
Parts near a threshold and low-confidence proposals are marked, so the manager's read starts where exposure concentrates.
Build an evidence trail
The classification, the ruling it was read against and the broker who filed it stay on the entry.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
ERP and item master
- SAP
- Oracle
- Dynamics
- Infor
- QAD
Customs filing
- ABI / ACE broker systems
- Descartes
- e2open
Trade content
- HTSUS
- CROSS rulings
- Federal Register
- CSMS
Agent
Trade, tariff and origin
- Reads the part
- Proposes the code
- Holds for approval
Supplier documents
- Mill certificates
- invoices
- Origin declarations
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 entry
Six controls, one within the next. What the whole stack lets through is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a code may stand
- L3
- Puts the release in a manager's hands
- L4 – L5
- Keep the classification and the ruling behind it
- L6
- Falls back to duty reporting when signals degrade
How Nestack evaluates it
Evaluate the classification workflow — not only the code at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the code on the entry
- E1Final-output evaluationDid the proposed heading match the one the manager adopted?
- E2Step-level evaluationDid the agent use the right weights, rules and any company-specific rate?
- E3Tool evaluationDid it read and write the correct part and the correct field?
- E4Confidence calibrationDo low-confidence proposals actually attract more amendments?
- E5Slice evaluationHow does performance change across specific part groups?
- E6Business outcomeHow many entries still carried an IEEPA line after 24 February 2026?
Floor — the exposure the importer 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
- TD-03
Superseded rate table
A 9903 line read after the rate it carried ceased.
Stage gathers
Item master, weights, certificates and rate tables
02 · Reasoning
2 modes
- TD-04
One origin, three tests
One origin field answers three different legal tests.
- TD-06
Exclusion read as saving
Exclusion netted without the 301 line it opens.
Stage proposes
HTS code, 9903 subheading, origin and confidence
03 · Tool / write
2 modes
- TD-02
Low-confidence adoption
Proposal adopted despite insufficient certainty.
- TD-05
Weight from nominal BOM
Metal weight taken from a BOM, not the shipment.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- TD-01
Content line uncapped
US content declared above the cap the proclamation sets.
Stage returns
The classification the manager adopts
05 · Change / Version
1 mode
- TD-07
Silent rule regression
A rate or threshold change widens what the agent proposes.
Stage tracks
Model, prompt, rate tables and threshold rules
- Sev-1 · acts outside the filing boundary
- Sev-2 · a wrong code reaches the entry
- Sev-3 · source degrades, part routes to review
Affected slices
One tariff line can carry most of the risk
A blended duty rate across the catalogue can look settled while a small number of tariff lines carry most of the exposure. Nestack reports the amendment 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 |
|---|---|---|---|---|
| USMCA dual-line derivatives | 9.2% | 3.8× | Review | |
| Metal content near the test | 7.4% | 3.1× | Review | |
| Multi-country supply chains | 4.6% | 1.9× | Watch | |
| Single-origin stock parts | 1.2% | 0.5× | Normal |
Bar: amendment-rate lift vs. single-origin-part baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Every cycle ends by adding a case
A cycle closes when the misclassified part is a case the next release has to catch. That suite is what the next entry prepared is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Amendment rate rises in a tariff-line slice.
- 02
Diagnose
The line that cleared for three years under the wrong code is read back to a single cause.
- 03
Improve
The fix carries a version number, and the entries that forced it travel with it.
- 04
Verify
Nothing goes out until every touched classification case has passed a second time.
- 05
Learn
It is kept for good, and the classification rules are amended in the same commit.
Learn → Detect
The return edge. The next entry is prepared 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, classification workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Trade workflow discovery and boundary set and boundary definition.Week 1
- 02Item-master and broker-system assessment.Week 1, Week 2
- 03HTS, 9903 subheading and origin-rule mapping.Week 1, Week 2
- 04Part-data ingestion and normalisation.Week 2
- 05Classification logic and evidence binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Trade-compliance manager approval.Week 3, Week 4
- 08Entry and broker-system integration.Week 5
- 09Classification and origin cases.Week 4, Week 5
- 10Guardrails and filing controls.Week 4
- 11Entry-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 part family, one origin | ProductionProduction entry systems | AdvancedMultiple plants / programmes |
|---|---|---|---|
| Introduced at Pilot | |||
| Proposals to your parts and rules | ✓ | ✓ | ✓ |
| Manager approval | ✓ | ✓ | ✓ |
| Classification-accuracy baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by tariff line | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Item-master integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multiple origins and regimes | — | — | ✓ |
| Multi-stage trade approvals | — | — | ✓ |
| High entry volume | — | — | ✓ |
| Multi-programme duty 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 item master and part weights
→Part-data ingestion and evidence mapping
Week 102Representative entries already filed
→Classification baseline and evidence binding
Week 203Your rate tables and origin rules
→HTS, 9903 and origin-rule mapping
Week 104Access to relevant APIs, feeds or exports
→Item-master and broker-system assessment, then integration setup
Week 205Entries you would not want examined
→Origin cases and failure-mode testing
Week 406What no declaration may rest on
→Confidence scoring, gap routing, guardrails and filing controls
Week 307A named trade-compliance manager to approve
→Manager approval 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 bands cover the weeks each phase really eats, which is why the fifth carries two of them.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Trade workflow discovery, rule mapping and the automation boundary
W2Item-master integration and the classification baseline
W3Classification workflow, confidence logic and approval controls
W4Evaluation suite, origin and threshold checks and failure-mode testing
W5Broker-system integration, pilot entries and targeted corrections
W6One entry cycle prepared under the trade-compliance manager, then Agent Care handover
- Reading the band
- Each band covers only the weeks its work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- Validation closes on live entries, and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Manufacturing AI agent
Build a tariff and origin agent around your entry process.
Show us your parts and who approves a code. If one origin field serves your marking rule, your melt-and-pour report and your non-preferential test, then two of the three are already wrong.
- Typical delivery
- 4–6 weeks
- Deployment
- Your ERP and broker
- Manager approval
- Pre-entry
- After launch
- Agent Care available
Nestack Agents · Trade, tariff & origin
AGT-MFG-13 · Agent Care available after launch