Electronics AI agent · Export classification
Export Classification & Licence AI Agent
Record why a part is classified the way it is — the parameters read, the technical note relied on, the date and the person — and leave the determination to a named classifier.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- A named person
- Rationale first
- After launch
- Agent Care
What this agent does
Holds the rationale, never the call
In
- 01
A part is classified by a named person, and the file keeps the parameters that produced the answer.
- 02
A self-classification carries no prescribed signature, so the reasoning is all a later reader can test.
Reason
- 03
A deadline can carry no signatory at all — the 1 February encryption report and the quarterly 743.9 one.
- 04
A record meets 762.5 only with an audit trail, written procedures and retrieval by party or country.
- 05
A transaction ends, and 762.6 runs five years from the latest of export, reexport or diversion.
Decide
- 06
A rule is unenforced and still codified — 740.27, 740.28, 740.29 and ECCN 4E091 were never removed.
- 07
A legacy procedure asks for an encryption registration; Supplement No. 5 to Part 742 is [Reserved].
Out
- 08
A file is kept whole — the parameters, the screens run, the drafts and the person who reasoned it.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the parameters and keeps the file; a named person classifies the part, decides whether a licence is required, signs and files, and the exporter answers for it.
Example workflow
One part, parameters to signature
- 1Part reaches the queueA new part number, a design change, a new destination, or a release to a foreign person on site
- 2Parameters assembledDatasheet values, the technical note relied on, the buyer, the destination and the source behind each
- 3Rationale draftedThe parameters read, the classification proposed, and what would change the answer
- 4Controls appliedList-currency checks, parameter-source checks, licence-clock checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is classified at any of them — the agent is assembling, and the trade compliance lane opens at the completeness gate.
Goes to the trade compliance lead to classify.
Adds an export counsel read first.
Trade compliance review
The file is held with its parameters, its screens and its gaps.
Record the classification · Append evidence · Send to counsel
Classified — by a named person
- 6Item master and order records updatedOnly where write access and export policy allow it
- 7Outcome evaluatedScreen currency, parameter completeness, counsel corrections and what a later review found
Corrections
Each counsel correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Determining the ECCN, or that a part is EAR99.
- Deciding that no licence is required.
- Signing Block 25 or a Form BIS-711 statement.
- Filing in SNAP-R, in AES or with any authority.
Automation boundary
Agent acts unaided
- ✓Assemble the parameters and name the source of each.
- ✓Re-run the end-user screen against every list currently on record.
- ✓Draft the rationale, the note relied on and the date.
- ✓Diarise the 750.7(g) licence validity against the employee tenure.
Nothing is classified or filed except by a named person, inside the agreed boundaries.
- Judging that a listed party has left a screen.
- Telling a customer that no authorisation is needed.
- Deciding that a release on site is object code, not source.
- Changes to screening rules, clocks or licence packages.
Example output
One part, annotated
Five years after the export this file is the reasoning, and no signature was ever prescribed for it.
Classification file · single part
Illustrative example- Part
- Encrypting network module
- Written up as
- Proposed 5A992, the 740.17(b)(1) route
- Parameters
- Read 19 August 2026
- Note of record
- Producer note, revision on file
- Confidence
- Held unclassified
- Held for
- The trade compliance lead, in person
As received
Taken from the datasheet revision on file and the classification the producer publishes.
What the file holds
- The parameters read
- The producer note
- The end-user screen
Why no classification here
Strict liability sits on the exporter, and 748.3 makes a CCATS no ruling on scope.
What the score decides
Below the configured threshold the file picks up a counsel read before.
Action
Record the classificationAppend evidenceSend to counselValue
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
Screen the party, not just the part
Shipment screening is our transportation agent, HTS origin our manufacturing one, and the FCC covered list a separate page.
The rationale is the product
Nothing is signed on a self-classification, yet Block 25 must be signed by hand, and a BIS-711 responsible official cannot delegate.
Where the exposure now reaches
The Affiliates Rule is stayed to 9 November 2026 and reimposes on 10 November 2026; the separate OFAC fifty-per-cent rule was never suspended.
Build an evidence trail
The part, the parameters that drove the classification and the person who reasoned it stay on file.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Filing systems
- SNAP-R
- AES / EEI
- Licence and CCATS records
Screening lists
- Entity List
- MEU List
- SDN
- Unverified List
Product and engineering
- PLM
- item master
- Datasheets
- technical notes
Agent
Export classification & licence
- Reads the parameters
- Builds the file
- Holds for the classifier
Order and shipping systems
- ERP
- order management
- Forwarder records
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six sieves between the model and the file
Six sieves, the widest of them first. Whatever is left at the end is written out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a file may stand
- L3
- Puts the determination in a person's hands
- L4 – L5
- Keep the classification and the reasoning behind it
- L6
- Holds the part at unclassified when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the file that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the file a reviewer reads
- E1Final-output evaluationDid the file record the parameters that produced the answer?
- E2Step-level evaluationDid the agent read the current lists, the right note and the right revision?
- E3Tool evaluationDid it read and write the correct part and the correct item record?
- E4Confidence calibrationDo low-confidence files actually attract more counsel corrections?
- E5Slice evaluationHow does performance change across specific destinations?
- E6Business outcomeHow many files needed a correction before the lead classified them?
Floor — the outcome the exporter answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each set at the stage where it actually starts.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- IG-03
Superseded list version
The screen is run against a list page already amended.
Stage gathers
The parameters, the note, the lists and the screens
02 · Reasoning
2 modes
- IG-04
Parameter taken on trust
A datasheet value is used with no source behind it.
- IG-06
Ownership not read through
A listed parent behind the buyer is never reached.
Stage proposes
The parameters, the screens run and completeness
03 · Tool / write
2 modes
- IG-02
Silence read as an answer
A gap in the file is treated as no licence needed.
- IG-05
Release on site missed
Technology goes to a foreign person untracked.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- IG-01
Classified, rationale unrecorded
The file shows an answer but not what produced it.
Stage returns
The file a named person stands behind five years on
05 · Change / Version
1 mode
- IG-07
Silent screening regression
A rule change moves a screen without moving the file.
Stage tracks
Model, prompt, screening rules and control lists
- Sev-1 · a part ships outside the boundary
- Sev-2 · a wrong classification lands on file
- Sev-3 · list degrades, file holds unclassified
Affected slices
One destination absorbs the corrections
A destination-level rationale-completeness figure can read clean while one route absorbs most of the corrections. Nestack reports the counsel-correction rate by destination, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Third-country assembly routes | 7.9% | 3.7× | Review | |
| Uncontrolled parts to listed fabs | 5.6% | 2.6× | Review | |
| Deemed exports on site | 3.5% | 1.6× | Watch | |
| Direct A:5 destinations | 2.3% | 1.1× | Normal |
Bar: counsel-correction-rate lift vs. direct A:5 baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an unreconstructable file costs
A cycle closes when the uncontrolled part to a listed party is a regression case. That suite is what the next classification written is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Counsel-correction rate rises on one destination.
- 02
Diagnose
The brush that carried no control number and went to a listed fab anyway is read back until one cause remains.
- 03
Improve
The change is numbered, and the parts that provoked it are filed underneath.
- 04
Verify
One part case still red will hold the whole release.
- 05
Learn
It becomes a permanent case, and the screening rules change with it.
Learn → Detect
The return edge. The next classification is measured 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, file assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Classification-workflow discovery and boundary work.Week 1
- 02Item-master, PLM and datasheet sources.Week 1, Week 2
- 03Control-parameter, screening-list and clock mapping.Week 1, Week 2
- 04Parameter ingestion and normalisation.Week 2
- 05Technical-note and revision binding.Week 2, Week 3
- 06Completeness scoring and counsel routing.Week 3
- 07Classifier release workflow.Week 3, Week 4
- 08Screening and ERP integration.Week 5
- 09Parameter and screening cases.Week 4, Week 5
- 10Guardrails and licence controls.Week 4
- 11Classification-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 product family, one destination | ProductionProduction screening workflow | AdvancedMultiple families / destinations |
|---|---|---|---|
| Introduced at Pilot | |||
| File assembly to your rules | ✓ | ✓ | ✓ |
| Named-classifier release | ✓ | ✓ | ✓ |
| Parameter-provenance baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by destination | — | ✓ | ✓ |
| Release workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Screening-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-list screening duties | — | — | ✓ |
| Cross-site deemed-export packs | — | — | ✓ |
| High part volume | — | — | ✓ |
| Multi-list end-user 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 part records and item master
→Parameter ingestion and part mapping
Week 102Representative parts and technical notes
→File baseline, note binding and rationale capture
Week 203Your screening lists and licence conditions
→Control-parameter, screening-list and clock mapping and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Item-master, PLM and screening assessment, then integration setup
Week 205Classifications you would not want re-derived
→Parameter cases and failure-mode testing
Week 406What no classification file may conclude
→Completeness scoring, counsel routing, guardrails and release controls
Week 307A named trade compliance lead to classify
→Classifier 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
Every bar is the true working span of one phase, so the fifth week has to hold a pair of them here.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Classification workflow discovery, parameter mapping and the automation boundary
W2Source integration and the parameter baseline
W3File assembly, screening logic and release controls
W4Evaluation suite, parameter cases and failure-mode testing
W5Screening integration, pilot parts and targeted corrections
W6One export quarter run under the empowered trade compliance lead, then Agent Care handover
- Reading the band
- A bar covers only the weeks its own work is named in. The doubled fifth week is real work, not padding.
- At the end of W6
- Once the classification record validates, Agent Care takes over the watch.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Electronics AI agent
Build an export classification agent around a file that nobody signs.
Show us how a part reaches classification today and who reasons it on 1 September. Who signed your last self-classification? Nobody did, because the EAR prescribes no signature on one, which is why the dated reasoning is the only part of it a reader can test.
- Typical delivery
- 4–6 weeks
- Deployment
- Your item master
- A named person
- Rationale first
- After launch
- Agent Care available
Nestack Agents · export classification & licence
AGT-EL-16 · Agent Care available after launch