Automotive AI agent · Safety recall
Safety-Recall & Field-Service-Action AI Agent
Assemble the chronology of principal events behind a defect determination, track the Part 573 and Part 577 clocks, and leave the determination itself with the engineer who has to make it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Engineer signs
- 5 working days
- After launch
- Agent Care
What this agent does
Assembles the chronology, not the determination
In
- 01
A field report lands, and it is ingested beside warranty claims, supplier 8D files and foreign field actions.
- 02
A claim is normalised to the manufacturer's date of receipt, the date 49 CFR § 573.6(c)(6) asks the file to carry.
Reason
- 03
A determination is dated, and the chronology of principal events behind it is assembled from those records.
- 04
A campaign opens, and the § 573.6(a) five-working-day filing clock runs from that decision, not from the signal.
- 05
A filing goes in, and the § 577.7 sixty-day owner clock is counted from the filing date, as the rule reads.
Decide
- 06
A software remedy ships, and it is treated as equipment under NHTSA Enforcement Guidance Bulletin 2016-02.
- 07
A quarter closes, and the package routes to the named engineer with the six § 573.7 returns tracked to the last.
Out
- 08
An event is retained with its source, its date of receipt and the engineer who dated the determination.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles and dates; a named engineer at the manufacturer determines, signs the Part 573 report and answers for it.
Example workflow
One campaign, signal to filing
- 1Field signal receivedWarranty claim, field service report, supplier 8D or foreign field action
- 2Chronology assembledPrincipal events, each with its source and the manufacturer's date of receipt
- 3Package draftedChronology, population, remedy, notice text and confidence
- 4Controls appliedCompleteness checks, date-of-receipt checks, Part 573 and Part 577 clock checks and confidence threshold
No human action required
Stages 1 to 4 run unaided and determine nothing — the agent assembles and dates, and the engineer's lane opens at the confidence gate.
Goes to the safety-office engineer.
Adds a product-investigations read first.
Engineer sign-off
The package is held with its chronology, its named gaps and the confidence.
Approve · Amend · Send to investigations
Signed — the clock starts
- 6Campaign systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedChronology completeness, amendment distance, filing timeliness and quarterly-return currency
Amendments
Every engineer amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Determining that a defect is safety-related.
- Filing the Part 573 report with the agency.
- Deciding an update is a customer-satisfaction campaign.
- Issuing a do-not-drive or park-outside advisory.
Automation boundary
Agent acts unaided
- ✓Assemble the chronology, each event carrying its date of receipt.
- ✓Track the § 573.6, § 577.7 and § 573.7 clocks from the dated determination.
- ✓Reconcile the § 573.15 VIN lookup and the § 579.5.
- ✓Flag the gaps and the missing sources, and hold.
Any write happens inside the boundaries agreed at implementation, never ahead of the signature.
- Setting the date the determination was made.
- Scoping the population a campaign covers.
- Committing to a remedy and a parts-availability date.
- Changes to reporting rules or automation thresholds.
Example output
One principal event, annotated
Everything the agent assembles is attached to the record it was drawn from.
Chronology output · single campaign
Illustrative example- Campaign
- Seat-belt anchorage
- Principal event
- Sixth warranty claim naming the same anchorage weld
- Date of receipt
- 14 Feb 2026
- Source of record
- Warranty claim file
- Confidence
- 93%
- Clock position
- No determination dated yet
As received
Taken from the warranty file and the supplier's 8D report — nothing on this side is written by the agent.
Source facts used
- Warranty claim intake
- Supplier 8D report
- Foreign field action
Why this date
It is the date the manufacturer received the claim, not the date the claim was opened.
What the score decides
Below the configured threshold the package picks up a product-investigations read before it.
Action
ApproveAmendSend to investigationsValue
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 signal
From the field record
Assemble from the record
Draw on warranty, field and supplier sources, and on the April 2026 agency orders that start a clock inside you.
The clock starts at the decision
Routine chronology assembly, notice drafting and quarterly returns arrive already prepared.
Send review to the dated gaps
Gaps, a stale § 573.15 lookup and low-confidence packages are marked — the lookup was itself a penalised item in November 2024.
Build an evidence trail
The event, the record it was drawn from and the engineer who determined stay on the chronology.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Quality and warranty
- Warranty claims
- TREAD
- Supplier 8D
- field reports
Regulatory filing
- NHTSA Part 573 portal
- Part 577
- quarterly returns
Owner and VIN data
- Registered-owner records
- VIN lookup
- § 573.15 feed
Agent
Safety recall & field service
- Reads the record
- Builds the chronology
- Holds for signature
Service and parts
- DMS
- dealer service
- Parts availability
- OTA
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six controls between the model and the filing
Six controls, each nested in the one before. What the whole run misses is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a package may stand
- L3
- Puts the signature in an engineer's hands
- L4 – L5
- Keep the event and the record behind it
- L6
- Narrows to chronology assembly when signals degrade
How Nestack evaluates it
Evaluate the assembly workflow — not only the filed chronology.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the filing the agency reads
- E1Final-output evaluationDid every event in the chronology match its source record and its date?
- E2Step-level evaluationDid the agent use the right sources, reporting rules and clock configuration?
- E3Tool evaluationDid it read and write the correct campaign and the correct field?
- E4Confidence calibrationDo low-confidence packages actually attract more engineer amendments?
- E5Slice evaluationHow does performance change across specific campaign types?
- E6Business outcomeHow many packages needed an amendment or a correction after filing?
Floor — the outcome the manufacturer 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
- RJ-03
Source omitted
Supplier 8D or foreign field action never reaches the file.
Stage gathers
Warranty claims, field reports, 8D and foreign actions
02 · Reasoning
2 modes
- RJ-04
Determination date drift
The clock is run from the flag, not the decision.
- RJ-06
OTA fix left unfiled
A software remedy is closed without a Part 573 filing.
Stage proposes
Chronology, population, clock dates and confidence
03 · Tool / write
2 modes
- RJ-02
Quarterly series cut short
Six consecutive § 573.7 returns are set up as four.
- RJ-05
Stale VIN lookup
The seven-day refresh lands after the filing did.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- RJ-01
Risk paragraph off-form
The § 577.5(f) wording is replaced by engineering prose.
Stage returns
The chronology the engineer signs and the agency reads
05 · Change / Version
1 mode
- RJ-07
Silent rule regression
A model or rule change loosens what counts as an event.
Stage tracks
Model, prompt, reporting rules and clock config
- Sev-1 · a filing goes out unsigned
- Sev-2 · a wrong date reaches the chronology
- Sev-3 · a source degrades, package held
Affected slices
The campaigns you did not open carry the units
A programme-wide filing rate can read as timely while a few campaign cohorts carry most of the late chronologies. Nestack reports the engineer-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 |
|---|---|---|---|---|
| Agency-influenced campaigns | 9.5% | 3.7× | Review | |
| Supplier-originated defects | 7.6% | 3.0× | Review | |
| Foreign field actions | 4.8% | 1.9× | Watch | |
| Manufacturer-opened campaigns | 2.1% | 0.8× | Normal |
Bar: amendment-rate lift vs. manufacturer-opened baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop closes on a test, not a memo
A cycle is done when the missed signal has become a case the next release must pass. That suite is what the next determination made is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
The § 573.15 lookup goes stale, or amendments rise in a slice.
- 02
Diagnose
The signal that was visible long before anyone named it is read back until one cause holds.
- 03
Improve
Stamp the change with a version; the campaigns behind it are filed against it.
- 04
Verify
Every touched case is put through again, and a single red stops the release dead.
- 05
Learn
It stays as a standing test, and the reporting rules are rewritten alongside it.
Learn → Detect
The return edge. The next signal is caught 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, assembly workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Recall workflow discovery and boundary definition.Week 1
- 02Warranty, field and supplier source review.Week 1, Week 2
- 03Part 573, Part 577 and quarterly reporting rules.Week 1, Week 2
- 04Field-record ingestion and date mapping.Week 2
- 05Chronology logic and source binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Engineer sign-off workflow.Week 3, Week 4
- 08Filing and campaign-system integration.Week 5
- 09Signal and deadline cases.Week 4, Week 5
- 10Guardrails and reporting controls.Week 4
- 11Campaign-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 campaign, one market | ProductionProduction filing systems | AdvancedMultiple markets / brands |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your sources and rules | ✓ | ✓ | ✓ |
| Engineer sign-off | ✓ | ✓ | ✓ |
| Chronology-completeness baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by campaign | — | ✓ | ✓ |
| Sign-off workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Field-data integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-market reporting rules | — | — | ✓ |
| Multi-stage safety-office review | — | — | ✓ |
| High campaign volume | — | — | ✓ |
| Multi-market campaign 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, warranty and supplier sources
→Field-record ingestion and date-of-receipt mapping
Week 102Representative closed campaigns
→Chronology baseline, source binding and clock modelling
Week 203Your reporting rules and notice templates
→Part 573, Part 577 and quarterly reporting rules
Week 104Access to relevant APIs, feeds or exports
→Warranty, field and supplier assessment, then integration setup
Week 205Determinations you would not want dated
→Timeliness cases and the evaluation suite
Week 406What no chronology may leave out
→Confidence scoring, gap routing, guardrails and sign-off controls
Week 307Named safety-office engineers to review packages
→Engineer sign-off 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 plan is drawn on elapsed weeks, not on slide space, so week 5 legitimately doubles.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Recall workflow discovery, rule mapping and the automation boundary
W2Source integration and the chronology baseline
W3Assembly workflow, confidence logic and sign-off controls
W4Evaluation suite, clock checks and failure-mode testing
W5Filing integration, pilot campaigns and targeted corrections
W6One campaign run under the safety-office engineer, then Agent Care handover
- Reading the band
- Each bar sits on the weeks its work is named in and no others. The week 5 doubling is real work, not padding.
- At the end of W6
- The closing checks pass on live campaigns, and Agent Care takes monitoring on.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Automotive AI agent
Build a safety-recall agent around your determination clock.
Show us your field sources, your reporting rules and who determines. The determination, its date and the signature on the Part 573 report stay with your safety-office engineer.
- Typical delivery
- 4–6 weeks
- Deployment
- Your field data
- Engineer signs
- 5 working days
- After launch
- Agent Care available
Nestack Agents · Safety recall & field service
AGT-AUT-10 · Agent Care available after launch