Safety-Coaching & Driver-Workforce AI Agent

What this agent does

Surfaces the event, never scores the person

Ingesting triggered events, telematics and video from supported ELD, camera and fleet-safety sources.

Carrying the record behind each event — the footage or the trace — through to the coaching queue.

Reason

Prioritising events for a supervisor's conversation, because the conversation is the part the evidence credits.

Stating what is being measured: a triggered event, not a crash, and those are two different claims.

Keeping the score out of hiring, discipline, discharge and load allocation, and holding anything that touches employment.

Decide

Auditing that no event penalises a required break, a refusal on hours or fatigue grounds, or a reported defect.

Routing each flagged event to the named supervisor, who decides whether it is coached at all.

Out

Retaining the event, its record and the coaching outcome under a schedule the fleet approves.

Executing write actions only inside the approval boundaries agreed during implementation.

→ Product statement

The agent surfaces events; a named supervisor holds the conversation, and no telematics score decides a person's employment.

Example workflow

One event, trigger to conversation

AgentHuman

1Event triggeredTelematics trigger, in-cab camera event, ELD signal or a reported concern

2Record assembledThe footage or trace, the route, the equipment, the conditions and the prior coaching history

3Event preparedEvent, record, exposure notes and confidence

4Controls appliedEmployment-use blocks, protected-activity checks, exposure notes and the confidence threshold

No human action required

Stages 1 to 4 run unaided and reach no driver — the event is prepared, and the supervisor's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the named supervisor to coach.

Low confidence

Adds a safety-desk read first.

Supervisor coaches

The event is held with its record, its exposure notes and the confidence.

Coach · Dismiss · Send to the safety desk

Coached — conversation recorded▼

6Fleet-safety records updatedOnly where write access and approval policy allow it

7Outcome evaluatedFlag precision, dismissed events, coaching completion and repeat events by cause

Dismissals

Every supervisor dismissal is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Hiring, discipline, discharge or promotion decisions.

Allocating loads or ranking drivers by a score.

Deciding whether an event is coached, and how.

Holding the coaching conversation with a driver.

Automation boundaryAgent acts unaided

✓Surface triggered events with the record behind each one.

✓Group events by cause so a supervisor can see the pattern into the review queue.

✓Note the route, equipment and conditions around the event.

✓Flag an event that penalises protected activity, and hold it for the named owner.

Any write happens inside the boundaries agreed at implementation, never ahead of the supervisor.

Turning on in-cab audio or a face-based model.

Directing how an owner-operator performs the work.

Setting or lifting a retention or litigation hold.

Changes to coaching rules, thresholds or approvals.

Example output

One triggered event, annotated

Everything the agent surfaces is attached to the record it came from.

Coaching output · single eventIllustrative example

Event

What triggered

Duration

Record on file

Confidence

Employment use

Following distance

Close following on a wet interstate grade, running downhill

12 seconds

Forward camera clip

89%

Blocked — coaching only

As receivedTaken from the telematics trigger and the clip on file — nothing on this side is a judgement about the driver.

What the record shows The forward clip The route and the grade The weather at the time

Why it is not a scoreOne event, on one road, in one condition and a person still decides.

ActionCoachDismissSend to the safety desk

What the score decidesBelow the configured threshold the event picks up a safety-desk read before it.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every triggered eventFrom telematics or the in-cab camera

03Coaching

Surface it for the conversation

Draw on the event record, the route and conditions around it, and the coaching rules your fleet configured.

01Approved path

Coach it, never score the person

Routine events arrive grouped by cause, with the record attached.

02Human review

Pass the rest to the supervisor

Ambiguous events, protected-activity flags and low-confidence triggers are marked, so the supervisor's read starts where risk concentrates.

04Build an evidence trail

The event, the footage or telematics behind it and the supervisor who coached stay on the record.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Telematics and ELDSamsara · Motive
Omnitracs · Platform Science

Video safetyLytx · Netradyne
Forward and in-cab cameras

Driver and training recordsDriver files · HR systems
Training and LMS platforms

Agent

Safety coaching and workforce

Reads the event
Groups by cause
Holds for a supervisor

Safety and claimsSafety management systems
Incident and claims records

Observability & evaluationOpenTelemetry · 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 driver

The controls are nested. What survives all of them appears in the map underneath.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeHold coaching for the supervisor when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, trigger-threshold and coaching-rule changes.Track

L4TraceabilityRecord the event, the footage or trace, the flags, the decision and who coached.Record

L3Supervisor gateHold events for a named supervisor; it governs the conversation, not whether the trigger was fair.Gate

L2Policy guardrailsTest events against the configured employment-use and protected-activity rules; a failure returns the event.Restrict

L1Confidence thresholdsRoute low-confidence events to a safety-desk read before the supervisor sees them.Require review

Model coreEvent produced — trigger, record, exposure notes and confidence

L1 – L2Test whether an event may stand

L3Puts the conversation in a supervisor's hands

L4 – L5Show the event and the record behind it

L6Holds coaching for the supervisor when signals degrade

How Nestack evaluates it

Evaluate the coaching workflow — not only the trigger that fired.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the event a supervisor sees

Depth of coverage ▼

E1Final-output evaluationDid the event match the record behind it?

E2Step-level evaluationDid the agent read the route, the conditions and the equipment around it?

E3Tool evaluationDid it read and write the correct driver record and the correct event?

E4Confidence calibrationDo low-confidence events actually attract more supervisor dismissals?

E5Slice evaluationHow does performance change across specific route and duty types?

E6Business outcomeHow many events were dismissed, or coached and then disputed?

Floor — the outcome the fleet answers for

Failure modes

Where each failure originates in the agent

Seven ways a flag goes wrong, set at its own stage.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

BF-03

Record not retrieved

The footage or trace behind an event is missing.

Stage gathersTriggers, footage, routes, conditions and coaching history

02 · Reasoning2 modes

BF-04

Exposure dropped

Route, weather and equipment fall out of the read.

BF-06

Protected activity flagged

A required break or a refusal is surfaced as a fault.

Stage proposesEvent, record, exposure notes and confidence

03 · Tool / write2 modes

BF-02

Score reaches employment

A telematics figure informs a decision about a person.

BF-05

Duplicate event raised

One incident is surfaced twice from two feeds.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

BF-01

Event without its record

A flag arrives with no footage or trace behind it.

Stage returnsThe event the supervisor reads and coaches on

05 · Change / Version1 mode

BF-07

Silent threshold regression

A model or threshold change widens what the agent will flag.

Stage tracksModel, prompt, trigger thresholds and coaching rules

Sev-1 · a score touches an employment decision Sev-2 · a wrong flag reaches a driver Sev-3 · a feed degrades, event routes to review

See the complete Transportation failure-mode catalogue →

Affected slices

One route type can carry the false flags

Base rates differ by route before any model runs. Night and long-haul work triggers far more often than regional daytime work, so a flag there means less. Nestack reports the dismissed-flag rate by route type, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Night and long-haul runs	8.2%	3.4×	Review
Urban stop-and-go routes	5.5%	2.3×	Review
Winter and wet conditions	4.3%	1.8×	Watch
Regional daytime runs	2.2%	0.9×	Normal

Bar: dismissed-flag-rate lift vs. regional-daytime baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

A cycle ends in a standing case

Explaining a bad flag does not close the cycle. A standing regression case does, and that suite is what the next event flagged is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Dismissed flags rise in one route type.

02Diagnose

If the flag was wrong, the event, the record behind it and the coaching note are read back together until the cause is one thing.

03Improve

The fix is versioned against the events that produced it.

04Verify

The release stops until every affected case is passing.

05Learn

The case stays in the suite, and the coaching policy is amended.

Learn → DetectThe return edge. The next event flagged 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, feeds, coaching workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Coaching workflow discovery and boundary definition.

02Telematics, video and ELD source review.

03Trigger, exposure and coaching-policy rule mapping.

04Event ingestion and record normalisation.

05Cause grouping and exposure logic.

06Confidence scoring and event routing.

07Supervisor coaching workflow.

08Telematics and video integration.

09False-flag regression cases.

10Guardrails and employment controls.

11Event-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 terminal, one fleet ProductionProduction telematics AdvancedMultiple terminals / fleets

Introduced at Pilot

Surfacing events to your coaching rules✓✓✓

Supervisor coaching gate✓✓✓

Flag-precision baseline✓✓✓

Introduced at Production

Reporting by event type—✓✓

Coaching workflow in your systems—✓✓

Approved record write-back—✓✓

Telematics integration—✓✓

Introduced at Advanced

Multi-state and union rule sets——✓

Multi-stage safety approvals——✓

High event volume——✓

Multi-terminal coaching controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.

Separate from buildBuild 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 bringWe build with it

01Your coaching policy and who holds the conversation → Coaching-rule mapping and the employment-use boundaryWeek 1

02Representative events, including the ones you dismissed → Flag baseline, cause grouping and exposure notesWeek 2

03Your route, duty and equipment mix → Exposure mapping and automation-boundary definitionWeek 1

04Access to relevant APIs, feeds or exports → Telematics, video and ELD assessment, then integration setupWeek 2

05Flags you would not want acted on → False-flag cases and the evaluation suiteWeek 4

06What must reach a supervisor before it touches employment → Confidence scoring, event routing, guardrails and employment-use controlsWeek 3

07Named supervisors to coach on events → Supervisor coaching workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases occupy the weeks the work really needs, so week 5 runs evaluation and launch side by side.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Coaching workflow discovery, rule mapping and the automation boundary W2Telematics integration and the flag baseline W3Coaching workflow, confidence logic and supervisor controls W4Evaluation suite, false-flag cases and failure-mode testing W5Video integration, pilot terminals and targeted corrections W6One coaching cycle run under safety, then handover

Reading the bandA bar runs across the weeks its work is named in and no more. The week 5 overlap is real work, not padding.

At the end of W6The cycle closes validation and Agent Care picks up the monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Transportation AI agent

Build a coaching agent around the supervisor who holds the conversation.

If you show us your coaching policy, your event feeds and who holds the conversation, we'll map the workflow, set the automation boundary and name what never touches employment.

Discuss this agent →

Nestack Agents · Safety coaching and driver workforceAGT-TR-17 · Agent Care available after launch