HR AI agent · Learning and skills
Learning and Skills Development AI Agent
Assemble the course path, record what each session actually covered, and hold each mandated completion for the named signer who certifies that a particular person was trained.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named signer
- Able, not done
- After launch
- Agent Care
What this agent does
Assembles the path, never certifies it
In
- 01
Most of a catalogue binds nobody, and a named list of mandated courses binds hard.
- 02
A mandated session runs, and 29 CFR 1926.503(b)(1) wants a name, the dates and a signature.
Reason
- 03
A completion is logged, and it records attendance without claiming the person is now able.
- 04
A hazard course is given, and 29 CFR 1910.1200(h) asks for no training record.
- 05
A workplace changes, and the 1926.503(c) retraining triggers start outside any learning system.
Decide
- 06
A retention clock is set per standard — Part 1904 covers injury records, not training ones.
- 07
A path becomes a prerequisite, and 42 U.S.C. 2000e-2(d) reaches training in its own subsection.
Out
- 08
An AI-literacy course is enrolled, and Article 4 has bound deployers since 2 February 2025.
- 09
Execute enrolment and record writes only inside the approval boundaries agreed at implementation.
Product statement
Assembly, sequencing and recording belong to the agent. Certification belongs to a named signer, and on the day a learning path becomes a prerequisite the ranked list it produced is a selection procedure within the definition at 29 CFR 1607.16(Q).
Example workflow
One mandated course, roster to certification
- 1Training requirement identifiedThe standard, the people it reaches, the frequency it names and the subjects it lists
- 2Course path assembled and clockedThe course chosen, the subjects it is said to cover and the day the requirement next falls due
- 3Session run and completions loggedWho attended, on which day, and the material the session is recorded as having covered
- 4Checks runSubject-coverage checks, frequency checks, retention-clock checks and record confidence
No human action required
Stages 1 to 4 run unaided, and nothing is certified at any of them — 1926.503(b)(1) puts a human signature on the record itself, and the signer lane opens at the certification gate.
Goes to the named signer to certify.
Adds a safety-lead read first.
Signer review
The record is held with the session it names, the subjects claimed and the checks that ran.
Certify · Return for correction · Send to safety review
Certified — by the named signer
- 6Learning and compliance records updatedOnly where write access and the retention policy allow it
- 7Outcome reviewedSubject coverage, frequency adherence, corrections and what review found
Corrections
Each signer correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Signing a training certification record.
- Judging that a worker can do the work.
- Deciding that earlier training was adequate.
- Making a course a condition of promotion.
Automation boundary
Agent acts unaided
- ✓Record what a session covered and the day it ran.
- ✓Hold each certification record unsigned for a named signer to read.
- ✓Test each course against the subjects a standard specifically names.
- ✓Take the retention clock from the standard itself.
Nothing is certified except by a named signer who can answer for it, inside the boundaries.
- Deciding whether a retraining trigger has fired.
- Telling an inspector that a worker is competent.
- Choosing who is routed to a remedial course.
- Changes to the mandate register or the clocks.
Example output
One completion record, annotated
This serves a learning team whose records may be read by an inspector years after the session ran; below is one completion record exactly as the agent leaves it, uncertified. Which person is matched to which role is a separate question and belongs to another agent.
Completion record · certification pending
Illustrative example- Stage
- Mandated course, session completed
- Recorded as
- Completion logged, not certified
- Next due
- Certification open
- Evidence on file
- Roster and session log, 5 August 2026
- Confidence
- Held uncertified
- Held for
- The named signer, by name
As received
Assembled from a roster and a session log, and it claims nothing about what the worker can now do.
What this record carries
- Roster of attendees
- Subjects claimed
- Date the session ran
Why no certification here
Certifying that a named worker is able is a judgement a human makes.
What the score decides
Annex III point 3 reaches training institutions, not an employer running its own courses.
Action
CertifyReturn for correctionSend to safety 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 completion record
From the session that produced it
Where the evidence lands
29 CFR 1910.132 has no paragraph (f)(4) and no written certification of PPE training; the certification at (d)(2) names a workplace and a hazard assessment, not a trained worker.
Finished is not the same as able
29 CFR 1910.178(l)(6) asks for the training date and the evaluation date, and (l)(4)(iii) puts the second one on a cycle of three years. The regulator wrote this distinction down before anyone built a dashboard for it.
Searched for, and not found
eCFR full-text returns nothing for learning path, skills gap, competency framework, skills inventory or reskilling, while the controls training program, retraining and training records return freely. The near misses are Head Start and federal-employee rules, not corporate learning.
Build an evidence trail
The completion, the session it records and the trainer who signed it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Learning systems
- LMS
- catalogue
- session logs
- Enrolments and completion dates
Compliance record stores
- Certification store
- registers
- Signed certification dates
Work and hazard systems
- Shift systems
- incident reports
- Exposure rosters and change orders
Agent
Mandated courses and learning paths
- Reads the mandates
- Assembles the path
- Holds for the signer
People and role records
- HRIS
- job records
- shift rotations
- Who a mandate covers, and when
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six clamps between the completion and the signature
Six clamps set around one beam, the last the tightest. What holds is drawn in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a record may go forward
- L3
- Leaves the certification to a named signer
- L4 – L5
- Keep the completion and the session behind it
- L6
- Records attendance and claims no competence when signals degrade
How Nestack evaluates it
Evaluate the whole learning run — not only the record that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the record an inspector would read
- E1Final-output evaluationDid the record name the subjects the session actually covered?
- E2Step-level evaluationDid the agent read the right mandate, the right people and the live catalogue?
- E3Tool evaluationDid it write to the correct worker and the correct course?
- E4Confidence calibrationDo low-confidence records actually attract more signer corrections?
- E5Slice evaluationHow does performance change across specific course classes?
- E6Business outcomeHow many records needed a correction before the signer certified?
Floor — the mandate a record is run against
Failure modes
Where each failure originates in the agent
Seven failure modes, placed where each first surfaces in the run.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- VL-03
Retraining trigger never arrives
The signal that restarts the clock sits outside the system.
Stage gathers
The roster, the course and the dates it ran
02 · Reasoning
2 modes
- VL-04
Right course, wrong subjects
Catalogue wording is read as a subject list.
- VL-06
Gap inferred from a proxy
Tenure or shift pattern stands in for a skill.
Stage proposes
The mandate, the subjects claimed and the clocks
03 · Tool / write
2 modes
- VL-02
Old completion read as current
A tick set long ago is reported as ability.
- VL-05
One retention clock for five
A single policy is run over unlike standards.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- VL-01
Certified, teaching unrecorded
The record carries a signature but not the teaching.
Stage returns
The record an inspector may later ask to see
05 · Change / Version
1 mode
- VL-07
Silent gate drift
A path becomes a prerequisite outside the product.
Stage tracks
Model, prompt, mandate rules and course data
- Sev-1 · a signature nobody could support
- Sev-2 · a mandated due date goes past
- Sev-3 · subjects short, record held back
Affected slices
Mandated safety courses absorb the corrections
No completion rate measures conduct, and what people know is found to fade, so an estate figure reads clean while mandated safety courses carry the corrections. Nestack reports by class.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Mandated safety courses | 7.7% | 3.7× | Review | |
| Multi-jurisdiction mandates | 5.4% | 2.6× | Review | |
| Retraining-trigger cases | 3.4% | 1.6× | Watch | |
| Ordinary catalogue courses | 1.6% | 0.8× | Normal |
Bar: correction-rate lift vs. catalogue-course baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a green dashboard costs
A loop closes when the mandated session recorded unsigned is a standing case. That suite is what the next training period is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on mandated safety courses.
- 02
Diagnose
The crew sent to a spill on completion records that nobody had signed is worked backwards until one cause is left standing.
- 03
Improve
Records go out numbered, and the sessions beneath them ride with it.
- 04
Verify
One signature case still failing is enough to hold the record back.
- 05
Learn
It is kept for good, and the recording rules are amended in the same commit.
Learn → Detect
The return edge. The next training period 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, systems, record build, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Certification-record and mandate-register boundary.Week 1
- 02Learning, roster and safety-record data.Week 1, Week 2
- 03Session-to-signature and mandate-coverage reporting.Week 1, Week 2
- 04Course and roster intake.Week 2
- 05Mandate, course and clock binding.Week 2, Week 3
- 06Coverage scoring and review routing.Week 3
- 07Signer certification workflow.Week 3, Week 4
- 08Learning-system integration.Week 5
- 09Signature and mandate cases.Week 4, Week 5
- 10Guardrails and enrolment controls.Week 4
- 11Completion-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 mandate, one training period | ProductionProduction learning workflow | AdvancedMultiple mandates / jurisdictions |
|---|---|---|---|
| Introduced at Pilot | |||
| Path assembly to your mandates | ✓ | ✓ | ✓ |
| Named signer certification | ✓ | ✓ | ✓ |
| Mandate-register baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by course class | — | ✓ | ✓ |
| Signer review workflow in your systems | — | ✓ | ✓ |
| Approved enrolment write-back | — | ✓ | ✓ |
| Catalogue-and-roster integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction mandate rules | — | — | ✓ |
| Cross-site certification packs | — | — | ✓ |
| Large course catalogues | — | — | ✓ |
| Multi-mandate frequency controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, course 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 mandated courses and the standards behind them
→Mandate capture and certification-boundary definition
Week 102Representative LMS, roster and safety-record access
→System binding, subject logic and the completion baseline
Week 203Your sites and the signers they name
→Mandate mapping, subject binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→LMS, roster and safety-record assessment, then integration setup
Week 205Records you would not want cited
→Mandate cases and the failure round
Week 406What no completion may demonstrate
→Coverage scoring, review routing, guardrails and release controls
Week 307A named signer who certifies
→Release to the named signer, 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
Week five holds two bands here because the two phases genuinely coincide, not to even the column out.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Mandate discovery, certification boundary and the automation boundary
W2System integration and the completion-record baseline
W3Subject logic, frequency clocks and release controls
W4Evaluation suite, signature cases and the failure round
W5Learning-system integration, pilot cohorts and targeted corrections
W6One training period run under the learning lead, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work names, and week five is shared by design.
- At the end of W6
- Once the completion record validates, Agent Care assumes the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Learning AI agent
Build a learning agent around the competence your last certificate assumed.
Show us one mandated course and the record it produced. If the signer cannot say which subjects were covered, the signature carries a claim the completion never made. OSHA priced that gap across three employers at Channelview, Texas on 26 June 2026.
- Typical delivery
- 4–6 weeks
- Deployment
- Your live mandates
- Named signer
- Able, not done
- After launch
- Agent Care available
Nestack Agents · Learning and skills
AGT-HR-13 · Agent Care available after launch