Main site

Construction AI agent · Safety compliance

Safety-Compliance AI Agent for SWMS & Permits

A safe work method statement is the output of consulting the crew, not a text artefact. The agent flags scopes, diffs submitted documents and holds each finding for the duty-holder who signs.

Typical delivery
4–6 weeks
Deployment
Your stack
Duty-holder review
Findings only
After launch
Agent Care

What this agent does

Finds the gaps; it does not write the SWMS

In

  1. 01

    Watch package scopes, submitted SWMS, permits and expiry registers as they arrive.

  2. 02

    Diff the scope wording against the high-risk-work limbs of the regime set for the project.

Reason

  1. 03

    Flag each scope that matches a limb, citing the limb, for a person to classify.

  2. 04

    Diff a submitted SWMS against the mandatory-content list: hazards, controls, monitoring and review.

  3. 05

    Flag controls that name no implementable step, and controls the site conditions do not support.

Decide

  1. 06

    Watch licence, ticket, induction and permit expiry against the named owner for each.

  2. 07

    Route every finding to the duty-holder's nominated signatory, not to a head-contractor approver.

Out

  1. 08

    Assemble the retention pack — versions, permits, consultation records — against the statutory clocks.

  2. 09

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

Product statement

The agent lists gaps; the PCBU carrying out the work signs the SWMS — and if a defective one was used, only that duty-holder stops the work and re-consults.

Example workflow

One HRCW package, scope to sign-off

  1. 1Scope of works receivedPackage scope, site conditions, a submitted SWMS or a permit request
  2. 2Jurisdiction resolvedThe regime set for the project — a model-WHS state, Victoria's OHS Regulations, OSHA or EM 385-1-1
  3. 3Gaps listedMatched limbs, missing mandatory content, expiries and confidence
  4. 4Controls appliedJurisdiction gate, mandatory-content checks, expiry checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is classified or signed at any of them — the agent is listing gaps, and the reviewer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the sub's nominated signatory.

Low confidence

Adds a site safety read first.

Duty-holder review

The finding is held with the limb cited, the site conditions and the confidence.

Accept · Dispute · Return to subcontractor

Accepted — back with the duty-holder

  1. 6Safety systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedGaps upheld, disputed findings, permit interfaces and defects found after use on site

Disputes

Every disputed finding is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Deciding whether work is high risk construction work.
    • Authoring or adopting the SWMS as the duty-holder.
    • Conducting the consultation with workers and the HSR.
    • Issuing, signing or cancelling a confined space permit.

  • Automation boundary

    Agent acts unaided

    • ✓Flag scopes whose wording matches a high-risk-work limb, for a person to classify.
    • ✓Diff a submitted SWMS against the mandatory-content list.
    • ✓Watch licence, induction, ticket and permit expiry dates.
    • ✓Assemble the retention pack against the statutory clocks for the configured regime.

    Any write happens inside the boundaries agreed at implementation, and never onto a signed document.

    • Judging a person competent, qualified or licensed.
    • Stopping work, and authorising its resumption.
    • Approving an AHA at the required authority level.
  • Closing out a hazard or non-conformance item.

Example output

One flagged scope, annotated

Everything the agent raises is attached to the scope it was read from.

Gap finding · single scope

Illustrative example
Package
Excavation subcontractor
Scope line
Trenching beside a live gas main
Matched limb
r.291(i)
Finding
No control named for the main
Confidence
91%
Jurisdiction
Model WHS — NSW

As received

Taken from the package scope and the site conditions on file — nothing here is written by the agent.

What was checked

  • Package scope wording
  • Site services drawing
  • Submitted SWMS version

Why this finding

The limb is cited from the regime set for this site, and a person decides it.

What the score decides

Below the threshold the finding picks up a site safety read before the signatory.

Action

AcceptDisputeReturn to subcontractor

Value

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 scope of works

From the package and the site

03Gap check

Apply the regime for the project

No national rule exists — model-WHS limbs, Victoria's own OHS list, or a US site where a JHA is not mandatory.

01Approved path

Find the gaps before the crew

Routine mandatory-content gaps are listed before the document is signed.

02Human review

Send review to the risky scopes

Compound limbs and low-confidence findings are marked, so the reviewer starts where risk concentrates.

04

Build an evidence trail

Each version, the consultation record and the reviewer stay with the document.

Integrations

Typical integrations

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

  • Safety and site operations

    • HammerTech
    • SafetyCulture
    • SWMS, permit and induction modules

  • Workforce and competency

    • Damstra
    • Assignar
    • Rapid Global
    • Pegasus

  • Project systems of record

    • Procore
    • Autodesk Construction Cloud
    • Document control and registers

Agent

Safety compliance — SWMS & permits

  • Reads the scope
  • Lists the gaps
  • Holds for the signatory

  • Registers and permits

    • Regulator HRCW registers
    • Permit-to-work systems

  • 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 crew

Each control wraps the one inside it. What a layer does not catch is named in the failure-mode map below it.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull the agent back to gap-listing only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, template and jurisdiction-configuration changes.Track
L4TraceabilityRecord the version, the limb cited, the reviewer and the closure decision behind each item.Record
L3Duty-holder reviewHold findings for the named signatory; it governs release, not whether the signed document is sound.Gate
L2Policy guardrailsTest findings against the configured regime and the mandatory-content list; a failure returns the finding.Restrict
L1Confidence thresholdsRoute low-confidence findings to a site safety read before the signatory sees them.Require review
Model coreFinding produced — matched limbs, missing content, expiries and confidence
L1 – L2
Test whether a finding may stand
L3
Puts the decision in the signatory's hands
L4 – L5
Show what changed, and when it was consulted on
L6
Narrows the agent to gap-finding when signals degrade

How Nestack evaluates it

Evaluate the compliance workflow — not only the finished document.

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

Surface — the document the crew works to

  1. E1Final-output evaluationDid each cited limb match the regime set for the project?
  2. E2Step-level evaluationDid the agent use the right site conditions, regime and template version?
  3. E3Tool evaluationDid it read and write the correct package and the correct register?
  4. E4Confidence calibrationDo low-confidence findings actually attract more reviewer disputes?
  5. E5Slice evaluationHow does performance change across specific work types?
  6. E6Business outcomeHow many documents needed rework, or a defect found after use on site?

Floor — the trail the controlling employer relies on

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed at the stage where it starts.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • WM-01

      Wrong jurisdiction's list

      Melbourne site checked against model-WHS limbs, not Victoria's.

    Stage gathers

    Scopes, submitted SWMS and the configured regime

  2. 02 · Reasoning

    2 modes

    • WM-02

      Negative HRCW call

      A shallow trench is read as non-HRCW, ignoring the live-services limb.

    • WM-03

      Control with no method

      “PPE and take care” is accepted as a control measure.

    Stage proposes

    Matched limbs, missing content and confidence

  3. 03 · Tool / write

    2 modes

    • WM-04

      Permit pre-filled

      A permit is populated from the prior shift's atmospheric results.

    • WM-05

      Bulk closure on archive

      Open toolbox and hazard items are closed in bulk, ending the trail.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • WM-06

      Unusable reading level

      The crew cannot read the document they must work to.

    Stage returns

    The gap list the signatory reads before signing

  5. 05 · Change / Version

    1 mode

    • WM-07

      Silent template rewrite

      An update rewrites controls on in-force SWMS with no reconsult.

    Stage tracks

    Model, prompt, template and jurisdiction config

  • Sev-1 · a held act performed by the agent
  • Sev-2 · a defective document reaches the crew
  • Sev-3 · source degrades, finding routes to review
See the complete Construction failure-mode catalogue

Affected slices

A safe average can sit on one bad cohort

Split the SWMS-defect rate and the cohort carrying it names itself — mixed jurisdictions, first-time subcontractors, compound scopes. Nestack reports the rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Mixed-jurisdiction portfolios8.6%3.9×Review
First-time HRCW subcontractors6.6%3.0×Review
Repeat HRCW, established sub3.3%1.5×Watch
Single-hazard, template-mature1.1%0.5×Normal

Bar: SWMS-defect-rate lift vs. an all-SWMS baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A cycle is not closed until a test exists

A cycle is done when the failure has become a test the next release has to survive.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    SWMS-defect rate rises in a work-type slice.

  2. 02

    Diagnose

    The defective document and the gap list behind it are read back until the cause narrows to one.

  3. 03

    Improve

    Version-stamp the change and attach the documents that exposed it.

  4. 04

    Verify

    The affected cases run again, and a fail stops the release.

  5. 05

    Learn

    It becomes a standing test, and the document template changes with it.

  6. Learn → Detect

    The return edge. The next detection runs against a suite one case larger.

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, compliance workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Safety workflow discovery and boundary definition.Week 1
  2. 02Safety and workforce source assessment.Week 1, Week 2
  3. 03Jurisdiction, HRCW-limb and permit-rule mapping.Week 1, Week 2
  4. 04Scope and document ingestion.Week 2
  5. 05Limb matching and content-diff logic.Week 2, Week 3
  6. 06Confidence scoring and finding routing.Week 3
  7. 07Duty-holder review workflow.Week 3, Week 4
  8. 08Safety and project-system integration.Week 5
  9. 09Document-defect regression cases.Week 4, Week 5
  10. 10Guardrails and duty-holder controls.Week 4
  11. 11Consultation-trail instrumentation.Week 5, Week 6
  12. 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 tierPilotOne site, one jurisdictionProductionProduction safety systemsAdvancedMultiple sites / jurisdictions
Introduced at Pilot
Gap findings against your regime✓✓✓
Duty-holder review✓✓✓
Document-defect baseline✓✓✓
Introduced at Production
Reporting by HRCW class—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
Regulator-pack export—✓✓
Introduced at Advanced
Mixed-jurisdiction portfolios——✓
Multi-stage principal-contractor review——✓
High document volume——✓
Enterprise WHS controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, document 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.

  1. 01Your jurisdictions and the sites they cover

    Jurisdiction gate and HRCW-limb mapping

    Week 1

  2. 02Representative signed SWMS and permits

    Gap-check baseline, limb matching and mandatory-content diff

    Week 2

  3. 03Your permit set and competency requirements

    Permit-interface mapping and automation-boundary definition

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Safety, workforce and project system assessment, then integration setup

    Week 2

  5. 05Packages where the document missed something on site

    Document-defect cases and failure-mode testing

    Week 4

  6. 06What must reach a duty holder before anything is signed

    Confidence scoring, finding routing, guardrails and approval controls

    Week 3

  7. 07Named signatories to review the findings

    Duty-holder review 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

Each phase is drawn across the weeks it really occupies. Week 5 carries evaluation and pilot at the same time.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Safety workflow discovery, jurisdiction mapping and the automation boundary

W2Source integration and the gap-check baseline

W3Limb matching, confidence logic and review controls

W4Evaluation suite, guardrails and failure-mode testing

W5Project-system integration, pilot packages and targeted corrections

W6One HRCW package run under the PCBU's own review, then handover

Reading the band
Each bar covers the weeks its work is named in and no others. The week 5 overlap is real, not padding.
At the end of W6
The last checks clear on live documents and monitoring moves to Agent Care.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Construction AI agent

Build a safety-compliance agent around your duty-holders.

Show us your jurisdictions, your HRCW scopes and your permit interfaces. The signatory for the PCBU carrying out the work stays the author — we map what the agent hands them, and what it never touches.

Typical delivery
4–6 weeks
Deployment
Your safety stack
Duty-holder review
Findings only
After launch
Agent Care available

Nestack Agents · Safety compliance, SWMS and permits

AGT-CN-02 · Agent Care available after launch

Build a safety-compliance agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your sites

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.