Construction AI agent · Change control
Building Safety Change-Control AI Agent
A major change carried out before it is granted cannot be regularised, and on a higher-risk building in England the approval shields nothing. The agent assembles the record; the client signs it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Client signs
- Per change
- After launch
- Agent Care
What this agent does
Drafts the record behind a change, not the answer to a question
In
- 01
A design moves, and the agent opens a change record against the agreed documents the project holds.
- 02
A change meets reg 26, and the agent drafts recordable, notifiable or major with the limb it relied on.
Reason
- 03
An adviser is consulted, and the agent records the name, the occupation and a summary of the advice.
- 04
A document is affected, and the agent lists it and checks the revised version exists before work starts.
- 05
A compliance position is needed, and the agent drafts it for the person in responsible charge to settle.
Decide
- 06
A notifiable change is drafted, and the agent holds the reg 20 notice unsent until the client signs it.
- 07
A major change is drafted, and the agent stages the reg 21 application; reg 18 bars the work until granted.
Out
- 08
A change closes, and the entry, the advice, the revisions and the signature stay against the project.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts and the client signs — and once a major change is carried out, no regularisation application may be made, so the work sits outside the approval.
Example workflow
One change, proposal to record
- 1Change proposedSite instruction, design revision, product substitution or dutyholder change
- 2Documents assembledThe agreed documents, the drawings they govern and the golden-thread revision each one sits at
- 3Category draftedRecordable, notifiable or major, the reg 26 limb relied on and confidence
- 4Controls appliedReg 19 element checks, consultation-record checks, document-impact checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is classified, notified or signed at any of them — the agent is drafting, and the client's lane opens at the confidence gate.
Goes to the named client.
Adds a principal designer read first.
Client decision
The change is held with its draft category, its consultation record and the confidence.
Adopt · Recategorise · Send to principal designer
Adopted — signed and released
- 6Log and golden thread updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedRecategorisations, missing elements, information requests and changes reworked at completion
Recategorisations
Every recategorisation made in review is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Deciding the category a change falls into.
- Signing the change record, notice or application.
- Stating that the changed work complies.
- Starting work before a major change is granted.
Automation boundary
Agent acts unaided
- ✓Propose the category against reg 26 and cite the limb relied on.
- ✓Assemble the reg 19 elements, including who advised and what they advised.
- ✓List the agreed documents affected and check a revised version exists.
- ✓Flag what a change record is missing, and hold it for the client.
Any write happens inside the boundaries agreed at implementation, never ahead of a signature.
- Taking over the log the principal contractor maintains.
- Signing a dutyholder compliance declaration.
- Deciding that a safety occurrence is reportable.
- Changes to category rules or approval thresholds.
Example output
One change, annotated
An RFI moves a question to its responder; this moves a change to the client who signs.
Change record · single controlled change
Illustrative example- Element
- External wall build-up
- Proposed change
- Insulation replaced with a board of the same reaction-to-fire class
- Agreed document
- FEF rev C
- Draft category
- Notifiable — reg 26
- Confidence
- 93%
- Consequence
- Notice before the change is carried out
As received
Taken from the agreed documents and the golden thread as they stand — nothing on this side is written by the agent.
What was read
- Fire and emergency file
- Approved wall build-up
- Product classification
Why this category
The replacement is not of a lower class, so reg 26 puts it below a major change — the client settles it.
What the score decides
Below the configured threshold the change picks up a principal designer read before the client sees it.
Action
AdoptRecategoriseSend to principal designerValue
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 proposed change
From the agreed documents
Draft against the approved set
Read the change against the agreed documents BSR approved, not against the latest drawing in the model.
The client signs each change
Routine recordable changes arrive already written up against reg 19.
Send review to the ones that stop work
Notifiable and major drafts are marked, so the client's read starts where the prohibition bites.
Build an evidence trail
The change, the advice it drew on and the client who signed 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.
Golden thread and agreed documents
- Zutec
- Operance
- Asite
- Autodesk ACC
Design and coordination
- Autodesk Revit
- Navisworks
- Solibri
- Bluebeam Revu
Project and change management
- Procore
- Aconex
- Viewpoint
- Fieldwire
Agent
Building safety change control
- Reads the documents
- Drafts the record
- Holds for the client
Submission and handover
- BSR online applications
- Reg 38 handover pack
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six rings between the model and the signature
Six rings read from the outside in. What clears the last of them is drawn in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a draft may stand
- L3
- Puts the decision in the client's hands
- L4 – L5
- Keep the change and the advice behind it
- L6
- Holds the change at unrecorded when signals degrade
How Nestack evaluates it
Evaluate the change workflow — not only the finished record.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the record the regulator reads
- E1Final-output evaluationDid the draft carry every element reg 19(2) prescribes?
- E2Step-level evaluationDid the agent read the agreed documents and the reg 26 limbs in force?
- E3Tool evaluationDid it read and write the correct project and the correct log?
- E4Confidence calibrationDo low-confidence changes actually attract more recategorisations?
- E5Slice evaluationHow does performance change across specific change types?
- E6Business outcomeHow many changes were recategorised, or reworked at completion?
Floor — the record the client signs and defends
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 →
01 · Retrieval
1 mode
- FY-01
Superseded agreed document
The change is read against a drawing, not the approved set.
Stage gathers
Agreed documents, drawings and golden-thread state
02 · Reasoning
2 modes
- FY-02
Category drafted too low
A change is drafted as recordable where reg 26 makes it notifiable.
- FY-03
Adviser recorded by role
The name and occupation are dropped, and so is the advice given.
Stage proposes
Category, affected documents and confidence
03 · Tool / write
2 modes
- FY-04
Compliance position asserted
A draft reads as a conclusion the responsible designer never gave.
- FY-05
Record dated after the act
The entry is accurate and complete, and made after the change was carried out.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- FY-06
Document impact missed
The drawing is revised and the fire and emergency file is not.
Stage returns
The record the client signs and the regulator reads
05 · Change / Version
1 mode
- FY-07
Designated change missed
A reg 25 designation never reaches the category rules.
Stage tracks
Model, prompt, category rules and reg 26 config
- Sev-1 · the agent performs a held act
- Sev-2 · a wrong category reaches the record
- Sev-3 · documents degrade, change to review
Affected slices
A clean project figure can sit on one change type
A project-level change-record completeness figure can read well while a few change types carry the missing elements. BSR publishes nothing at all on change control, so Nestack reports yours by slice.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Product substitutions | 9.6% | 3.7× | Review | |
| External wall changes | 7.5% | 2.9× | Review | |
| Dutyholder changes | 4.2% | 1.6× | Watch | |
| Internal refurbishment | 1.6% | 0.6× | Normal |
Bar: missing-element-rate lift vs. an internal-refurbishment baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop ends in a regression case
A cycle closes when the unrecorded change is a regression case. That suite is what the next change worked is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Missing-element rate rises in a change type.
- 02
Diagnose
The detail that moved after the approval it was drawn against is traced back to one step.
- 03
Improve
The change ships numbered, and the projects that forced it ride with it.
- 04
Verify
Nothing releases while one touched change case is still red.
- 05
Learn
It is kept for good, and the category rules are amended in the same commit.
Learn → Detect
The return edge. Every later detection meets a suite that is 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, change workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Change-control discovery and boundary definition.Week 1
- 02Golden-thread and CDE source assessment.Week 1, Week 2
- 03Agreed-document and reg 26 category rule mapping.Week 1, Week 2
- 04Change ingestion and document matching.Week 2
- 05Category drafting and reg 19 assembly.Week 2, Week 3
- 06Confidence scoring and change routing.Week 3
- 07Client review workflow.Week 3, Week 4
- 08Golden-thread and CDE integration.Week 5
- 09Category and consultation cases.Week 4, Week 5
- 10Guardrails and recording controls.Week 4
- 11Change-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 project, one change log | ProductionProduction golden-thread systems | AdvancedMultiple projects / clients |
|---|---|---|---|
| Introduced at Pilot | |||
| Drafting to your agreed documents | ✓ | ✓ | ✓ |
| Client review gate | ✓ | ✓ | ✓ |
| Consultation-record baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by building | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Design-register integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-building rule sets | — | — | ✓ |
| Multi-stage dutyholder review | — | — | ✓ |
| High change volume | — | — | ✓ |
| Multi-project change controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, change 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 agreed documents and change control plan
→Agreed-document mapping and change ingestion
Week 102Representative changes and their records
→Category-drafting baseline and reg 19 element assembly
Week 203Your reg 26 category rules and decision procedure
→Category-rule mapping and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Golden-thread and project system assessment, then integration setup
Week 205Change records you would not want pulled
→Category cases and failure-mode testing
Week 406What no change record may omit
→Confidence scoring, change routing, guardrails and approval controls
Week 307Named client signatories to review drafts
→Client 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 band here is sized to the weeks its own phase absorbs, so the fifth honestly carries a pair.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Change-control discovery, rule mapping and the automation boundary
W2Source integration and the category-drafting baseline
W3Reg 19 assembly, confidence logic and review controls
W4Evaluation suite, guardrails and failure-mode testing
W5Golden-thread integration, pilot changes and targeted corrections
W6One change cycle run under the client duty-holder, then Agent Care handover
- Reading the band
- A bar spans the weeks its own work is named in. The week 5 overlap is worked time, not padding.
- At the end of W6
- Validation closes on live changes, and Agent Care picks up the watch.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Construction AI agent
Build a change-control agent around the record reg 19 already prescribes.
The Lords committee found in December 2025 that applications are rejected for an inability to evidence how safety was considered. On a higher-risk building in England, reg 19 already prescribes that record and reg 18 forbids proceeding without it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your golden thread
- Client signs
- Per change
- After launch
- Agent Care available
Nestack Agents · Building safety change control
AGT-CN-09 · Agent Care available after launch