Construction & InfrastructureFuture of AI & Visionary Thinking

Morphic materials and AI-driven design in construction and infrastructure: what you can actually build with

Morphic materials are construction materials that change their own properties or geometry in service — self-healing concretes, shape-memory alloys, phase-change facades, printed components that move on cue. AI has compressed their discovery to months. It has not compressed the certification pathway that decides whether a designer may specify them, and that gap decides what you can build with.

Construction & Infrastructure · Future of AI & Visionary Thinking

Atomic Loops EngineeringIndustrial AI practiceReviewed by Atomic Loops Solutions ArchitecturePublished 14 August 202665 min read

Key takeaways

Morphic materials are materials that change their own properties or geometry in service — self-healing concretes, shape-memory alloys, phase-change facades and 4D-printed responsive components. All four families are real; none of them is universally specifiable, and the difference is certification rather than performance.
AI's genuine contribution sits at the two ends of the pathway: predicting properties to shrink the candidate search, and shaping geometry through topology optimisation and generative structural design. It contributes almost nothing to the middle, where standardised testing, third-party assessment and code recognition consume most of the elapsed time.
Google DeepMind's GNoME released 2.2 million predicted crystal structures, 380,000 of them predicted stable, and reported that external researchers had independently made 736 of them in a laboratory. The gap between prediction and physical realisation, then between realisation and a construction-products assessment, is the whole story.
The certification pathway has six gates — performance requirement, standardised test data, third-party assessment, approving authority, insurer and warranty, and finally code or standard recognition — and each has a different owner. A material fails at whichever gate was engaged last, which is almost always the insurer.
A material with no second supplier and no trained installer population is a research result, not a product. Supply and skills readiness decides whether a certified material survives value engineering, and it is the readiness question projects check last.

Abbreviations used on this page

CPR: Construction Products Regulation (the EU and GB product-marking regimes)
DoP: Declaration of Performance — the manufacturer's declared, testable product performance
ETA: European Technical Assessment — the route for products outside a harmonised standard
EAD: European Assessment Document — the assessment method an ETA is issued against
BBA: British Board of Agrément (issuer of UK agrément certificates)
ICC-ES: International Code Council Evaluation Service (US evaluation reports)
TRL: Technology readiness level — the 1–9 maturity scale used in research funding
SMA: Shape-memory alloy (nickel-titanium and copper-based families)
PCM: Phase-change material — stores and releases latent heat at a set temperature
GNN: Graph neural network — the model class used for crystal property prediction
GNoME: Graph Networks for Materials Exploration (Google DeepMind's discovery model)
MMC: Modern methods of construction — offsite and platform-based delivery

What morphic materials are — and why certification, not discovery, sets the pace

A definition, the four families that genuinely exist, and the two clocks whose difference decides what a designer may specify.

Morphic materials are construction materials that change their own properties or geometry in response to the conditions they meet, without anybody instructing them to. Four families are genuinely in service somewhere in the built environment today: self-healing concretes that seal their own cracks using bacterial or encapsulated healing agents; shape-memory alloys that recover a trained shape and are used for seismic damping and active prestress; phase-change materials that absorb and release latent heat inside facades and floor build-ups; and additively manufactured components engineered to deform in a designed way under heat, moisture or load — the family usually marketed as 4D printing.

The reason this page is not mostly about the science is that the science is not the constraint. Two clocks govern whether any of these materials reaches a permanent work. The discovery clock — finding a candidate, predicting its properties, making a specimen — has been compressed dramatically by machine learning: Google DeepMind's GNoME (opens in a new tab) released 2.2 million predicted crystal structures in a single publication, and Berkeley Lab's autonomous A-Lab (opens in a new tab) synthesised 41 novel compounds from 58 targets in seventeen days of unattended operation. The certification clock — standardised testing, third-party assessment, code recognition, insurer acceptance — has not been compressed at all, because none of its steps is a computation. It is a sequence of institutional decisions, each requiring physical evidence generated on production material over calendar time.

That asymmetry is the whole subject. A material that cannot be certified cannot be built with, however clever it is, and the certification pathway routinely takes longer than the discovery that produced the material. Everything downstream — which candidate is worth pursuing, what a pilot must be designed to produce, when to involve an insurer, whether a second supplier matters more than a better formulation — follows from taking that seriously rather than treating it as bureaucracy that will eventually get out of the way.

Buildable value released along the certification pathway

The curve is flat for a long time and then steep. A material at the specified-novel rung has produced no buildable value at all, however good its laboratory results, because nothing has been installed. Value inflects at the point where acceptance stops being an individual argument and becomes a standing route — which is why programmes measured in papers published rather than assessments obtained report progress without producing anything a designer can specify.

Buildable value released by stage

Stage 1 · Conventional — 46% of operators. Conventional is a project where every material in the permanent works comes from the palette existing codes already cover, and no AI touches the material specification.
Stage 2 · Specified-novel — 31% of operators. Specified-novel is where a material outside the standard palette has been named in a specification, but its acceptance rests on a supplier's data pack rather than on a clause anyone can point to.
Stage 3 · Piloted in works — 14% of operators. Piloted in works is a material installed in permanent works under a bespoke, project-specific acceptance, with monitoring in place and a conventional fallback agreed in advance.
Stage 4 · Code-recognised — 7% of operators. Code-recognised is a material with a standing acceptance route — a harmonised standard, a European Technical Assessment, an agrément certificate or an evaluation report — so approval no longer depends on an individual argument.
Stage 5 · Standard practice — 2% of operators. Standard practice is a material in the default palette: multiply sourced, competitively priced, installed by trained operatives, accepted by insurers and specified without special justification.

Curve shape: logistic, plotted from the stage data above. Distribution: Pathway stages consistent with the construction-products assessment routes published by EOTA.

The two clocks: from a predicted material to a specified one

Read the lanes as speeds, not as stages. The top lane runs in months and has been transformed by machine learning. The middle lane runs in years and has not. The bottom lane is what a designer is actually allowed to write into a specification, and it is fed only by the middle lane.

Data & feeds
AI / model
Human in the loop
Where value leaks
System-of-record action

The process, in words

The discovery lane runs on data and computation. Structure databases, project test records and in-service performance feed a property-prediction model, which ranks candidate materials by predicted stability and behaviour. This lane now produces candidates faster than any laboratory can make them, and far faster than any assessment body can evaluate them.
The certification lane runs on physical evidence and institutional decisions. A small number of candidates enter a standardised test programme, executed on production material to published methods; the resulting evidence pack goes to a third-party assessment body, which issues a document with a stated scope; code recognition and insurer acceptance follow, or do not. None of these steps is compressible by a better model.
The site lane is what a designer may actually write down. A candidate with no test programme is a research result, whatever its predicted properties. A material with a bespoke assessment can be specified on one project only. A material with a standing route and an accepted scope can be specified as standard — and only then does the discovery lane's output become buildable value.

Step-by-step insights

Materials and durability data — the asset record you already own: The public structure databases are the visible half of the data layer and the less useful half for a construction business. The half that differentiates is the in-service record: forty years of inspection reports, defect photographs, repair histories and the original mix designs and specifications those assets were built to. Almost none of it is joined up, which means the question a contractor or asset owner most wants answered — how has this material behaved in our exposure conditions, on our assets — cannot be asked of a model. Digitising and joining that record is unglamorous, it is not an AI project, and it is the highest-value data work most infrastructure owners are not doing.
Property prediction — what a graph neural network is actually good at: The models that produced the recent step change treat a crystal as a graph of atoms and bonds and learn to predict formation energy and stability from structure. That is a genuine capability and it is narrow in a way that matters for construction: it predicts thermodynamic stability of inorganic crystals, not fifty-year chloride resistance in a marine splash zone, not fire performance of a build-up, not creep under sustained load, and not what happens when a real operative places it on a wet Tuesday. Property prediction shrinks the search space at the top of the funnel. It does not produce evidence anybody downstream will accept.
The handover — why so few candidates enter testing: The narrowing between the discovery lane and the certification lane is severe and it is economic rather than scientific. A standardised test programme against the clauses an assessment needs is a multi-year, six-or-seven-figure commitment, and somebody has to fund it in the expectation of selling the resulting product. That funder is almost always a manufacturer, and manufacturers select for materials they can produce at volume with existing plant, not for materials with the best predicted properties. The filter between prediction and testing is a business-case filter, and no model changes its shape.
Third-party assessment — what you are actually buying: An assessment body issues a document that states what the product is, what it has been shown to do, and under what conditions. What the buyer is really acquiring is a transfer of accountability: a named organisation, competent and insured, has reviewed the evidence and put its name to a scope. That is why an assessment number moves an approving engineer when supplier data does not, and it is why the scope section is the important part of the document. A certificate used outside its scope transfers nothing, because the assessment body never assessed that use.
The bespoke route — useful, but it expires: The dashed line from assessment to project-specific acceptance is how most genuinely novel materials first reach permanent works: a bespoke assessment or an agreed departure, negotiated for one asset with one approving authority. It is a legitimate and often the only available route, and it is worth taking. What it is not is a precedent. Design the pilot so that its instrumentation, sampling and reporting produce evidence a third party could later assess against a published method, and the bespoke route becomes the first step of a standing one instead of a dead end with a nice case study attached.
Insurer acceptance — the gate that is engaged last and decides most: The insurance and warranty position appears at the far right of the certification lane because that is where projects usually discover it, not because it belongs there. Warranty providers and professional indemnity insurers set their appetite from claims history, and a material with an excellent certificate and no in-service record is, from their position, an unpriced risk. Engaging them at concept rather than at financial close costs nothing and changes the shape of the proposal: sometimes they will accept with an inspection regime, sometimes with a reduced scope, and occasionally they simply will not — which is far cheaper to learn in month two than in month twenty.

The five rungs in detail: Conventional to Standard practice

For each rung: what it looks like inside a real project, the diagnostic signals a reviewer can check in an afternoon, the anti-pattern that traps organisations there, and what leaving costs.

Each rung below describes a material's status inside a project rather than the material's quality. That distinction is the point of the ladder: the same self-healing admixture can sit at rung 2 on one scheme and rung 4 on another, because the ladder measures the strength of the acceptance argument, not the chemistry. The hallmarks describe observable conditions, the diagnostic signals are checks you can run against your own specifications and approval records this week, and the anti-pattern is the specific mistake most often made trying to leave that rung.

Read your organisation's rung as the level at which the acceptance capability actually sits, not the best example you can recall. One instrumented pilot on one asset does not make a business a rung-3 organisation if the other nine projects that year specified from the standard palette without asking. The ladder is about repeatable capability, which is why the assessment scores four dimensions separately and takes the lowest as the binding one.

Select a rung

Every rung's full detail is in the page source — the selector only changes which panel is visible, so nothing here depends on JavaScript to exist.

Stage 1

Conventional

46% of operators sit here

Conventional is a project where every material in the permanent works comes from the palette existing codes already cover, and no AI touches the material specification.

Stage 1 is not backwardness. It is the rational default of an industry where the cost of being wrong about a material is measured in decades and occasionally in lives. The palette is small on purpose: a handful of concrete strength and exposure classes, a handful of steel grades, a masonry range, a set of proprietary systems each carrying a certificate with a number on it. Every item in that palette has a test history, a supply chain, an installer population and an insurer who has already priced it.

The diagnostic tell at this rung is that nobody in the business can describe the route by which a new material would be accepted. Ask a design manager what it would take to get an unfamiliar admixture into a below-ground pour and the answer is either "the client won't have it" or a shrug. That is not obstruction. It is the honest answer of an organisation that has never had to find out, so the knowledge does not exist to be withheld.

Staying here is cheap and frequently correct. What makes it expensive is asymmetry. When a client, a funder, a planning condition or a carbon target eventually demands a solution the standard palette cannot deliver, a stage-1 organisation has to learn the entire acceptance route under programme pressure, on the project where it can least afford to. The learning happens either way; the only variable is whether it happens on a scheme with slack in it.

In practice

The admixture that never got past the design manager

A specialist subcontractor proposes a crystalline waterproofing admixture for a basement box, arguing it removes a tanking trade from the programme. The design manager asks who signs it off. The structural engineer says the client's technical assurance team; the client says building control; building control asks which standard the durability claim is made against. Nobody can answer inside two weeks, so the specification reverts to a conventional tanking system the week before the first pour. No one made a bad decision; there was simply no route for the proposal to travel down.

What it looks like

Specifications draw exclusively from standard strength classes, grades and proprietary systems that already hold certificates
Nobody in the business has taken a product through third-party assessment
Material choices are justified by precedent — "we used it on the last one"
AI, where it exists at all, sits in programme and cost tools rather than in material decisions

Diagnostic signals you can check this week

Ask who in the business has ever taken a product through third-party assessment. If the answer is nobody, you are here
Open the last three specifications and count the proprietary systems cited by certificate number. If none are, acceptance is being assumed rather than evidenced
Ask a design manager to describe, end to end, how an unfamiliar material would be approved. Vagueness is the signal, not reluctance
Check whether any material decision on the last project referenced measured in-service performance rather than precedent

Anti-pattern · Running an innovation programme instead of an acceptance route

The instinctive fix looks like innovation: a materials horizon scan, a supplier open day, a slide deck of promising technologies, possibly an innovation manager. None of it changes anything, because awareness was never the constraint. The constraint is that no one has walked a single material through the gate list and found out what each gate actually costs. Pick one material, one asset and one approving authority, and write down the real route before proposing anything to anybody.

What holds you here

Nobody in the organisation knows the route by which a new material would be accepted, so every proposal dies at the first question about who signs it off.

Highest-leverage next move

Pick one material and one asset and write the acceptance route down end to end — clause, test method, assessment body, approving authority, insurer — before proposing anything.

Cost of leaving

Effort: 3–6 months
Team: A materials-literate design manager and a procurement lead, part-time
Risk: Low — nothing enters permanent works at this rung, so the exposure is time
To next stage: 3–6 months

If this is you, the next step is

A two-week exercise: one material, one asset, the real gate list with owners and elapsed times.

Map your first acceptance route

Stage 2

Specified-novel

31% of operators sit here

Specified-novel is where a material outside the standard palette has been named in a specification, but its acceptance rests on a supplier's data pack rather than on a clause anyone can point to.

Stage 2 is where most construction innovation actually lives, and it is far more fragile than it looks from a marketing deck. The material is in the specification — sometimes as the base offer, more often as an "equal and approved" alternative — and it is supported by a folder of supplier test reports. What it is not supported by is a route. No clause in the governing code states what this material must satisfy, so acceptance depends on an individual deciding that the folder is convincing enough to put their name against.

That dependency surfaces as programme risk rather than technical risk. The tests may be excellent and the chemistry may be sound. The difficulty is that the approving authority, the client's technical assurance function and the insurer each form their own view, in sequence, at different points in the programme, and any one of them can decline at a moment when saying yes to the conventional alternative is cheaper than continuing the argument. Nobody has to be hostile for the outcome to be a reversion.

The characteristic failure of this rung is that the material survives the specification and dies at procurement. A contractor asked to price an unfamiliar material with a single supplier, an unverified lead time and no installer population will price that risk, and the priced risk usually exceeds the benefit being claimed. Value engineering then removes it, and the organisation records the episode as "the innovation did not stack up" when what actually happened is that no route existed for it to travel.

In practice

The equal-and-approved that was never equal

A lower-carbon cement replacement is named in a structural specification, backed by a supplier's own durability test pack. At technical query stage the approving engineer asks which standard the chloride-ingress claim is made against. The answer names a supplier method rather than a standardised one, so the engineer cannot map the result onto the exposure class in the design. The substitution is withdrawn at tender and the frame is priced conventionally. The material was probably fine; the evidence was addressed to the wrong audience.

What it looks like

The material appears in a specification, usually as an "equal and approved" alternative
Evidence is supplier-generated and mapped to no external clause or standardised method
No assessment body, insurer or warranty provider has yet seen the proposal
The material is routinely removed during value engineering

Diagnostic signals you can check this week

Search the last specification for "or equal and approved" against novel materials — the phrase transfers the acceptance argument to whoever prices the job
Ask which clause, in which standard, each performance claim is made against. An answer naming a supplier's own method places you here
Check whether the insurer or warranty provider has seen the proposal. At this rung they almost never have
Count how many of the last five novel materials survived value engineering. Below half is the signature of this rung

Anti-pattern · Strengthening the data pack instead of changing its author

When a proposal is questioned, the reflex is to ask the supplier for more test data. It rarely moves the decision, because the objection is not really about the numbers — it is about who carries the consequence if the numbers are wrong. A third-party assessment transfers a defined portion of that consequence to an assessment body whose business it is to carry it. That is why a certificate number changes conversations another two hundred pages of supplier testing does not.

What holds you here

Acceptance depends on one person's judgement of a supplier's data pack, so it can be reversed at any point up to and including procurement.

Highest-leverage next move

Convert one supplier data pack into the start of a third-party assessment — name the assessment body, the standardised test programme and the clause the result is intended to satisfy.

Cost of leaving

Effort: 6–12 months
Team: Design manager, a materials or technical specialist, procurement lead, and a named insurer contact
Risk: Medium — the exposure is programme and commercial rather than structural, because nothing has been built yet
To next stage: 9–18 months

If this is you, the next step is

We read the data pack the way the approver will, and list exactly what is missing.

Pressure-test one specification

Stage 3

Piloted in works

14% of operators sit here

Piloted in works is a material installed in permanent works under a bespoke, project-specific acceptance, with monitoring in place and a conventional fallback agreed in advance.

Stage 3 is the first rung where the material is load-bearing in the literal sense. It is in the ground, in the frame or on the facade, and a named person has signed for it. The acceptance is genuine but bespoke — a project-specific assessment, an agreed departure from standard, a bespoke test programme negotiated with the approving authority, or a client's own technical approval issued for this scheme only. It cost real money and real months to obtain, and it is worth exactly one project.

The engineering discipline here looks more like commissioning than design. Instrumentation goes in with the material. A conventionally built control section sits alongside it so the claim can be attributed later rather than asserted. An intervention plan states what happens if the expected performance does not appear in year three. Teams that skip the control section discover five years later that they cannot demonstrate the material did anything at all, because the same asset also received better drainage and a revised maintenance regime in the same window.

The trap is that the approval does not travel. The next project restarts the argument, frequently with a different approver who reaches a different view on identical evidence. Organisations can pilot the same material four times and remain at stage 3 throughout, because nothing they did converted a project decision into a standing route. The pilot is only worth what its evidence is worth to somebody who was not in the room.

In practice

The instrumented bay

An infrastructure owner accepts a crack-sealing admixture in one bay of a retaining structure under a project-specific assessment, with strain and moisture instrumentation cast in and an adjacent bay built conventionally as the control. Two years of readings show the treated bay behaving as claimed. The next framework project has a different approving engineer who has not seen the readings, was not party to the acceptance basis, and reasonably declines to accept an argument made for another asset. The evidence exists; the route does not.

What it looks like

The material is in permanent works under a named, project-specific acceptance
Instrumentation and a conventionally built control section were installed alongside it
An intervention plan exists for the case where the performance does not appear
Nothing about the acceptance generalises to the next project

Diagnostic signals you can check this week

Read the acceptance document and check whether it names the project. If it does, the route is project-specific and expires with the scheme
Check whether a conventionally built control section exists. Without one there is no attributable result, only a story
Ask what the intervention plan is if the material underperforms in year three. Silence means the pilot has no exit
Count how many separate projects have repeated the same approval argument for the same material

Anti-pattern · Calling the pilot a precedent

A project-specific acceptance gets cited in the next bid as though it were a certificate, sometimes with the project name quietly dropped. Approvers notice, and the credibility cost lands on the whole innovation portfolio rather than on the one claim. Either convert the pilot's output into evidence a third party can assess against a published method, or present it honestly as one instrumented installation with a stated scope. Both are defensible. Overstating it is not.

What holds you here

The acceptance is project-specific, so it does not travel to the next scheme and the argument restarts from zero with a different approver.

Highest-leverage next move

Design the pilot so its output is admissible evidence for a standing route — agree test methods, sampling and reporting with an assessment body before anything is installed.

Cost of leaving

Effort: 12–24 months
Team: Technical specialist, site engineering, a named monitoring owner, and a working relationship with the approving authority
Risk: Higher — the material is in permanent works, so the intervention plan has to be real and funded
To next stage: 2–5 years

If this is you, the next step is

Control section, instrumentation and acceptance basis agreed before the pour, not after it.

Design a defensible pilot

Stage 4

Code-recognised

7% of operators sit here

Code-recognised is a material with a standing acceptance route — a harmonised standard, a European Technical Assessment, an agrément certificate or an evaluation report — so approval no longer depends on an individual argument.

At stage 4 the argument has been made once, by an organisation whose job is to make it, and the result is a document with a number on it. That number changes the economics of every subsequent project. The designer cites it, the approver recognises it, and the conversation shifts from "should this be permitted at all" to "is it being used inside its assessed scope". That is a much shorter conversation, and it is the first point at which the material starts to look cheap.

The work to get here is unglamorous and long. It is a test programme scoped against the specific clauses the assessment will satisfy, executed to standardised methods in accredited laboratories, on production material rather than laboratory batches, with ageing and durability evidence that necessarily runs on calendar time. This is precisely the part of the pathway AI does not compress. A fifty-year service-life claim is defended with accelerated ageing protocols which themselves had to be validated, and validation is a committee process conducted by people who are paid to be sceptical.

Stage 4 is also where scope becomes the operative word. A certificate covers stated uses, exposure classes, substrates, thicknesses and details, and nothing else. Using a certified material outside its assessed scope is a stage-2 argument wearing a stage-4 badge, and it is the single most common way a code-recognised material ends up in an incident report. Read the scope section, not the summary page.

In practice

The certificate with a narrow scope

A facade system holds an assessment covering one substrate, one insulation thickness and one fixing pattern. A design team applies it to a different substrate on the grounds that the system "is certified". The approving body reads the scope section, notes that the assessed fire and wind-load evidence does not cover the substitution, and the design changes at a cost measured in months. Nothing was mis-sold. The certificate was read as a permission rather than as a boundary.

What it looks like

A standing assessment exists with a citable number and a written scope of use
Evidence was generated on production material, to standardised methods, in accredited laboratories
Approvers recognise the route without a project-specific argument being made
Supply is still thin — frequently a single manufacturer holds the assessment

Diagnostic signals you can check this week

Ask for the certificate number and read the scope section rather than the summary. Most scope breaches are visible on one page
Check whether the underlying evidence was generated on production material or on laboratory batches
Ask whether the insurer and the warranty provider have accepted the certificate explicitly, in writing, for this use
Count how many manufacturers hold a comparable assessment. One is a commercial dependency, not a market

Anti-pattern · Treating the certificate as the finish line

A certificate creates the right to specify. It does not create the ability to build. Programmes stop at stage 4, declare the innovation delivered, and then discover on site that there is one supplier with a twenty-week lead time, no trained installers within the region, and a warranty provider who wants an extra inspection regime nobody priced. The certificate was the expensive half of the work and the visible half; supply and skills are the half that decides whether the material is ever used twice.

What holds you here

The certificate exists but the supply base and the installer population do not, so the material stays expensive to build with at any scale.

Highest-leverage next move

Build the second source and the installer training before the first large specification, and get the insurer's acceptance in writing against the assessed scope.

Cost of leaving

Effort: 2–8 years
Team: A standards and technical lead, an accredited test partner, an assessment body, and a manufacturer prepared to fund the programme
Risk: Concentrated — the cost is front-loaded and largely irrecoverable if the assessed scope comes back narrower than the business case assumed
To next stage: 3–10 years

If this is you, the next step is

What the assessment will actually cover, what evidence it needs, what it costs and how long it takes.

Scope the certification programme

Stage 5

Standard practice

2% of operators sit here

Standard practice is a material in the default palette: multiply sourced, competitively priced, installed by trained operatives, accepted by insurers and specified without special justification.

Stage 5 is invisible from the outside, which is the point. Nobody writes a case study about specifying a C32/40 concrete. A material reaches standard practice when the questions stop: the designer does not justify it, the estimator has a rate for it, the site team has installed it before, the clerk of works knows what good looks like, and the warranty provider does not flag it in their review.

Getting here needs three things a certificate cannot provide. Competitive supply, so procurement stops treating the material as a risk item. A trained installer population, so workmanship stops being the dominant failure mode. And claims history, which is the slowest of the three. Insurers price what they have seen fail. A material with a fifteen-year in-service record is priced differently from a material with an immaculate certificate and no record, and no amount of laboratory evidence substitutes for elapsed years.

Standard practice is also revocable, and organisations forget this. Materials leave the default palette — sometimes abruptly, after a failure, an inquiry or a regulatory change — and the businesses that cope are the ones that kept the evidence base current rather than treating acceptance as permanent. The reassessment of external wall materials in the UK after 2017 is the reference case: build-ups that were routine one year were unspecifiable on tall residential buildings the next, and the difference between a week of re-specification and a year of it was whether alternates and evidence had been maintained.

In practice

The material that left the palette

A cladding build-up that was routine across a residential portfolio became unspecifiable on buildings above the height threshold following a change to the governing guidance. Organisations that held current evidence, a maintained alternates list and a warranty provider they spoke to regularly re-specified within weeks. Organisations that had treated the build-up as permanently accepted spent the better part of a year unpicking live schemes. The material had not changed. The acceptance had.

What it looks like

Three or more suppliers can price the material on the current programme
Trained installers exist outside the manufacturer's own crews
Insurers and warranty providers accept it without special terms or extra inspection regimes
Designers specify it without writing a justification for doing so

Diagnostic signals you can check this week

Count the suppliers who can quote the material on the current programme. Fewer than three is not standard practice, whatever the certificate says
Ask the estimator whether a published rate exists. A bespoke quote every time means the market has not formed
Ask whether site operatives have been trained by anybody other than the manufacturer
Ask what would happen to the specification if the governing guidance changed next month. Organisations at this rung can answer without a study

Anti-pattern · Assuming acceptance is permanent

Evidence packs go stale, standards are revised on a cycle, and insurers change their appetite after events nobody forecast. Materials in the default palette need a periodic evidence review exactly as much as novel ones do, and the review costs a fraction of a re-specification programme carried out under pressure. Put a named owner and a date on it, and keep a live alternate for every material whose acceptance rests on a single point of regulatory failure.

What holds you here

Acceptance is treated as permanent, so the evidence base ages quietly until a regulatory or insurance change makes a routine material unspecifiable overnight.

Highest-leverage next move

Put the default palette on a periodic evidence review with a named owner, and hold a live alternate for every material with a single point of regulatory failure.

Cost of leaving

Effort: Continuous
Team: A technical assurance function, supply-chain development, and a standing relationship with insurers and warranty providers
Risk: Low frequency, high consequence — the risk is regulatory and reputational rather than technical

If this is you, the next step is

Which of your standard materials would survive a change to the governing guidance next quarter.

Review your default palette

Where construction projects actually sit on the ladder

The distribution across the rungs, and why the step from specified-novel into permanent works is the largest single loss.

Most projects sit on the first two rungs, and the shape of the distribution is not a criticism of the industry. Construction is a business in which the consequence of a material failure is borne for fifty years by somebody who was not in the room when the decision was made, and the palette is conservative for that reason. What the distribution does show is where the losses happen: a large population of proposals reach a specification and a much smaller population reach permanent works, because the gap between those two states is an acceptance route that most organisations have never built.

Illustrative distribution of construction projects across the five rungs

Rung 1 is the mode and rung 2 is the plateau. The drop from specified-novel to piloted in works is the largest single transition loss on the ladder, and it is a commercial and procedural loss rather than a technical one. Figures are illustrative and synthesised from published construction-innovation and construction-products assessment research; they are not a survey.

Share of projects

46% — 1 · Conventional (the default palette)
31% — 2 · Specified-novel (the plateau)
14% — 3 · Piloted in works
7% — 4 · Code-recognised
2% — 5 · Standard practice

Source: Illustrative distribution, synthesised from published construction-innovation research

The plateau at rung 2 has a specific mechanism, and it is worth naming precisely because it is so often misdiagnosed as risk aversion. A material named in a specification without a standing acceptance route imposes a cost on whoever prices the job: an unfamiliar product, an unverified lead time, a single supplier, an unknown workmanship risk and an approving authority who has not yet said yes. Those costs are estimable, and an estimator will estimate them. The resulting number is usually larger than the benefit being claimed, so value engineering removes the material — not because anybody doubted it, but because the proposal asked a commercial party to carry an uncertainty that a certificate would otherwise have absorbed.

The same pattern appears in the broader productivity literature on the sector. The World Economic Forum's construction work (opens in a new tab) and McKinsey's engineering and construction practice (opens in a new tab) have both spent a decade documenting an industry whose technology adoption lags its capability, and the mechanism they describe is consistently structural rather than attitudinal: fragmented supply chains, project-by-project contracting and a liability model that punishes the first adopter. Certification is the specific instrument through which that structure expresses itself in material choices.

What is genuinely real today — and what is still a research result

Four families of morphic material are in service somewhere. For each: what actually exists, what AI contributed, the rung it typically occupies, and the constraint that binds it.

Four families of morphic material are genuinely installed in permanent works somewhere in the world today, and a fifth — AI-discovered chemistries entering construction products — is not yet, in any meaningful sense. Keeping those apart is the most useful editorial discipline on this subject, because the marketing material for all five reads identically. The test is not whether a laboratory result exists; it is whether a designer somewhere has specified it, an approving authority has accepted it, and an insurer has priced it.

Family	What genuinely exists in service	What AI actually contributed	Typical rung	The binding constraint
Self-healing concrete (bacterial and encapsulated agents)	Crack-sealing admixtures and repair mortars used on water-retaining structures, basements, tunnels and pavements, with commercial products on the market	Very little to date. The healing mechanisms came from microbiology and polymer chemistry; AI's contribution is mostly in predicting crack behaviour and prioritising inspection	2–3	Long-term durability evidence: a fifty-year sealing claim needs ageing data that only accrues on calendar time
Shape-memory alloys (seismic damping, active prestress)	Damping and restraint devices in seismic retrofit and bridge applications, plus SMA reinforcement in research-scale demonstrator structures	Alloy composition search and fatigue-life prediction; also structural optimisation of the devices themselves	2–3	Cost and cyclic fatigue qualification; devices are assessed individually rather than as a recognised material class
Phase-change materials (facades, floors, thermal mass)	Encapsulated PCM in wallboard, ceiling panels, ventilation systems and facade build-ups, in commercial products across several markets	Building-performance simulation and control optimisation — deciding where PCM helps and how the plant should respond to it	3–4	Fire performance of the encapsulation and long-term cycling stability, both assessed as part of the build-up rather than the material
4D-printed and programmed responsive components	Demonstrators and small-scale architectural elements; responsive shading and self-forming components largely outside permanent structural works	Geometry generation and the inverse design that determines a print pattern from a desired deformation	1–2	No assessment route exists for a component whose geometry is intended to change in service; standards assume geometry is fixed
AI-discovered inorganic chemistries	Nothing in construction permanent works. The output is predicted structures and, in a small number of cases, laboratory specimens	This is the family AI created: property prediction at a scale no laboratory programme could screen	1	The entire pathway. A predicted structure is separated from a specifiable product by synthesis, scale-up, standardised testing, assessment and supply
Adaptive and kinetic facade systems	Motorised and bimetallic shading, responsive glazing and electrochromic units on completed commercial buildings	Control optimisation and occupancy prediction; increasingly, generative facade geometry	3–4	Maintenance liability and whole-life cost rather than certification; the moving parts are the risk, not the material

The morphic material families, read against the buildability ladder. 'Typical rung' is where the family sits for a mainstream project today, not where its best single example has reached. The binding constraint column is the gate that actually stops it.

Two readings of that table matter more than the rest. The first is that AI's contribution has been small in exactly the families that are furthest along, and large in the family that is furthest behind — which should temper any expectation that better models will accelerate the ones already in service. Self-healing concrete did not need a graph neural network; it needed a decade of durability data and a manufacturer willing to fund an assessment.

Self-healing concrete is the family closest to routine use
Bacterial healing agents, developed from work at TU Delft (opens in a new tab) and commercialised by spin-outs including Basilisk (opens in a new tab), and capsule- and crystalline-based systems from established admixture manufacturers, are on the market as products with declared performance. The technical argument is largely settled; the remaining argument is about how a designer credits the healing in a durability calculation, which is a standards question that bodies such as fib (opens in a new tab) and RILEM (opens in a new tab) have technical committees working on.
Shape-memory alloys are a device story, not a material story
In practice SMA reaches structures as a damper, a restrainer or an active prestressing element — an engineered device with a certificate of its own — rather than as a material a designer specifies by grade. That framing matters, because a device can be assessed and type-tested far more readily than a new structural material class can be brought into a code. Earthquake engineering work at institutions including NIST's Engineering Laboratory (opens in a new tab) has kept the performance case moving, but the economics still confine SMA to applications where its cost is small against the consequence it prevents.
Phase-change materials are constrained by the build-up, not the chemistry
PCM in a facade or a ceiling is never assessed alone; it is assessed as part of a build-up whose fire performance, cycling stability and long-term encapsulation integrity are the real questions. Programmes such as the US Department of Energy's Building Technologies Office (opens in a new tab) have supported the performance evidence, and the products that have travelled furthest are the ones whose manufacturers took the whole build-up through assessment rather than certifying a material and leaving the designer to argue the assembly.
4D printing is real research and pre-specification practice
The inverse-design problem — computing a print pattern that produces a desired deformation under a stimulus — is genuinely solved for a growing set of cases, and groups including MIT's Self-Assembly Lab (opens in a new tab) have shown convincing demonstrators. What does not exist is any assessment framework for a component intended to change geometry in service. Every structural code assumes the geometry you designed is the geometry that will be there in year forty, and until an assessment route accommodates intentional change, these components stay outside permanent structural works.

The second reading is about honesty in the specification. Every family above has at least one supplier presenting laboratory-stage capability with the vocabulary of a mature product, and the fastest way to tell them apart is not to interrogate the science. Ask for the assessment number and read its scope; ask which standardised method each performance figure was produced under; ask how many other organisations hold a comparable assessment. Three questions, ten minutes, and they sort the table above without any materials expertise at all.

The certification gate: the six approvals a new material must clear

The centre of this page. Six gates, six different owners, six different kinds of evidence — and the one that gets engaged last is the one that most often says no.

A new construction material becomes buildable when six separate gates have each said yes, and the reason novel materials fail so reliably is that organisations engage them in the wrong order. The gates are: a performance requirement written in terms the codes recognise; standardised test data produced by accredited laboratories on production material; a third-party assessment with a written scope; acceptance by the approving authority for the specific project; acceptance by the insurer and warranty provider; and finally recognition in a code or standard, which is what turns the previous five into a repeatable route rather than a one-off.

Gate	What it actually asks	Who decides	Evidence it accepts	Typical elapsed time
1 · Performance requirement	What must this material do, expressed in the units the governing code uses — exposure class, fire class, design life, load case	The designer of record, with the client's technical assurance function	A written performance specification traceable to a code clause, not to a product	Weeks
2 · Standardised test data	Has the claimed performance been measured, by a method somebody else can repeat, on material from the production line	Accredited testing laboratories, to published methods	Test reports against named standardised methods, on production material, with sampling recorded	1–3 years, longer where ageing evidence is required
3 · Third-party assessment	Will a competent, insured body put its name to a stated scope of use for this product	An assessment body — an ETA issuer under an EAD, an agrément certifier, or an evaluation service	The full evidence pack, factory production control, and a scope negotiated line by line	1–3 years after the test programme completes
4 · Approving authority	Is this acceptable on this building or asset, under this regulatory regime	Building control, the building safety regulator, or the infrastructure owner's technical approval body	The assessment plus a project-specific justification showing use inside the assessed scope	Weeks to months per project
5 · Insurer and warranty	What is the claims exposure, and will we carry it at what price	Professional indemnity insurers, contractor's insurers, and structural warranty providers	Assessment, in-service record, an inspection regime, and sometimes a bond	Months — and it is the gate most often engaged last
6 · Code or standard recognition	Should this material class be written into the standards everyone works to	Standards committees at national and international level	A body of assessed use, research consensus and, usually, a decade of in-service data	5–15 years, and it is not a project activity

The six gates of the certification pathway, with owners, evidence and typical elapsed time. Elapsed times are ranges for a genuinely novel material, not for a variant of an existing certified product; they run in parallel only where noted.

The regimes differ by market but the shape does not. In Europe, products outside a harmonised standard reach the market through a European Technical Assessment issued against an EAD (opens in a new tab), by an assessment body notified through the NANDO database (opens in a new tab), with the harmonised standards themselves developed through CEN and CENELEC (opens in a new tab) and the structural design rules in the Eurocodes (opens in a new tab). In Great Britain the Construction Products Regulation regime (opens in a new tab) sits alongside the Approved Documents (opens in a new tab) and, for higher-risk buildings, the Building Safety Regulator (opens in a new tab); agrément certification through bodies such as the British Board of Agrément (opens in a new tab) is the customary route for products without a standard. In the United States the equivalent is an ICC-ES evaluation report (opens in a new tab) read against the International Code Council's model codes (opens in a new tab), with test methods from ASTM International (opens in a new tab) and design rules from bodies such as the American Concrete Institute (opens in a new tab).

The order that actually works

Write the performance requirement before you look at products
Gate 1 is cheap, fast and almost always skipped, because a supplier has usually already been in the room. Writing the requirement first — exposure class, design life, fire class, load case, and what failure would look like — does two things. It makes the candidate comparison honest, and it produces the document every later gate will ask for. A test programme scoped against a real requirement is half the size of one scoped against a wish list.
Engage the insurer at gate 1, not gate 5
The insurance and warranty position is the gate most often discovered last and the one most likely to be fatal. It costs nothing to put the proposal in front of a warranty provider such as NHBC (opens in a new tab) or your professional indemnity insurer at concept stage, and the answer reshapes the proposal usefully: acceptance with an inspection regime, acceptance with a reduced scope, or a clear no that saves two years.
Scope the test programme against the clauses, not the science
A test programme designed by researchers answers interesting questions. A test programme designed against the clauses the assessment will satisfy answers the questions an assessment body must be able to tick. They overlap far less than either party expects. Agree the method list with the assessment body before the first specimen is cast — this single step is the largest available compression of the certification clock.
Design the pilot to produce admissible evidence
If a project-specific acceptance is the only route available, take it — but instrument it so its output is admissible later. Standardised sampling, a conventionally built control section, agreed reporting and a named data owner turn a pilot into the first tranche of an assessment's evidence base rather than an anecdote with photographs.
Treat supply and skills as gate zero
Everything above assumes somebody can make the material at volume and somebody else can install it correctly. If neither is true, the certification programme buys a certificate for a product nobody can build with. Verify a second source and an installer training route before the test programme is funded, not after the certificate arrives.

Which novel material is actually worth pursuing here

Plot the certification route against supply and skills readiness. Three of the four quadrants are traps, and the two that look most promising in a supplier presentation are the two on the diagonal.

Available but unapprovable

Volume supply exists, often from another sector
No construction assessment route — nobody has funded one
Fix: fund the assessment, or accept it stays out of permanent works

Specifiable now

Standing assessment plus a real market
The only quadrant where value engineering does not remove it
Fix: nothing — specify it and record the in-service performance

Research result

Interesting, published, occasionally beautiful
Where almost all AI-discovered chemistries sit today
Fix: do not put it in a specification; sponsor the research honestly instead

Certified but unbuildable

A certificate exists; one supplier and no installers do
The most expensive trap, because the money is already spent
Fix: second source and installer training before the next specification

Supply & skills readiness — top: Multiple suppliers, trained operatives, bottom: Single supplier, no trained installers

Certification route — left: No route — bespoke argument each time, right: Standing route — assessment with a scope

The bottom-right quadrant is the one worth dwelling on, because it is where well-run innovation programmes end up. A business funds a certification programme, gets a certificate, celebrates, and then finds the material is available from one manufacturer on a twenty-week lead with no trained installer population inside the region. The certificate is real and the material is unbuildable at any useful scale. Certification and supply are not sequential activities where one earns the other; they are parallel programmes that have to finish at roughly the same time.

Where AI actually helps — and where it changes nothing

Five capabilities with genuine evidence behind them, what each changes, what it does not change, and where its output has to land to be worth anything.

AI's real contribution to construction materials sits at the two ends of the pathway and almost nowhere in the middle. At the front it shrinks the search: property prediction over crystal structures reduced a screening problem that was previously bounded by laboratory throughput to one bounded by compute. At the back it shapes geometry: topology optimisation and generative structural design produce members and assemblies that use materially less material for the same performance. In between — testing, assessment, code recognition — it contributes essentially nothing, because those steps are institutional rather than computational.

Capability	What it genuinely changes	What it does not change	Where the output must land
Property prediction over crystal structures (GNN)	Screens candidate inorganic materials at a scale no laboratory programme could reach; ranks by predicted stability	Produces no evidence any assessment body accepts; predicts thermodynamic stability, not fifty-year durability in a marine splash zone	A funded synthesis and standardised test programme — otherwise it is a paper
Topology optimisation and generative structural design	Reduces material in members and assemblies for the same structural performance; explores far more of the design space than a human sequence of iterations	Does not make the result buildable, fabricable or inspectable, and does not satisfy code clauses on detailing, cover or robustness	The analysis model and the fabrication package, with buildability constraints inside the objective function
Surrogate models for durability and service life	Turns expensive physics simulations into fast approximations, so design teams can explore exposure scenarios interactively	Does not substitute for physical test evidence, and must never enter a compliance claim as though it were measurement	The design record — clearly labelled as model-derived, versioned, and validated against physical tests
Computer vision for in-service condition and defect detection	Converts inspection imagery into structured condition data at a rate manual inspection cannot match; builds the durability record materials work depends on	Does not assess residual capacity or decide interventions; it detects and classifies, and the engineering judgement remains	The asset register and the maintenance system, joined back to the original specification and mix
Language models over standards and evidence packs	Finds the applicable clauses, cross-references between codes and assessments, and drafts the traceability matrix nobody wants to build by hand	Does not interpret a clause authoritatively and must not be a source in a compliance argument; the citation must always be checked against the standard	A referenced traceability matrix that a human technical lead signs, with every clause verified against the published text

AI capabilities in construction materials and design, read honestly. 'Where the output must land' is the test: a capability whose output stops at a report has not changed anything a designer can build with.

The generative-design row deserves elaboration because it is the capability most likely to be oversold on a construction project. Topology optimisation is a mature field with a substantial research literature — machine-learning approaches to it are documented in work such as Neural networks for topology optimization (opens in a new tab) and TopologyGAN (opens in a new tab) — and the results are genuine: less material for the same performance, obtained faster than iterative manual design. The catch is that the objective function almost never includes the things that decide whether a structure gets built. Formwork complexity, reinforcement congestion, transport envelope, connection standardisation, tolerance, and whether an inspector can see the critical detail after installation are all buildability constraints, and an optimiser that ignores them will confidently produce a member that is thirty per cent lighter and forty per cent more expensive to make.

To close the gap between the rates of computational screening and experimental realization of novel materials, we introduce the A-Lab, an autonomous laboratory for the solid-state synthesis of inorganic powders.
An autonomous laboratory for the accelerated synthesis of inorganic materials, Nature (2023)

That opening sentence is the most honest framing of the field available, and it is written by the people doing the work. The gap it names — between how fast candidates can be screened and how fast they can be physically realised — is exactly the gap this page is about, one stage earlier. The A-Lab addresses the laboratory half of it; the Materials Genome Initiative (opens in a new tab) has been addressing the data and infrastructure half since 2011, and standards-side work at NIST's Material Measurement Laboratory (opens in a new tab) addresses the measurement half. None of that work touches the construction-specific half, which is that even a fully realised, well-characterised material still has to clear six institutional gates before a designer may write it into a specification.

There is one place where AI does genuinely compress the certification clock, and it is worth naming because it is unglamorous and available now. The evidence pack for an assessment is a traceability problem: a set of performance claims, each mapped to a test method, each mapped to a clause, each mapped to a report and a sampling record. Assembling and maintaining that matrix consumes months of specialist time on every programme, and it is exactly the kind of structured cross-referencing that language models handle well under human verification. That is not a discovery story and nobody will present it at a conference, but it takes real weeks out of gates 2 and 3.

What this looks like in public

Three publicly reported programmes, read against the ladder. None is an Atomic Loops engagement — each links to the programme's own published material.

Three publicly reported programmes show the two clocks running at different speeds, and each one lands on a different rung. A materials-discovery programme at the top of the funnel; a printed structural component that reached permanent works through an entirely bespoke certification route; and a self-healing concrete technology that went from a university laboratory to commercial products with declared performance. Read together they describe the pathway more usefully than any of them does alone.

Three programmes read against the buildability ladder

Outcomes as reported by the programmes themselves; verify figures against the linked source before reusing them, as we have not independently audited them. No operator photography exists in our image library for any of these programmes, so each card carries an illustrative industry scene — the images depict the type of work described, not the programmes themselves, and imply no endorsement.

Industry scene: an infrastructure design review against modelled performance data — illustrative, not a photograph of the programme described

Google DeepMind (GNoME)AI research programme · materials discovery at scale11

Challenge: Discovering stable inorganic crystals has historically been bounded by laboratory throughput, with months of experimentation behind each new stable compound and roughly 48,000 computationally stable crystals known from prior databases and computational approaches.
Approach: GNoME applied graph neural networks to predict the structure and stability of candidate crystals at scale, with predictions checked using density functional theory and the results released publicly to the Materials Project database.
Reported outcome: Google DeepMind reported the discovery of 2.2 million new crystals, 380,000 of them predicted to be the most stable and released as candidates for synthesis, and stated that external researchers had independently created 736 of GNoME's materials in the laboratory.; What it shows about the curveThis is rung 1 for construction, and deliberately so. The programme transformed the discovery clock and touched none of the certification clock — which is exactly why a construction business should read it as a supply of candidates rather than as a supply of materials.

Google DeepMind — Millions of new materials discovered with deep learning (opens in a new tab)

Industry scene: gantry-mounted additive manufacturing of a structural element on a construction site — illustrative, not a photograph of the programme described

MX3D (Amsterdam printed steel bridge)Additive manufacturing · pedestrian bridge in permanent works13

Challenge: A wire-arc additively manufactured stainless steel footbridge had no governing standard: existing structural codes assume material produced by conventional processes with established property distributions, and none of them covers a printed component whose properties vary with print path.
Approach: MX3D and its partners built a bespoke certification route — extensive physical testing of printed material and of the structure itself, with academic partners including Imperial College London — and instrumented the completed bridge with a sensor network so its in-service behaviour could be monitored as a data-centric engineering programme with the Alan Turing Institute.
Reported outcome: The bridge was installed over a canal in central Amsterdam and opened to the public, with the sensor programme and testing evidence published by the partners as an ongoing research and monitoring exercise.; What it shows about the curveA textbook rung 3. The acceptance was real, hard-won and specific to one structure, and its lasting value is the published evidence base rather than the crossing itself. The bespoke route is available to anybody willing to fund it — and it does not generalise unless the evidence is designed to.

MX3D — company site; monitoring programme reported by the Alan Turing Institute (opens in a new tab)

Industry scene: an illustrative render of autonomous crack sealing in a concrete element — not a photograph of the programme described

Basilisk (TU Delft self-healing concrete)University spin-out · self-healing concrete products14

Challenge: Cracking in concrete is the primary route by which water and chlorides reach reinforcement, and conventional repair is reactive, disruptive and expensive on water-retaining structures, basements and tunnels where access is hard.
Approach: Research at TU Delft developed bacterial healing agents that precipitate limestone when water reaches a crack, and the spin-out commercialised them as admixtures, repair mortars and liquid repair systems supplied as products with declared performance rather than as a research technology.
Reported outcome: Basilisk publishes a commercial product range and project references for self-healing concrete and repair systems across Dutch and international infrastructure and building work.; What it shows about the curveThe clearest rung 4 example in this family, and it took roughly two decades from the underlying research to a supplied product line. The technical breakthrough was the shortest part of the story; productisation, declared performance and route to market were the long part.

Basilisk — product and project material; underlying research at TU Delft (opens in a new tab)

Set the three side by side and the ratio is the argument. The discovery programme produced 2.2 million candidates in one publication. The printed bridge produced one crossing after years of bespoke testing with partners including Imperial College London (opens in a new tab) and a monitoring programme with the Alan Turing Institute (opens in a new tab). The self-healing concrete programme took about two decades from laboratory to product line. Nothing in the first programme shortens the second or third, and any plan that assumes otherwise is planning against the wrong clock.

The specification stack, layer by layer

What has to exist for a morphic material to be specifiable — six layers, each annotated with the rung that first requires it.

Getting a morphic material into permanent works requires six layers of capability, and the order in which a business builds them determines whether its innovation compounds or evaporates. The stack below is deliberately unfashionable: nothing in it is specific to a vendor or a modelling technique, and every layer is defined by what it must guarantee rather than by what product provides it. The layers that organisations skip are almost always the fourth and the fifth, which is why the sixth so often arrives as a surprise.

The layers a specifiable material depends on

Each layer is annotated with the rung that first requires it. A business trying to reach rung 3 without the physical evidence and in-service assurance layers is building a rung-2 proposal with more slides.

Materials data
Stage 1+
- Public structure and property databasesMaterials Project, ICSD-derived sets, published test data
- Project and supplier test recordsMix designs, certificates, batch records, sampling
- In-service performance historyInspection reports and defect records, joined back to the original specification
→
Prediction & design tools
Stage 2+
- Property-prediction modelsCandidate screening against stated performance requirements
- Topology optimisation and generative designGeometry and member sizing, with buildability in the objective
- Durability and service-life surrogatesFast approximations, clearly labelled as model-derived
→
Physical evidence
Stage 3+
- Standardised test programmeAccredited laboratories, published methods, production material
- Ageing and durability evidenceAccelerated protocols plus real elapsed time; not compressible
- Instrumented trial with a control sectionConventionally built comparator, so the claim is attributable
→
Assessment & certification
Stage 4+
- Assessment body engagementEAD/ETA, agrément certification or evaluation report
- Declaration of performance and markingFactory production control and the declared, testable characteristics
- Code or standard recognitionCommittee work; measured in years and not a project activity
→
Supply & skills
Stage 4+
- Qualified second sourceTwo or more manufacturers with evidenced production capacity
- Installer training and workmanship controlTrained operatives outside the manufacturer's own crews
- Insurer and warranty positionWritten acceptance against the assessed scope, obtained early
→
In-service assurance
Stage 3+
- Monitoring planWhat is measured, by whom, at what interval, for how long
- Intervention regimeWhat happens when the expected performance does not appear
- Feedback into the evidence baseIn-service data returning to the materials data layer

Pipeline described

Materials data (stage 1+) — Public structure and property databases: Materials Project, ICSD-derived sets, published test data; Project and supplier test records: Mix designs, certificates, batch records, sampling; In-service performance history: Inspection reports and defect records, joined back to the original specification
Prediction & design tools (stage 2+) — Property-prediction models: Candidate screening against stated performance requirements; Topology optimisation and generative design: Geometry and member sizing, with buildability in the objective; Durability and service-life surrogates: Fast approximations, clearly labelled as model-derived
Physical evidence (stage 3+) — Standardised test programme: Accredited laboratories, published methods, production material; Ageing and durability evidence: Accelerated protocols plus real elapsed time; not compressible; Instrumented trial with a control section: Conventionally built comparator, so the claim is attributable
Assessment & certification (stage 4+) — Assessment body engagement: EAD/ETA, agrément certification or evaluation report; Declaration of performance and marking: Factory production control and the declared, testable characteristics; Code or standard recognition: Committee work; measured in years and not a project activity
Supply & skills (stage 4+) — Qualified second source: Two or more manufacturers with evidenced production capacity; Installer training and workmanship control: Trained operatives outside the manufacturer's own crews; Insurer and warranty position: Written acceptance against the assessed scope, obtained early
In-service assurance (stage 3+) — Monitoring plan: What is measured, by whom, at what interval, for how long; Intervention regime: What happens when the expected performance does not appear; Feedback into the evidence base: In-service data returning to the materials data layer

Step-by-step insights

Materials data — the in-service record is the differentiated asset: Every business can reach the same public databases, so no competitive position exists there. The differentiated asset is the in-service record: decades of inspection reports, defect photographs, repair histories and the mix designs and specifications the assets were built to. Joined up, that record answers the question public data never can — how this material behaves in our exposure conditions on our assets — and it turns a generic candidate list into a shortlist somebody can defend. It is also the layer that requires no AI to build and no AI to be valuable, which is why it keeps losing funding contests to work that demonstrates better.
Prediction and design tools — the failure is at the boundary, not in the model: Optimisation and generative tools fail in construction almost entirely at their boundaries. Upstream, they are given an objective function that omits formwork, reinforcement congestion, transport envelopes and inspectability. Downstream, their output is redrawn conventionally because the analysis model and the fabrication package cannot consume it. Fixing either boundary is worth more than any improvement to the optimiser: an average optimiser whose result reaches the fabrication data changes more buildings than an excellent one whose result reaches a competition board.
Physical evidence — the layer with an irreducible clock: This is the layer that does not compress. Accelerated ageing protocols exist and are useful, but they were themselves validated against real elapsed time, and an assessment body will read a fifty-year claim supported entirely by accelerated data with appropriate scepticism. The practical consequence for planning is blunt: if the programme needs durability evidence, start generating it in year one on a specimen set you keep, rather than in year three when the assessment body asks for it. The specimens are cheap; the years are not.
Assessment and certification — buy the scope, not the certificate: The negotiation with an assessment body is really a negotiation about scope, and scope is where the business case lives. A narrow scope — one substrate, one exposure class, one detail — can be obtained faster and cheaper, and may be worthless if the intended application sits outside it. A wide scope costs more evidence and more time. Deciding which scope the business actually needs, before the test programme is designed, is the single most valuable hour anybody spends on the whole pathway, and it is routinely skipped in favour of getting started.
Supply and skills — the layer that decides whether it happens twice: A certified material with one supplier and no trained installers survives exactly as long as its champion stays in post. Two things make it durable: a second source, which removes the commercial dependency procurement is pricing, and an installer population trained by somebody other than the manufacturer, which removes the workmanship risk the insurer is pricing. Both take about as long as the certification programme, and both should be run in parallel with it rather than started when the certificate arrives.
In-service assurance — the loop that makes the next one cheaper: The monitoring plan looks like a cost on the first project and is the entire asset on the fourth. Every instrumented installation with a conventional control section produces evidence that reduces what the next assessment has to generate from scratch, and evidence generated in real assets is precisely what insurers and standards committees weight most heavily. Feeding it back into the materials data layer closes the loop, and closing that loop is what separates a business with an innovation programme from a business with an innovation portfolio.

The layer most often skipped is in-service assurance, and it is the one that determines whether the second project is cheaper than the first. Without a monitoring plan and a control section, a completed installation generates a photograph and a testimonial; with them, it generates evidence an assessment body can weigh. The cost difference is instrumentation and a data owner. The value difference is whether the organisation is climbing the ladder or repeating rung 3 indefinitely.

A 90-day plan: get one adaptive material into permanent works on one asset

The rung 2 → 3 transition made concrete on a single construction problem — a crack-sealing self-healing admixture in a below-ground water-retaining structure. Contains no materials research at all.

Ninety days is enough to get one adaptive material accepted into permanent works on one asset, provided the scope is exactly that: one material, one asset, one approving authority. To make it concrete, the plan below runs the transition on a problem almost every infrastructure owner and civils contractor recognises — early-age and service cracking in a below-ground water-retaining structure, where the conventional answer is a tanking system plus reactive injection repair, and where a self-healing admixture with declared performance is a genuinely available product rather than a research proposition.

The quarter contains no materials development. The chemistry exists, products exist, and suppliers publish declared performance. What does not exist, in most organisations, is the acceptance route — and that is what these ninety days build. If any phase overruns its window, narrow the scope further: one pour, one bay, one exposure condition. Extending the plan is how a ninety-day route becomes a two-year programme.

Rung 2 → rung 3 on one water-retaining structure, in one quarter

One asset, one material, one named approver, one named data owner. Every phase produces a document the next phase needs; nothing here is a study.

Days 1–20
Write the requirement and open every gate
Write the performance requirement in code terms — exposure class, crack-width limit, design life, watertightness class — with no product named. Then, in the same fortnight, put the proposal in front of all six gate owners at once rather than in sequence: the designer of record, the client's technical assurance lead, building control or the owner's technical approval body, the professional indemnity insurer, the warranty provider and procurement. The purpose is not agreement; it is to surface the fatal objection while it is still cheap.
A code-referenced requirement and a written objection list from six named people
Days 21–45
Assemble and gap-test the evidence pack
Collect the candidate products' declarations of performance and any assessment certificates, and build the traceability matrix: each performance claim mapped to the standardised method that produced it and the clause it is intended to satisfy. Mark every claim supported only by a supplier method. That marked list is the gap, and it is usually shorter than feared — most of the missing evidence is about the specific exposure condition rather than about the material.
A traceability matrix with the evidence gap explicitly marked
Days 46–70
Agree the acceptance basis and design the pilot
Take the marked gap back to the approving authority and agree, in writing, what would close it for this asset: usually a bounded application, a defined monitoring regime and a conventional fallback specified in the same package. Design the installation to produce admissible evidence — standardised sampling, a conventionally built control section on the same structure, agreed reporting intervals, and a named data owner who is not the supplier.
A written acceptance basis, a pilot design with a control section, a named data owner
Days 71–90
Install the bounded pour and register the result
Install on the agreed bounded scope with the workmanship control regime the insurer asked for, take the baseline readings before backfill, and register the installation in the organisation's evidence base — asset, mix, batch records, control section, monitoring plan, reporting dates and owner. Registration is the phase most often skipped and the one that decides whether project two starts from this project's evidence or from zero.
Material in permanent works, baseline readings taken, evidence registered and owned

Why the order matters

Six gates in parallel, not in sequence
The instinct is to build a strong proposal and then take it to the approvers. That sequence guarantees the fatal objection arrives after the effort has been spent. Opening all six gates in the first three weeks feels premature and is the single largest time saving available on this plan, because the objection you cannot answer is worth knowing on day fifteen.
The control section before the claim
A conventionally built section on the same structure, in the same conditions, poured in the same week, is what converts a completed installation into evidence. Without it, five years of good readings are indistinguishable from five years of good luck, better drainage or a revised maintenance regime — and an assessment body will say so.
The fallback is what unlocks the approval
Specifying the conventional solution as an explicit fallback in the same package looks like a lack of confidence. It is actually the political key: an approver and an insurer will accept a bounded novelty they can revert, and will decline the same proposal presented as irreversible. The fallback costs a paragraph and buys the acceptance.
Registration before celebration
The installation is worth one project until it is registered — asset, batch, control section, monitoring plan, owner and dates in a place the next project's design manager will find. Registration takes an afternoon and it is the difference between climbing the ladder and repeating rung 3 on a new scheme every eighteen months.

Procurement and supply: the readiness nobody checks until it is too late

A material with no second source and no trained installer is a research result, whatever its certificate says. The readiness questions, the evidence to demand, and a checklist you can run before the next specification.

A material with no supply chain and no installer training is a research result, whatever its certificate says, and this is the readiness dimension that projects check last. The pattern is consistent enough to be predictable: technical due diligence is thorough, certification due diligence is at least attempted, and supply due diligence consists of asking the manufacturer whether they can supply it. They always say yes. The questions below are the ones an estimator and a site manager will ask later, at a point in the programme where the honest answers are expensive.

Readiness question	Evidence to demand	Rung it unlocks	Who owns it
Can more than one manufacturer supply this?	Two named manufacturers with comparable assessments, or a written plan and date for the second source	4	Procurement, with technical
What is the real lead time at our volume?	A quoted lead time at programme volume, plus one reference project at similar scale, plus the production constraint that sets it	3	Procurement
Who installs it, and who trained them?	Named subcontractors with recorded operative training, delivered by somebody other than the manufacturer's own crew	4	Site management, with the supply chain team
How is workmanship verified on site?	A written inspection and test plan with hold points, sampling frequency and an independent verification step	3	Quality and site engineering
What is the warranty and insurance position?	Written acceptance from the warranty provider and the professional indemnity insurer, against the assessed scope, for this use	4	Commercial, with technical assurance
What happens if it fails in year seven?	An intervention plan with an access strategy, a repair method and a costed provision	3	The asset owner
What is the conventional fallback and its trigger?	The fallback specified in the same package, with the decision point and the person who makes it named	3	The designer of record

Supply and skills readiness: what to ask, what evidence answers it, and which rung the answer unlocks. Every row has an owner inside the business, and the owner is rarely the person who proposed the material.

The second-source question is the one that changes outcomes most and gets asked least. A single-manufacturer material is a commercial dependency, and every commercial party downstream prices dependencies: the contractor prices supply risk, the client's commercial team prices the absence of competitive tension, and the insurer prices the concentration. Those three prices, added together, routinely exceed the technical benefit being claimed — which is the actual reason novel materials disappear during value engineering. It is not scepticism about the chemistry. It is arithmetic.

Metric	Rung 2	Rung 3	Rung 4	Where to read it
Acceptance basis	Supplier data pack	Project-specific approval	Assessment cited by number	The technical submission and its approval record
Evidence provenance	Supplier method	Standardised method, lab batch	Standardised method, production material	Test reports and sampling records in the evidence pack
Insurer engagement point	Not engaged	At detailed design	Before specification, in writing	Correspondence dates against the design programme
Suppliers who can price it	1	1–2	2+	Tender returns for the material line
Control section installed	None	Yes, on the same asset	Yes, plus registered in the evidence base	The as-built record and the monitoring plan

Verification metrics for the rung transitions. All five are readable from documents the business already produces, so none of them depends on a self-report.

Specification readiness checklist

Run this before a novel material enters a specification. If you cannot tick all eight, the material is not ready to be specified — which is different from the material not being good. Tick as you go; this list works without JavaScript.

The performance requirement is written in code terms, with no product namedExposure class, design life, fire class, load case — the language every later gate uses.
Every performance claim is mapped to a standardised test methodClaims resting on a supplier's own method are marked as gaps, not treated as evidence.
A third-party assessment exists and its scope has been readThe scope section, not the summary page. Most breaches are visible in one page.
The approving authority has stated, in writing, what would make this acceptable
The insurer and warranty provider have given a written positionAgainst the assessed scope, for this specific use, before financial close.
A second source exists, or a dated plan for one doesOne manufacturer is a dependency that procurement will price out.
Trained installers exist outside the manufacturer's own crewsWith recorded competence and an inspection and test plan with hold points.
A conventional fallback is specified in the same package, with a named decision makerThis is what makes a bounded novelty acceptable to an approver.

0 of 8 ticked

Nothing ticked — start with the requirement, not the product

A blank list usually means a supplier arrived before a requirement existed, which is the normal way these proposals start and the reason so many die. Write the performance requirement in code terms first, with no product named. Everything else on this list becomes answerable once that document exists, and several items answer themselves.

Write the requirement with us

Failure modes that turn a certified material into an incident

Five recurring failures, none of which is the material being bad. Each with its likelihood, its impact and the cheap preventive measure.

Five failure modes account for almost every case where a novel material caused a problem on a construction project, and in none of them was the material the cause. They are failures of scope, of evidence provenance, of supply, of buildability and of institutional memory. All five are cheap to prevent and expensive to discover, and all five are visible in documents the project already produces — which means a reviewer with an afternoon and the right questions can find them before the concrete does.

Likelihood: highImpact: high

The certificate used outside its assessed scope

A material holds a genuine assessment for one substrate, one exposure class or one detail, and is applied to another on the reasoning that it is certified. The assessment body never assessed that use, so the accountability transfer everyone assumed had happened did not. This is the most common route from a well-intentioned specification to an incident report.

PreventionRead the scope section against the actual application before the material enters the specification, and record the comparison in the design record.

Likelihood: mediumImpact: high

Model-derived figures entering the evidence pack as measurement

A surrogate model's durability or service-life output is quoted in a technical submission without being distinguished from physical test data. It survives design review and fails technical audit, and when it fails it discredits the genuine evidence sitting next to it.

PreventionLabel every model-derived figure in issued documents, exclude it from compliance claims by default, and state its validation basis where it is used at all.

Likelihood: highImpact: medium

The single-supplier dependency discovered at tender

A material with one manufacturer reaches tender, where the contractor prices supply risk, the client prices the absence of competitive tension and the insurer prices the concentration. The combined price exceeds the benefit and the material is value-engineered out, taking the certification investment with it.

PreventionEstablish a second source, or a dated plan for one, before the material enters a specification — and tell procurement about it early.

Likelihood: mediumImpact: medium

Optimised geometry nobody can build or inspect

A topology-optimised member is thirty per cent lighter and cannot be formed, reinforced, transported or inspected economically. The saving is real in the model and negative on site, and the episode discredits generative design internally for years afterwards.

PreventionPut formwork, reinforcement congestion, transport envelope and inspectability into the objective function, and review early output with a fabricator rather than a designer.

Likelihood: highImpact: medium

The pilot whose evidence was never registered

A material is installed successfully under a project-specific acceptance and nothing is recorded in a place the next project will find. Two years later the same argument is made from scratch, often by a different team who do not know the first installation exists. The organisation repeats rung 3 indefinitely and reads its own persistence as innovation.

PreventionRegister every installation — asset, batch, control section, monitoring plan, owner and dates — as the closing task of the project, not as an aspiration.

The through-line is that four of the five are documentation failures rather than engineering ones, and the fifth is a supply-chain failure. That is characteristic of this subject: the materials science is done by people who are good at it, and the losses accumulate in the handovers between the laboratory, the design record, the approval file, the tender and the site. Anybody looking for the highest-return intervention in a construction materials innovation programme should look at those handovers before looking at the models.

Glossary

Hover a term for its definition — or expand the map full screen. The full definitions are written out below.

Morphic material: A material that changes its own properties or geometry in response to conditions in service — heat, moisture, load, cracking — without external instruction. Used here as the umbrella term for self-healing, shape-memory, phase-change and programmed-responsive families.
Self-healing concrete: Concrete containing an agent that seals cracks autonomously, most commonly bacterial spores that precipitate limestone when water reaches a crack, or encapsulated polymers released by cracking. Supplied as admixtures, repair mortars and liquid repair systems with declared performance.
Shape-memory alloy (SMA): An alloy, typically nickel-titanium or copper-based, that recovers a trained shape on heating or unloading. In construction it reaches structures mainly as engineered devices for seismic damping, restraint and active prestress rather than as a specified material grade.
Phase-change material (PCM): A material that absorbs and releases latent heat at a set temperature, used encapsulated in wallboard, ceiling panels and facade build-ups to add thermal mass without weight. Assessed as part of the build-up rather than as a material on its own.
4D printing: Additive manufacturing of a component engineered to change shape in a designed way under a stimulus such as heat, moisture or load. The design problem is inverse: computing the print pattern that produces the intended deformation.
Topology optimisation: A computational method that distributes material within a design domain to meet a performance objective under constraints, producing structurally efficient and often organic geometry. Its usefulness in construction depends entirely on whether buildability constraints are inside the objective function.
Graph neural network (GNN): A model class that treats a crystal as a graph of atoms and bonds and learns to predict properties such as formation energy and stability from structure. The technique behind the recent step change in materials-discovery throughput.
European Technical Assessment (ETA): The route to market for a construction product not covered by a harmonised standard: an assessment body issues a documented, scoped assessment against a European Assessment Document, enabling a declaration of performance and product marking.
Agrément certificate: A third-party certificate covering products without a governing standard, issued in the UK by bodies such as the British Board of Agrément. It states an assessed scope of use and is the customary route by which a novel product becomes routinely acceptable to approving authorities.
Declaration of Performance (DoP): The manufacturer's formal statement of a product's essential characteristics and declared performance, produced under a construction products regime and backed by factory production control. It is the document a specifier should be citing rather than a brochure.
Assessed scope: The stated boundary of a certificate: the uses, substrates, exposure classes, thicknesses and details the assessment covers. Use outside the assessed scope transfers no accountability, because the assessment body never evaluated it.
Control section: A conventionally built portion of the same asset, poured or installed in the same conditions, deliberately excluded from the novel material so its performance can later be attributed rather than asserted. The construction equivalent of a holdout.

Frequently asked questions

The questions design managers, technical leads and asset owners ask most often about putting an adaptive material into permanent works.

What are morphic materials in construction?

Morphic materials are materials that change their own properties or geometry in service in response to conditions, without external instruction. Four families are genuinely in use: self-healing concretes that seal their own cracks, shape-memory alloys used in seismic damping and active prestress, phase-change materials that add latent-heat storage to facades and floors, and printed components engineered to deform in a designed way. The term is an umbrella rather than a technical classification, and the families differ far more in their certification status than in their science.

Is self-healing concrete actually used on real projects?

Yes. Bacterial and encapsulated healing agents are sold as admixtures, repair mortars and liquid repair systems with declared performance, and are used on water-retaining structures, basements, tunnels and pavements. The remaining technical debate is not whether the healing occurs but how a designer may credit it in a durability calculation — a standards question that technical committees at bodies including fib and RILEM are working on. Until that credit exists in a code, the material typically improves durability without reducing the conventional provision beside it.

How long does it take to certify a new construction material?

For a genuinely novel material, plan on five to fifteen years from first evidence to code recognition, and two to six years to a third-party assessment with a useful scope. The dominant cost is not laboratory time but the sequence: a standardised test programme on production material, then assessment, then approving-authority and insurer acceptance, each waiting on the previous one. Variants of already-certified products are far quicker. This is why the certification clock, not the discovery clock, governs what a designer may specify.

What did DeepMind's GNoME actually achieve, and does it matter for construction?

GNoME used graph neural networks to predict the structure and stability of inorganic crystals at scale, and Google DeepMind reported 2.2 million new crystals, of which 380,000 were predicted most stable and released to the Materials Project. It also reported that external researchers had independently made 736 of them in a laboratory. It matters to construction as a supply of candidates, not of materials: a predicted structure is separated from a specifiable product by synthesis, scale-up, standardised testing, assessment and a supply chain.

Can AI speed up the certification of a new material?

Only in one place, and it is unglamorous. The evidence pack for an assessment is a traceability problem — each performance claim mapped to a test method, a clause, a report and a sampling record — and assembling it consumes months of specialist time. Language models handle that structured cross-referencing well under human verification, and it takes real weeks out of the assessment gate. Nothing else in the pathway compresses: testing needs physical specimens and elapsed time, and assessment and code recognition are institutional decisions.

Why do novel materials keep getting removed during value engineering?

Because the proposal transfers an unpriced risk to a commercial party. A material with one supplier, an unverified lead time, no trained installers and an approving authority who has not yet said yes imposes estimable costs, and an estimator will estimate them. The contractor prices supply risk, the client prices the absence of competitive tension, the insurer prices the concentration, and the sum routinely exceeds the technical benefit claimed. The fix is a second source and an early insurer position, not a better technical argument.

When should we involve the insurer or warranty provider?

Before the material enters the specification. The insurance and warranty position is the quietest veto on the pathway and the one most often engaged last, at or after financial close, when a no is most expensive. Engaging at concept costs nothing and reshapes the proposal usefully: acceptance with an inspection regime, acceptance with a reduced scope, or a clear refusal that saves two years. Get the position in writing, against the assessed scope, for the specific use.

Is generative or topology-optimised structural design ready for real projects?

The optimisation is mature; the integration usually is not. Machine-learning approaches to topology optimisation are well documented in the research literature and the results are genuine — less material for the same structural performance. The failure is at the boundaries: objective functions that omit formwork, reinforcement congestion, transport envelope and inspectability, and output that gets redrawn conventionally because the analysis model and fabrication package cannot consume it. Fixing either boundary is worth more than improving the optimiser.

What is the difference between an ETA, an agrément certificate and an ICC-ES report?

They are the same instrument in three regimes. A European Technical Assessment is issued by a designated body against a European Assessment Document, for products outside a harmonised standard. A UK agrément certificate, from bodies such as the BBA, performs the equivalent role in Great Britain alongside the construction products regime and the Approved Documents. An ICC-ES evaluation report does the same against the International Code Council's model codes in the United States. In each case, the document that matters is the assessed scope.

How do we design a pilot so the approval travels to the next project?

Design it to produce evidence a third party could assess. That means standardised sampling to published methods rather than convenient ones, a conventionally built control section on the same asset, agreed reporting intervals, a named data owner who does not work for the supplier, and registration of the whole record where the next design manager will find it. A pilot designed this way becomes the first tranche of an assessment's evidence base. A pilot designed for a case study becomes an anecdote with photographs.

Where does 4D printing sit — is any of it specifiable today?

Not in permanent structural works, and the obstacle is structural in both senses. The inverse-design problem is genuinely solved for a growing set of cases, and convincing demonstrators exist. What does not exist is any assessment framework for a component intended to change geometry in service: every structural code assumes the geometry you designed is the geometry present in year forty. Until an assessment route accommodates intentional change, these components stay in demonstrators, facades and non-structural applications.

Does the pathway differ for a contractor, a designer and an asset owner?

The six gates are identical; the leverage is not. An asset owner can write a performance requirement into a framework and fund evidence across many schemes, so they can reach code recognition. A designer controls the requirement and the design record, which is where evidence provenance is won or lost. A contractor controls supply, workmanship and the price that value-engineers a material out. The most common failure is each assuming another party owns the acceptance route, so nobody builds one.

About the author

Atomic Loops Engineering

Industrial AI practice

Atomic Loops builds production AI systems for construction, infrastructure, manufacturing and energy operators — property prediction, generative and optimisation-led design, computer-vision inspection and decision support running against live project data, delivered into the design and assurance tools a project already uses rather than as a research demonstrator.

· Production deployments across contracting, asset owner and design-practice estates
· Materials and assurance data pipelines built with client technical teams
· Delivery framed around what a design record and an approving authority will accept
· 36 cited sources on this page

Sources

Regulations, compliance and governance for construction AIThe governance charter that decides who approves an AI-derived design output — and how model-derived figures are kept out of a compliance claim.AI readiness and transformation roadmap for construction & infrastructureHow to sequence the data, tooling and supplier work behind a materials programme so the certification route is not the thing you start last.Construction & infrastructure practiceProduction AI systems Atomic Loops builds for contractors and infrastructure owners — delivered into the design record and the assurance file, not as demonstrators.

Find out what you could actually specify next quarter

We take one material and one asset, walk the six certification gates with your technical, procurement and insurance leads, and leave you with a costed acceptance route: what evidence exists, what has to be generated, who signs each gate and how long each takes. You keep the route whether or not we build anything.

Book a materials review See our construction work

Published 14 August 2026 · Last updated 14 August 2026

Key takeaways

Abbreviations used on this page

Score your project on the buildability ladder

What morphic materials are — and why certification, not discovery, sets the pace

The five rungs in detail: Conventional to Standard practice

Conventional

Specified-novel

Piloted in works

Code-recognised

Standard practice

Where construction projects actually sit on the ladder

What is genuinely real today — and what is still a research result

The certification gate: the six approvals a new material must clear

The order that actually works

Where AI actually helps — and where it changes nothing

What this looks like in public

The specification stack, layer by layer

Materials data

Prediction & design tools

Physical evidence

Assessment & certification

Supply & skills

In-service assurance

A 90-day plan: get one adaptive material into permanent works on one asset

Why the order matters

Procurement and supply: the readiness nobody checks until it is too late

Failure modes that turn a certified material into an incident

The certificate used outside its assessed scope

Model-derived figures entering the evidence pack as measurement

The single-supplier dependency discovered at tender

Optimised geometry nobody can build or inspect

The pilot whose evidence was never registered

Glossary

Frequently asked questions

What are morphic materials in construction?

Is self-healing concrete actually used on real projects?

How long does it take to certify a new construction material?

What did DeepMind's GNoME actually achieve, and does it matter for construction?

Can AI speed up the certification of a new material?

Why do novel materials keep getting removed during value engineering?

When should we involve the insurer or warranty provider?

Is generative or topology-optimised structural design ready for real projects?

What is the difference between an ETA, an agrément certificate and an ICC-ES report?

How do we design a pilot so the approval travels to the next project?

Where does 4D printing sit — is any of it specifiable today?

Does the pathway differ for a contractor, a designer and an asset owner?

About the author

Sources

Related reading

Find out what you could actually specify next quarter