From Prompt to Passport

How AI can help us design more circular buildings

Architecture has a habit of turning tools into myths. The pencil became an extension of the hand; the plotter became a promise of precision; parametric scripts became a new form of authorship. AI is now being pulled into the same narrative machine. The debate oscillates between fear and celebration, between replacement anxiety and salvation fantasies. Both positions are convenient. Both avoid the harder work of describing what AI is actually good for in the built environment, and what it is not.

My interest in AI is not driven by the question of whether algorithms can “design.” The more urgent question sits elsewhere: can digital technologies make circular construction practical at scale? Can they reduce the friction that currently makes reuse feel heroic, slow, and too often financially irrational?

Circularity changes the starting point of architecture. The design no longer begins with a blank page. The brief begins with an inventory: what material already exists, where it is, what condition it is in, when it becomes available, how it can be recovered, and how it can be redeployed without destroying its value. That inversion is uncomfortable for a profession trained to specify. Yet the inversion is also where creativity becomes real, because creativity comes from constraint, negotiation, and care.

From demolition sites to material banks

Reuse will never scale if end-of-life buildings remain unpredictable. Too many components are discovered too late, after the schedule is fixed, after the tender is written, after the excavator is already on site. Material value collapses into waste because the construction sector is not designed to see buildings as future supply.

Computer vision and machine learning can support a different approach. AI-models can scan street-level imagery, real estate photos, and other large-scale data sources to classify façade materials, detect windows, and infer typologies. Combined with public records, renovation histories, and permitting data, these indicators can help anticipate where materials might re-enter the market. A city then becomes searchable by not only land use and density, but also material availability.

Vision systems are brittle. Performance varies across building types and across contexts. Models misread materials. Condition assessments from imagery remain probabilistic. Bias enters through what gets photographed, at which resolution, in which neighbourhoods, and with which data coverage. Municipal decision-making that leans too heavily on such models risks turning data gaps into planning gaps. Circularity could unintentionally privilege the most documented districts and penalise the ones made invisible.

We therefore need calibrated uncertainty and clear boundaries. Urban-scale detection can guide where to look, when to intervene, and where to invest in audits. On-site verification remains essential when safety, liability, and heritage value are at stake.

Components need information that can travel

Reuse fails less because of ideology than because of trust. A reclaimed beam is never just a beam. The beam carries a geometry, tolerances, connection details, a structural history, a damage profile, and a regulatory story. Without reliable information, procurement stalls. Engineers hesitate. Insurers hesitate. Contractors hesitate. The component remains stuck in limbo between potential and risk.

This phase is where AI can be most impactful, and also where its role looks least glamorous. Building information lives across BIM files, PDFs, scans, photographs, spreadsheets, energy certificates, and site notes. AI in digital passports helps standardise different naming systems and connect components to the right material information by recognising similar meanings. Large language models can help organise how parts relate to each other, so products can be understood, repaired, or recycled more easily over time.

This phase also brings the most uncomfortable questions. Tracking technologies, whether QR codes, RFID tags, or DNA-based carriers, raise data-governance issues immediately. Who owns the data, who maintains it, who protects it, who updates it, and for how long? What happens when a building changes hands multiple times? How does traceability avoid becoming surveillance?

Circular construction only works if information about materials is open and shared in common formats. If a few companies control the data about materials and products, they could block that data for profit, which goes against the whole point of a circular economy.

Reverse design and buildable improvisation

The design phase is where many people expect AI to “get creative,” often meaning image generation and rapid formal variation. Circular construction asks for a different kind of creativity. The task is not infinite novelty. The task is feasibility under irregular constraints.

Reclaimed components arrive with non-standard dimensions, uncertain quantities, incomplete histories, and unpredictable timing. Computational search can explore geometries that reduce cutting, avoid waste, and improve constructability. AI-assisted tools can support participatory processes by translating sketches and informal ideas into options that can be discussed and refined. Layout exploration can help produce fit-outs that anticipate change and avoid premature demolition of interiors.

The strongest contribution of AI here is coordination. The value sits in matching messy supply to messy demand. Constraint solvers, knowledge graphs, and procurement-aware design tools may matter more than another stream of seductive images.

Speed still carries risk. An explosion of plausible options can replace judgment with selection fatigue. Cheap concept generation can encourage more proposals, more speculative work, and more generic results. Circular design requires the opposite mindset. Circular design requires specificity, patience, and an acceptance that the material stock, not the moodboard, sets the agenda.

What needs to change in practice

A meaningful use of AI for circular construction demands shifts in how architects work, what clients procure, and what institutions regulate. Architects need to treat AI outputs as hypotheses rather than truths. Uncertainty should be explicit and auditable. Failure cases should be documented, not hidden.

The sector needs investment in the “boring” infrastructure: material data that works across different systems, product passports, disassembly maps, and shared naming systems. Prompt literacy will not replace data standards. Model-assisted decisions should be declared, not for fashion but for accountability and transparency. Verification, whether by engineers, fabricators, or auditors, should remain visible in the process.

Certain decisions must remain explicitly human because responsibility is social. Safety, ethics, heritage judgments, and community priorities cannot be delegated to statistical pattern matching. Automated inference can support those decisions, but cannot replace the obligation to justify them.

Critique needs to become tool-literate. Architectural culture already knows how to critique form. Architectural culture now needs to critique datasets, model behaviour, and incentive structures with the same sharpness. This is also what we, as professors in architecture and engineering schools, need to teach the next generation of designers.

Circularity without restraint becomes an alibi

Circularity is essential, and circularity can still be misused. Reuse can become a justification for continuing demolition. “Reusable” can become another label that protects business as usual. AI could lower the friction of reuse while also lowering the friction of building more. Rebound effects are real.

The debate around AI in architecture should therefore move away from the fantasy of automated authorship and toward the question of systemic direction. AI can help turn cities into material banks. AI can help build information infrastructures that make reuse reliable. AI can help reverse the design process so that reclaimed components become drivers rather than obstacles.

The outcome will not be decided by the model. The outcome will be decided by standards, contracts, incentives, and governance. That is the work worth arguing about. – Catherine De Wolf