Report No. 02 · Aedifice Research · 2026
Machine-Readable Buildings
How AI Accelerates the Circular Economy in New York — and What That Means Everywhere Else
April 20, 2026
New York has about one million buildings. The city keeps a public record of how energy is used, how safe the facade is, and whether the building is a landmark for roughly sixty thousand of them — about seven percent. The other nine hundred thousand are, for most purposes, invisible to anyone but the people who walk into them.
That is a problem. Buildings produce most of New York's climate emissions. Climate laws — the ones that set rules about insulation, heating systems, demolition, and reuse — can only work on buildings the city can actually see. Today, the city sees one in fourteen.
Artificial intelligence is not a silver bullet. It will not build a single new apartment. What it can do, cheaply, is turn the messy public data the city already publishes — permit filings, utility bills, facade inspections, boiler records — into something a human, an owner, or a policymaker can actually act on. This report is about that specific use of AI: not AI as hype, but AI as a cheap way to close a measurement gap that has blocked climate progress in the building sector for a decade.
It is written with New York as the working example. The playbook federates: the same techniques apply in Copenhagen, London, Tokyo, and anywhere else that has opened its building data to the public. Five chapters, roughly one hour to read.
The headline
798,474 of New York's 858,644 buildings — 93 percent — appear in no public compliance, energy, or landmark record at all. The bottleneck for a circular built environment is not AI capability. It is the measurement gap AI can close.
Author
Jeremy Edwards
Founder, Aedifice
Publication
Aedifice Research, Report No. 02. Published April 2026. Free to read.
Lineage
Sequel to Building Prosperity in New York. Report No. 01 identified the $36.8B addressable circular opportunity; this report is its execution layer.
Thesis
The decision cost, not the data cost
Buildings are the largest under-measured, under-modeled, and under-automated asset class in the climate economy. AI does not build a new category of buildings. It collapses the cost of deciding what to do with the ones we already have.
In New York that collapse is worth tens of billions of dollars per year. Globally it is the difference between a circular built environment that exists on paper — in the EU Energy Performance of Buildings Directive recast, in Denmark's Bygningsreglement, in California's CALGreen, in Local Law 97 — and one that executes. The data NYC already publishes could carry most of that distance. The measurement stack it has not yet built is the rest.
Report No. 01 established the $36.8 billion per year of circular-economy opportunity in New York's built environment. This report examines the execution layer that makes that opportunity bankable: the AI methods that turn PLUTO, LL84, DOB violations, FISP filings, boiler rosters, and the landmark designation record into decision-ready structure, and the policy framework that would extend the substrate past the regulated minority.
In no public record
93%
of NYC buildings
LL84 panel · ML-ready
2.2M
monthly rows
LLM-classifiable condition
1.09M
violation records
Intelligence to governance
5
chapters
Table of contents
Five chapters.
Ten NYC datasets
01
The Intelligence Gap
What the city measures, and what it doesn't
858,644 PLUTO parcels. 28,669 under LL97. 27,922 benchmarked under LL84 in 2024, with a 62 percent compliance rate against the required roster. 12,114 LL33 grades resolve to PLUTO; 52 percent grade D or F. Across every compliance, energy, and landmark dataset combined, 60,170 buildings are reached. The remaining 798,474 — 93 percent of the registry — are dark to every instrument simultaneously.
Published
02
The AI Toolkit for Circular Buildings
Six method families, anchored in NYC data
Computer vision against 85,769 FISP facade records. LLMs against 1,089,210 DOB violation descriptions. ML on 2,207,184 LL84 monthly readings. Combinatorial optimization over PLUTO and landmark rosters. Remote sensing and digital twins. What is production-ready, what is pilot-ready, and what remains research.
Published
03
Six New York Use Cases
The deployable half of the toolkit
LL97 retrofit sequencing under a carbon budget. Computer-vision FISP inspection. Auto-generated material passports from BIM plus DOB filings. LLM-based office-to-residential feasibility screening. AI-driven reclaimed-material matching. Embodied-carbon RFP scoring for NYC DDC procurement. Each case mapped to one of Report No. 01's six strategies.
Published
04
The Global Layer
What federates past New York
Cross-jurisdiction rule engines spanning LL97, Denmark's BR18 §297, EU EPBD 2024/1275, CALGreen, and Tokyo's Top Runner. Federated material-passport registries. Transferable climate-resilience models. A peer-city scorecard for Copenhagen, Amsterdam, Paris, Singapore, Tokyo. The standards battle between IFC, Autodesk Forma, and ESRI ArcGIS Urban.
Published
05
Risks and Governance
Six risks, six governance answers
Hallucinated code compliance. Training-data bias toward LL84 typologies. Inference energy versus abatement delivered. Data-ownership ambiguity across owner, tenant, and city. Regulatory capture on BIM-style vendor lock-in. False confidence in under-measured categories. The four-element governance framework: disclosure, audit, open standards, ownership.
Published
Methodology, in brief
Every headline figure in this report is anchored to a specific public table published by New York City, with the row count verified against the Aedifice Research Supabase mirror at the time of writing. The AI-method evidence is drawn from peer-reviewed work at the NeurIPS and ICML Climate Change AI workshops, the IEA's AI for Climate and Energy series, LBNL ComStock and NREL ResStock, the Carbon Leadership Forum's WBLCA benchmarks, and published academic AEC-AI literature. Where a specific venue is uncertain the claim is flagged rather than fabricated.
The report's global-layer claims reference the EU EPBD recast (Directive (EU) 2024/1275), Denmark's Bygningsreglement §297, California CALGreen, Japan's Top Runner program, and the buildingSMART IFC 4.3 and BCF 2.1 specifications. The risks chapter engages with NIST AI RMF, the EU AI Act Annex III, and the Anthropic Responsible Scaling Policy as governance anchors.
How to cite
Edwards, J. (2026). Machine-Readable Buildings: How AI Accelerates the Circular Economy in New York — and What That Means Everywhere Else. Aedifice Research, Report No. 02.