Making Housing (Artificially) Intelligent

Much of the talk about artificial intelligence’s role in the built environment has, unsurprisingly, been about the massive amounts of computing power the new technology requires, and the real estate, construction, and energy that compute requires.

AI’s relationship with housing is fraught; polling shows that the public expects job elimination which will make it harder to afford housing (59%) at a two-to-one ratio to expecting AI to strengthen the economy and improve housing affordability (30%) [1]. In a darkly comic twist for the many of us who have struggled against NIMBY mindsets for years, AI data centers have overtaken multifamily buildings as the building type most likely to be opposed in a neighborhood [2].

AI is suffering from a legitimacy problem and a lack of public acceptance. Workers are worried about displacement, and communities resist infrastructure they perceive as increasing energy costs rather than driving local public benefit. Earlier this week, President Trump signed an executive order increasing government oversight over large language models [3].

The AI industry requires a parallel civic strategy, and could align efforts towards reducing rising housing costs, the crisis that most acutely affects Americans in their everyday lives. Housing and construction are prime candidates for change, though AI would not be the first technological revolution to find itself lost in the industry’s complexities. However, these giant processing machines have unique potential to address housing’s particular problems: antiquated zoning and permitting, challenges to pricing and financing risk, thousands of pages of construction documents, inefficient design generation, and labor-intensive construction.

The housing construction process is rife with complications: it is risky with unknown approvals, costs and delays, and significant financial exposure; layered with multiple key parties across a lengthy process; and bespoke with over-customization due to the regulatory and methodological environment. Every step in the housing process compounds complications that slow processes down and increase costs. Over time these layers of risks, middlemen, and customization have bogged down the system and limited the number of players who can reasonably navigate it. As part of an effort organized by the National Housing Crisis Task Force, some of the co-authors of this piece hosted a discussion this past fall exploring these challenges in more depth. A presentation outlining the themes further can be found here.

Other industries are outpacing housing in the interplay between technology and affordability – not least AI data centers, which are innovating to provide computational power quicker and more cheaply. Early last year, we wrote about Housing for Chips [4] — the idea that capital markets were mobilizing at unprecedented scale to build data centers but not doing the same with the housing challenge – a dynamic which is playing out with technological innovation.

AI applications have, though, begun to impact the housing process with targeted solutions to its series of inefficient steps, and the industry could further drive more transformative, systems-level shifts. The industry, from cutting-edge research labs to start-ups to incumbents, should target three essential components of the housing construction process: Govern Better, Finance Better, and Build Better. AI’s ability to ingest data and customize outputs brings the promise of adapting housing processes to what we want to build, rather than adapting what we build to our antiquated processes.

Govern Better: Remove Barriers

Time is money and permitting and compliance take a lot of time. Reviewing plans is a lengthy part of the process of building housing. One study attributes 40.6% of the total cost of multifamily development to regulation [5]. It was estimated that in Los Angeles County, permitted vacant land sold for 50% higher than unpermitted land – representing the pure economic value of navigating the process [6].

AI can both be a catalyst for regulatory reform and a direct substitute for slow processes. AI tools that digitize and standardize permitting and code compliance are compressing timelines. Large language models can scan construction drawings and ingest thousands of disparate zoning codes to flag conflicts and pre-screen applications, so they arrive complete on a reviewer’s desk. This results in fewer correction cycles, shorter project queues, and reviewers freed to focus on more complex judgment calls.

Early pilots in a few geographies are showing promise. Honolulu has had issues with elongated cycles of back-and-forth with permit applicants. In 2023, Honolulu partnered with the start-up CivCheck, a software company that uses AI to streamline permitting for specific jurisdictions [7]. With technology offering a compliance suggestion but always leaving final judgment to the department’s reviewer, Honolulu reduced review time to 15 to 20 minutes. In Los Angeles, Governor Newsom deployed Archistar’s AI-powered plan review to accelerate rebuilding permits after the 2025 wildfires. Early results showed permitting times reduced by more than 36%, with the tool provided free to homeowners through philanthropic funding [8]. Maryland is working with Anthropic and Percepta to “improve permitting and licensing timelines, and support Maryland’s efforts to expand housing development” [9].

Just as regulations can cause bottlenecks in housing, institutional capacity is a lesser understood impediment. Public housing authorities, planning departments, and housing finance agencies often lack the capacity to maximize the tools and authorities they already have. Forward-thinking changes like rezoning are hamstrung by the challenges of understanding the effects of various permutations.

At the National Housing Crisis Task Force and in this newsletter, we have highlighted Atlanta’s Urban Development Corporation [10], which uncovered untapped powers from a 1937 Georgia Housing Authorities Law to provide tax exemptions for affordable housing and innovate building code regulations. Atlanta’s team dug up this statute via diligent work and happenstance. AI models could comb through laws and statutes to help countless localities better understand potential areas with latent authorities, helping prioritize policy and legal staff time on high-leverage reforms.

Likewise, AI-enhanced innovations like digital twins, virtual replicas of places and their infrastructure, can help jurisdictions model the impacts of zoning changes before committing them politically. While current processes may hinge on the results of a dated environmental review or traffic study, digital twins could enable dynamic, multivariable forecasting to improve understanding of true impacts of development. MIT’s Real Estate Innovation Lab has received a grant to develop digital twins as a “living data integration platform” that forecasts the implications of new development and has partnered with the City of Boston Housing Innovation Lab to workshop solutions for affordable housing production [11].

Many of the point solutions in the site selection, zoning, and entitlement process focus on making what are currently highly bespoke processes more efficient. But a truly transformative future state could see code adapting to the built environment’s needs, instead of the opposite. AI’s processing power could drive outcome-based code, in which code adapts to particular sites and desired buildings, while maintaining the policy and safety intentions of the original underlying regulation.

Finance Better: De-Risk

Housing finance is remarkably narrow for a sector of its size. On the market-rate side, over 90% of residential construction loans originate from banks and thrifts [12], and those banks have tightened standards across multiple consecutive quarters [13]. Construction loans remain short-term, variable-rate, and often personally guaranteed, with every deal individually underwritten with rigid timelines that don’t allow for innovations, such as modular construction. On the affordable side, the challenge compounds: the average number of funding sources in a LIHTC deal doubled from two to four between 2000 and 2017 [14], each carrying its own compliance regime. And across both market-rate and affordable, multifamily insurance premiums have doubled since 2021, destabilizing pro formas and reflecting difficulty in underwriting risk [15].

In the near term, AI is driving point solutions for more automated underwriting. AI can more efficiently pull and match sales and rental comparables as well as construction costs. AI can also drive diligence on title, code, and other construction considerations. Other operational compliance processes can be dramatically improved; for instance, Pronto is automating affordable housing compliance for landlords and tenants, which is driven by a complex stack of regulation [16].

Frontier models and start-ups alike are working to make financial underwriting more efficient. Banking and private equity analysts traditionally spend weeks in Excel; the potential to automate financial modeling could make institutions more efficient and implicitly reduce their cost of capital. Insurance built on top of more advanced machine learning and AI could reward projects that are both geographically and structurally more resilient.

Today, financing comes in rigid buckets defined by the regulatory environment, institutions’ need for efficiency, and historical norms. AI models could bring on a structural shift in underwriting by digesting more, better data, making financial modeling more precise and risk better understood. Instead of planning projects to satisfy the inflexible needs of financial institutions, AI capabilities could enable financial institutions and insurers to dynamically price risk and even drive new financial products. For instance, modular housing construction has struggled due to the development of factories being so financially distinct from traditional construction loans for individual projects. In a future state with more accurately priced risk, we could see financial innovation keep pace with technical innovation.

Build Better: Increase Productivity

Finally, the white whale of housing innovation has long been improvement in construction productivity. The industry has been on the cusp of change before but has yet to find efficiency leaps like it did in the postwar period or the Sears catalogue. AI might finally be the unlock for the nascent (domestically, at least) modular housing sector, while also bringing massive efficiency and improvement to existing construction design and documents processes.

Researchers at the Technical University of Denmark developed a multi-agent AI system for affordable housing site selection that evaluates parcels against 127 federal and local regulatory constraints. Deployed in partnership with New York City’s 2026 affordable housing initiative, the system reduced site selection time from 18 months to 72 hours while identifying 23% more viable locations than human experts [17].

We have long heard about modular construction’s capacity to compress timelines by 30-50% and reduce costs, but we have yet to see scaled, consistent results in the U.S. In Sweden, roughly 84% of homes are built with prefabrication methods [18]. The problems that have dogged modular were never solely about the factory; they were about configurations, logistics coordination, financing, regulatory variance across jurisdictions, and design inconsistency. These types of information problems could be solved by AI’s processing power to bring together (and structure) disparate data and efficiently provide optimized results across a complex set of factors. AI can also optimize factory operations, as it has in other industries [19]. Startups like Reframe are driving robotic production in microfactories for “missing middle” housing [20] [21], and Aro Homes has already successfully proven a modular production and sale cycle for high-end single-family homes in the Bay Area [22] [23]. Carnegie Mellon University is seeking to connect leading construction automation research to real-world deployment through its new Robotics Innovation Center at Hazelwood Green [24].

On the design side, construction drawings are a long, expensive, iterative process. Generative design software driven by AI can produce compliant baseline plans in hours instead of weeks, leaving architects to focus on creative and site-specific work. Likewise, the construction industry is dogged by incredibly detailed documents and plans. Trunk Tools lets construction workers query tens of thousands of project documents in plain language; a Gilbane pilot found it avoided over $100K per month in rework [25] [26]. These iterations, whether in drawing or reworking, are enormous cost drivers.

Outside of new construction, renovations and retrofits are being made more efficient by companies like TailorBird, which turn publicly available visual data into architectural drawings [27]. Applications can drive better long-term operations as well. In the UK, Wolverhampton Homes used AI to predict risky spots for damp and mold with 98% accuracy in 21K homes [28]. Other significant damages like floods and fires are more frequent in certain types of risk-prone housing. AI powers leak and flood sensing systems, paired with sensor hardware [29].

In the future, the AI industry has the potential to lay the groundwork for a more automated housing construction system. From design to construction, AI models can be the backbone of a system that builds more efficiently — modular or site-built — across specific sites and jurisdictions. Anori, a Google X spinout, aims to solve this systems level problem, seeking to harness AI to pull all constraints — entitlements, financing, cost, materials, and design – upfront, allowing developers, architects, municipalities, and investors to address them in a shared space [30]. The potential future state would mean building in a cheaper, faster, and more energy- and resource-efficient way, bringing direct benefits to local economies. A committed investment in construction technology can unlock long-stagnating productivity growth for the industry.

Jolting an Industry into the Future

Housing is uniquely positioned to be a sector in which AI earns back some public trust. The housing crisis is now permeating across geographies and income levels. It is harrying federal, state, and local officials and the private sector alike. An AI for the common good can drive genuine strides in our housing crisis.

Of course, the deployment of AI-driven tools needs to be paired with regulatory, governance, and capital innovation. However, this is a technology powerful enough to not only complement but enhance existing reform efforts while also solving gap problems on its own. True impact will come from AI showing up as a willing partner to open-minded practitioners. Cities need to pursue permitting efficiency. Institutional capacity needs to be identified and built. Capital markets need to adapt to construction innovation. But AI can reverse-engineer the system, allowing us to match processes to the built environment we want, rather than compromising what we build due to antiquated and inefficient processes.

The AI sector should focus on applying its breakthrough technology to everyday problems faced by the populace, and the housing crisis’ unique set of complexities could make it the ideal use case to show what’s possible. After years hearing about how AI might take our jobs and drive up our energy costs, how meaningful would it be if it instead made our housing more affordable?