What AI could mean for America's housing crisis

America's 4-million-home shortage is often framed as a construction problem: labor shortages, material costs and the need to build faster. While those matter, much of the cost and delay accumulates before construction begins.

The process of designing, reviewing and approving buildings still runs on fragmented workflows. AEC professionals spend countless hours navigating regulations, coordinating documents and resolving issues that could have been caught earlier.

Regulations play an important role in this process. The complex compliance system of creating and enforcing laws ensures we build safer, more resilient buildings. But this disjointed workflow is creating downstream effects for everyday Americans.

The National Association of Home Builders (NAHB) estimates that government regulations add 26.4% to the price of an average new single-family home, with building code changes over the past decade accounting for the largest share of cost, $40,288.

Permitting delays add more. The NAHB estimates regulatory delays during lot development average roughly seven months when they occur and puts the "pure cost of delay" (the financing cost of that lost time alone) at about $2,480 per home during development and $1,632 during construction.

The answer isn't fewer code updates or cutting corners on reviews. It's tools that let builders and officials review plans and apply codes more efficiently.

AI is only as good as its data

There's real excitement about AI's potential in architecture, engineering and construction. But AI is only useful when it can reach reliable, structured information.

Because generic LLMs process language sequentially, building codes are hard for them to interpret correctly. A single requirement might have five parent categories, three exceptions and a footnote. Standard models often struggle to connect these overlapping layers.

To test this, in October 2025, UpCodes compared ChatGPT, Gemini and Microsoft Copilot against its own AI across 43 real-world code scenarios, validating each response with an experienced code expert. Generic models averaged 36% accuracy, while UpCodes' purpose-built AI achieved 93%, more than double the next-best model. Where the general models went wrong, it was usually missing context: an exception, a cross-reference or a local amendment.

Without structured, well-maintained regulatory data, generic models often produce confident-sounding but incorrect answers. Output quality depends on information quality.

Reducing friction across the building lifecycle

One clear use for AI is cutting the time spent on repetitive tasks that slow projects and eat into margins. Plan review is one example: checking a design against applicable codes often means navigating thousands of pages of regulations, amendments and project documents.

The industry is at an inflection point. AI can now automate first-pass QA/QC analysis by cross-referencing drawings, specs and locally adopted codes to flag potential issues.

The same opportunity exists throughout the permitting process. Many jurisdictions face staffing shortages and growing backlogs, and officials must protect public safety while reviewing ever more complex projects.

Disasters show how little slack these systems have. After the January 2025 fires destroyed roughly 11,000 homes across Los Angeles, fewer than 10% of owners had permits to rebuild eight months later. By September 2025, LA County had approved only 405 of 1,972 applications in areas it oversees, and Malibu had issued just two permits for nearly 600 destroyed homes.

Maui saw the same strain: two years after the Lahaina fire, only 45 homes, about 2% of what was lost, had been rebuilt, in a county where permits averaged about a year to process before outside contractors were brought in.

Better technology helps reviewers work faster without lowering standards. Every issue caught during design is one less cause of delay, rework or costly change during construction.

Construction professionals spend 35% of their time on non-productive work, such as searching for information and fixing mistakes, resulting in about $177.5 billion in labor costs annually. Nearly half of all rework traces back to bad data and miscommunication. Catching a code conflict during design, when it's cheap to fix, beats catching it in the field.

That requires integrated systems, not standalone apps layered onto broken processes. Progress comes when tools dynamically integrate building codes, product data and project documentation into one workflow. With that foundation, AI acts as a nervous system, connecting information to enable faster, more informed decisions.

The right information leads to better outcomes

The construction industry adopts new tools when they solve real problems. Building Information Modeling improved coordination between disciplines; cloud collaboration made project information easier to share. AI is the next step, but only if it's built on the right information. When codes, local amendments, product details and project documents are connected, teams catch issues earlier, cut delays and make better decisions. At scale, that lowers costs, shortens schedules and lifts productivity.

No single technology solves the housing crisis. The opportunity is more practical: modernize the workflows we already have, organize the information behind them and let AI take the friction out of design, permitting and construction.

That's the work we're focused on at UpCodes: connecting codes, plans and specs into one workflow to help teams build more efficiently and affordably.

Experience UpCodes AI-native QA/QC platform live at Greenbuild (booth #1136) or visit UpCodes to learn more.