A Holy Grail of streamlined local permitting and housing approvals has recently shifted from a pipe dream to a real-world, real-time solution as U.S. states and municipalities adopt a growing array of AI-powered tools.
A Bipartisan Policy Center session this week, titled “Housing Innovators: Building Better with AI,” explored how the latest technology can accelerate approval processes and pathways, but perhaps more importantly, where chokepoints remain.
AI has already proven that it can reduce review times. For example, Honolulu cut its permit review times in half using AI. However, while AI can speed up reviews, its effectiveness hinges on accurate, consistent and understandable zoning and building code data to guide its decisions.
The session’s panelists noted that, because those rules and data are often outdated, contradictory or riddled with exceptions, AI can tend to introduce errors in judgment, which, in turn, require even more human review and additional rounds of corrections.
For homebuilders, these constraints are more than a minor technical problem. They can add time to an already lengthy approval process, where even small administrative hurdles can compound into weeks or months of delays, and jeopardize projects altogether. As panelists explained, those delays often stem not from one major obstacle, but from a series of smaller inefficiencies that slow review times.
Permitting is a “death by 1,000 cuts”
Jenny Song, Founder and CEO at Infilla, an AI company that partners with municipalities to streamline housing permitting and approvals, argued that the permitting and entitlement process is often slowed or stalled not by a single large stumbling block, but by a host of small problems.
“It’s death by 1000 cuts for both government staff and for those going through the entitlements and permitting process,” she said.
For example, according to Song, only about 10% to 15% of applications that come to a municipality are complete at initial plan review. This means municipal staff must spend extra time finding missing information and going back and forth with builders, eating up more time from already-limited municipal staff capacity.
Many applications, particularly those with missing information, go through multiple reviews, with each cycle usually taking between 30 and 45 days. Timelines can stretch out far longer than anticipated, particularly when multiple government departments or agencies get involved.
This problem is compounded by staffing shortages in local government. As longtime employees retire, municipalities can lose institutional knowledge, while remaining staff spend their limited time and bandwidth on administrative work rather than substantive reviews.
AI and other technologies can help reduce some of that burden. In Sonoma County, California, for example, Infilla’s system saved the municipality about 40 hours of staff time per week during intake, allowing employees to focus more on substantive reviews and less on administrative work.
However, Song quickly clarified that AI and technological advancements alone can’t solve the permitting problem.
“It’s a mix of people, policy, process, and platform technology that will really make systematic change,” she said.
Inefficient zoning adds a layer of complexity
AI’s limitations become more apparent when the underlying rules and data governing development are notoriously difficult to understand.
Before a project can be reviewed, developers and builders must first determine what they are actually allowed to build, and that question can sometimes be surprisingly difficult to answer.
During the session, Sara Bronin, Founder of Land Use Labs, a company that digitizes zoning and land-use codes across the United States, tied permitting delays to this deeper problem.
Zoning rules can be hard to understand and administer. Permitting is where those rules are applied, so if an applicant can’t tell what a parcel allows, or if staff and applicants read the same code differently, applications can arrive incomplete or incorrect, forcing staff to spend extra time on reviews.
Part of the problem is the zoning itself. Different jurisdictions use differing terminology for similar concepts, and codes can run hundreds of pages, full of cross-references, tables, footnotes and exceptions. Standards vary from one district to the next, and developers and builders often have to compare written text with maps and GIS data just to learn what is allowed on a single parcel.
Atlanta exemplifies the full range of this complexity. The city has 268 zoning districts and an 850-page zoning code. Bronin favors “a shorter, simpler code” with clearer districts, clear graphics and terms that people can easily understand.
The problem is not simply that zoning codes are complicated. The underlying information can also be inconsistent.
According to data Land Use Labs reviewed, about 33% of jurisdictions have inconsistencies between zoning text and maps. Nearly 20% of districts listed in zoning text have no corresponding map, and some districts appear on maps but no longer exist in the text.
“Even the basic GIS data is out of date because it doesn’t match the official code, if there’s GIS data at all,” she said.
Those inconsistencies tie directly back to permitting. A municipality can invest in AI to make the review process faster, but if the underlying zoning information is inaccurate, outdated or inconsistent, those tools can only go so far.
That limitation becomes especially apparent when a project encounters a zoning provision that is easy to overlook.
For Vikas Enti, CEO of Massachusetts-based off-site homebuilder Reframe Systems, a project in Somerville, Massachusetts, became a case study in how easily hidden zoning details can slip through. Reframe built two three-story homes that were allowed under the city’s general zoning rules, but a separate rule on facades and front entries also applied.
AI missed it at first, and the city flagged it only after the homes had cleared state approval and manufacturing. A five-degree rotation of the side-entry door brought the project into compliance, but the example illustrates the pitfalls of AI reviews, particularly when the underlying zoning rules are complex or difficult to understand.
“All of a sudden, you end up with this tension of how you actually reconcile the town’s intent and zoning’s intent to allow for more housing. But then you get caught between what building codes actually require,” Enti said.
Looking beyond technology
Enti’s experience also illustrates a broader mismatch between how quickly homes can be produced and how long they take to move through permitting and approval.
Reframe Systems’ use of robotics and factory automation allows it to build homes up to three times faster than traditional homebuilders. Enti argued that homebuilding is increasingly “a game of speed,” and builders that can build the quickest have an advantage.
But those gains can be difficult to realize when projects remain stuck in lengthy approval and permitting processes.
“We’ve consistently demonstrated…that we can manufacture our homes significantly faster than we can get site permits,” Enti said.
That disconnect illustrates a central limitation for homebuilders – faster construction fails to matter if the regulatory process is riddled with errors or can’t keep pace.
One potential approach Enti proposed is to shift certain parts of the permitting process away from local jurisdictions.
He pointed to factory-built housing as an example of how a more centralized approach to plan review can work. The industry already incorporates this model in roughly 40 states, where state-level factory certification programs rely on third-party agencies to review plans, state agencies to oversee the process and inspectors to visit factories. Enti described the resulting approval process as being fairly efficient.
A similar approach could help streamline conventional housing permitting, he argued.
Rather than require municipal staff to handle every aspect of a review, qualified third parties or state agencies could take on some of the technical work, particularly building-code compliance. Builders could choose to engage third-party reviewers, while local officials could devote more time to land use, zoning and other decisions that require local oversight.
Enti’s proposal reflects a broader lesson from the panel: that technology alone can’t fix what’s slowing housing approvals. AI can make permitting faster, but getting the full benefit will also mean rethinking how reviews are done, tackling the regulatory obstacles that keep delaying projects and making sure zoning and building code data is accurate, consistent and easy to understand.
