The Federal Government Just Bet $3 Million on AI Permitting. The Best Evidence Comes From 19 Houses in Honolulu.

In June, the U.S. Department of Housing and Urban Development published a Notice of Funding Opportunity offering state and local governments up to $3 million to deploy automated building code permitting systems. Applications closed July 13, and somewhere in a federal office, someone is now deciding which cities get taxpayer money to let software check their building plans.

Every piece of evidence supporting that decision fits on a single spreadsheet.

Honolulu's Department of Planning and Permitting launched CivCheck, an AI-assisted permit preparation tool built by Clariti, in December 2025. By mid-2026, the department had processed 19 residential permits through the system. Nineteen. It compared those against 17 similar permits that didn't use CivCheck. Results: average review time dropped from 73 days to 32.5 days, and plan review cycles fell from 3.4 to 1.4. Davis Pitner, a DPP public information specialist, called these "the most current performance results available for AI-assisted permit review."

He is almost certainly right, and that should concern everyone involved.

Two Programs, $13 Million, Four Named Vendors

HUD actually launched two separate AI homebuilding programs with the same July 13 deadline. Its first program, the Automated Permitting Systems Demonstration (NOFO PDR-2600-DC-029O), offers $300,000 to $1.5 million per city for three-year software licensing, staff salaries, and operational expenses tied to AI permitting platforms. Its second, the Mass Market Solutions for Leveraging Robotics and AI Technologies for Home Construction Demonstration (PDR-2600-DC-029Q), awards up to $10 million per project for robotics and AI in factory-built housing. HUD expects to identify successful applicants by September.

Buried in the permitting NOFO is an unusual clause. It states that "participating jurisdictions could deploy systems such as PermitFlow, Blitz Permits, CivCheck, Permitify, or similar platforms." Federal funding announcements rarely name specific commercial products. An "or similar platforms" qualifier provides legal cover, but the signal to applicants is unmistakable: these four companies have the federal government's attention.

One of those companies is Clariti, CivCheck's parent. Clariti's vice president of government affairs, Julia Richman, served as deputy executive director at the Colorado Governor's Office of Information Technology before joining the company. She told StateScoop in June that Clariti is "helping governments to apply for the grant funds." A former senior state technology official, now employed by a vendor named in a federal grant, coaching cities on how to apply for that grant, in an arrangement where nobody has alleged wrongdoing but the geometry would make a procurement officer reach for the ethics manual.

What the Numbers Actually Show

Strip away the press releases and the data from Honolulu is genuinely interesting, if limited.

A 55% reduction in average review time is large enough to matter: a permit that takes 32 days instead of 73 saves the homeowner roughly $8,200 in carrying costs on a median-priced Honolulu home, assuming a 7% mortgage rate and $830,000 purchase price, which is the kind of money that covers a kitchen appliance upgrade or half a bathroom remodel. Fewer review cycles may matter more: each cycle represents a round trip where the applicant fixes errors, resubmits, and reenters the queue. Fewer cycles means fewer opportunities for a project to stall.

CivCheck also made a measurable dent in Honolulu's backlog. Honolulu's prescreen queue dropped from a six-month wait in 2023 to seven days. That figure reflects broader workflow improvements, not just the AI tool, but the trajectory is clear.

But the problem is statistical, not directional: nineteen permits against 17 controls is a sample that wouldn't survive the methods section of an undergraduate statistics paper, and a confidence interval on a 55% improvement from n=19 is wide enough to drive a bulldozer through. No randomization governed the comparison group, and we have no idea whether the 19 CivCheck applications were simpler projects, more experienced applicants, or submissions that would have sailed through the old process anyway, which is exactly the kind of selection bias that turns a promising pilot into a misleading headline. Pitner acknowledged the limitation: "As additional applications are completed, DPP will continue to validate and report these results using a much larger dataset."

Until that larger dataset exists, the federal government is writing checks against a promissory note.

What Other Cities Know (and Don't)

Austin formally adopted Archistar's eCheck as a pre-check layer for single-family home permits, though no published outcome data exists. Seattle's former mayor Bruce Harrell signed an executive order in June 2025 directing development applications through an AI pilot program, with full rollout expected sometime this year, and Denver is also using CivCheck, but neither city has published outcome data either.

Absent data is not the same as absent results, and these programs are genuinely new: CivCheck launched barely seven months ago, and you cannot have large-scale evidence for something that barely exists at scale. That policy catch-22 is real: waiting for ironclad evidence before funding pilots guarantees the evidence never materializes.

But the catch-22 argument has a limit. A 2025 National Association of Home Builders survey found that only 1% of single-family builders reported using AI to operate equipment. One percent. Between federal funding ambitions and industry adoption sits not a crack but a canyon.

What This Means for Your Permit

If you are pulling a residential building permit in 2026, AI is unlikely to touch it. Cities experimenting with these tools number in the single digits. HUD's $3 million program, if it awards the maximum to two cities, adds two more. Even an optimistic projection puts AI-assisted permitting in fewer than a dozen jurisdictions by the end of 2027. There are roughly 19,500 cities, towns, and villages in the United States with building departments.

A more immediate question is whether AI pre-checks actually reduce your timeline or just shift the bottleneck. CivCheck catches incomplete submissions and code violations before they enter the formal review queue. That front-loading is valuable when application quality is the primary cause of delay. But if your jurisdiction's bottleneck is reviewer capacity, not submission quality, the AI catches errors faster while the corrected application still sits in the same understaffed queue. At that point, the tool becomes an expensive filter ahead of an unchanged pipe.

Dheekshita Kumar, CivCheck's CEO, has been careful to frame the tool as augmenting reviewers rather than replacing them. That is the correct framing for political reasons and probably the correct framing technically. A harder question, which nobody involved has a financial incentive to ask, is whether the $300,000 to $1.5 million a city spends on AI permitting software would save more time if spent hiring two additional plan reviewers.

What No Software Can Fix

Permitting delays are a real crisis. In March, an Independent Institute analysis noted that "developers build long delays into project timelines, and smaller projects are often avoided altogether." A Virginia homebuilder reported losing a permit mid-review because the county staffer handling it resigned and didn't transfer the file, a breakdown that no amount of AI pre-screening would have prevented because the failure was personnel management wearing a technology hat.

Subjective interpretation of building codes compounds everything, because different reviewers read the same code differently, and one Independent Institute writer compared the situation to baseball umpires with varying strike zones, where a batter's experience depends less on the rule than on who is calling balls and strikes that day. AI could standardize that interpretation, and in some limited demonstrations it appears to do so. But standardization carries its own risk: a machine that consistently applies a wrong interpretation is worse than a human who inconsistently applies the right one, because the machine's errors are invisible until something fails an inspection or, in a worst-case scenario, a structural element.

A March 2026 study from the University of East London, published in Frontiers in Built Environment, reviewed 60 peer-reviewed studies on AI in construction management and found that risk prediction systems and scheduling optimization systems still operate as "separate dashboards." The AI that identifies your permitting problem doesn't talk to the AI that adjusts your construction schedule. Connecting them is a research problem, not a deployment problem, and no amount of federal grant money for permitting platforms addresses it.

Limitations of This Analysis

This article relies on published summaries of the HUD NOFOs rather than the full NOFO text, and the number of cities that applied before the July 13 deadline has not been disclosed. Honolulu's CivCheck results are self-reported by the DPP and have not been independently verified, while cost data for the AI tools themselves, including the per-permit licensing fees and integration expenses borne by the jurisdictions, is not publicly available. No randomization governed the comparison between CivCheck-processed and non-CivCheck permits, and selection effects may inflate the reported improvement; meanwhile, Austin's eCheck and Seattle's AI pilot have not published comparable outcome data, which means Honolulu's results stand alone and cannot be validated against other deployments.

Where This Leaves You

HUD's grant programs are a bet, not a verdict. Early data from Honolulu points in the right direction: faster reviews, fewer revision cycles, shorter backlogs. If the improvement holds at scale, it could save homeowners weeks of delay and thousands of dollars in carrying costs per permit. Those are outcomes worth chasing.

But the federal government is funding a nationwide demonstration based on evidence from 19 houses in one city, with a NOFO that names four specific vendors, while the former technology chief of a state government helps those vendors' customers apply for the money. Sound policy? Possibly. Thin evidence trail? Absolutely.

Your building department is not getting AI this year. Want your permit faster? Submit a complete application, respond to corrections the same day, and find out which plan reviewer handles your project type. That last piece of advice has a shelf life. Probably not a long one. But for now, the most effective permitting technology is still a phone call to the front desk at 8 a.m. on a Tuesday.