Forty Cities Use AI to Review Building Permits. Nineteen Thousand Don't. The Math Isn't the Problem.

In Honolulu, a residential plan review that once consumed ninety minutes of a building department staffer's day now takes twenty. Honolulu deployed Clariti's CivCheck AI tool to pre-screen permit submissions, and it cut residential plan review times by 70%, according to Clariti's May 2026 announcement. Total review time dropped 64%. Dawn Takeuchi Apuna, the department staffer who oversees the process, called it a game of consistency and confidence, not replacement.

Austin ran 190 submissions through Archistar's eCheck pre-review tool. Zero negative comments from applicants. Fifty percent reduction in staff review time. Janet Heit, chief administrative officer for Austin Development Services, told Homes.com News in April that staff didn't just accept the tool. They adopted it as a second set of eyes for double-checking calculations. Austin pays $1.1 million a year for the service.

Naples, Florida, paid $514,096 to license Blitz AI's compliance platform. Resubmission rates fell 85%.

Three cities. Three different vendors. Three independently verified improvements ranging from dramatic to transformative. It works. That is no longer a question anybody seriously contests.

So here is the question that matters: why do roughly 19,960 other permit-issuing jurisdictions in the United States not use it?

The universe is enormous

The U.S. Census Bureau's Building Permits Survey tracks approximately 20,100 permit-issuing places across the country. Shovels.ai, a construction data company, independently mapped over 10,000 Authorities Having Jurisdiction. These are the local governments (cities, counties, townships, boroughs) that stand between an approved set of plans and someone's foundation pour.

Against that universe, the count of jurisdictions meaningfully deploying AI in permitting is somewhere between 30 and 40. I assembled a list from vendor announcements, government press releases, and news coverage: Honolulu, Austin, Denver, Naples, Los Angeles, San Jose, Lancaster (California), Louisville, Bellevue (Washington), Pueblo County, Arlington (Texas), San Mateo County, Harris County (Texas), and a handful of others in various stages from signed contracts to exploratory RFIs. Boston published a formal Request for Information in May 2026, seeking proposals but not yet deploying. Seattle and Calgary are running CivCheck pilots.

That is an adoption rate of approximately 0.2%. If this were a pharmaceutical, it would still be in Phase I trials with fifteen volunteers. If it were a consumer product, investors would call it pre-traction. For a technology with a White House executive order behind it, bipartisan congressional legislation supporting it, and dozens of states listing permitting reform as a policy priority, 0.2% is a number that demands explanation.

What the delay actually costs

Residential building permits in the United States take two to eight weeks to process, according to mykukun.com's 2026 analysis. Delays beyond the initial review (resubmissions, corrections, interdepartmental handoffs) can extend timelines by months. Los Angeles routinely exceeds a year for single-family permits, with fees climbing into the tens of thousands, per the Independent Institute's March 2026 analysis.

I ran the arithmetic. Roughly 1.5 million residential building permits are issued annually in the United States. If AI pre-check tools save an average of one to two weeks per application, a conservative estimate given Austin's reported "days to weeks" of savings and Honolulu's 64% total time reduction — the carrying cost alone is significant. One week of delay on a $400,000 home financed at 7% costs approximately $538 in mortgage interest that the buyer will eventually absorb, because builders price delay risk into their bids. Add contractor standby costs, scheduling disruption for trades waiting on approvals, and the opportunity cost of capital sitting idle, and a conservative per-permit cost of avoidable delay lands around $2,000.

Multiply that across 1.2 million permits that could plausibly benefit from AI pre-screening, excluding the simplest applications where review is already fast, and the aggregate cost of non-adoption is approximately $2.4 billion annually. Not money flowing from buyers to cities. Not revenue. Pure deadweight loss, transferred from homebuyers and builders to nobody at all, evaporating in the gap between what technology can do and what governments choose to purchase.

The incentive problem nobody discusses on the record

Scott Beyer, writing for the Independent Institute, is one of the few analysts to name the political economy directly. "The reason [adoption remains limited] is rooted in incentives," he wrote. "Some groups, from a public choice perspective, don't want this efficiency."

He identified three constituencies that benefit from slow permitting. Municipal departments, where faster approvals could reduce staffing needs, budgets, and institutional influence. Existing homeowners, for whom lengthy permitting stifles housing production without requiring anyone to publicly voice opposition to a specific project. And elected officials, who depend on both groups for support and have little to gain from antagonizing either.

The beneficiaries of faster permitting (builders, future homeowners, renters priced out of markets where supply cannot keep up with demand) are diffuse. They do not attend city council meetings about permitting software procurement. They do not have lobbyists. Many of them do not yet live in the jurisdictions whose decisions will determine what they pay for housing.

This is not a conspiracy. It is a standard collective action problem that political economists have documented for decades: concentrated benefits and diffuse costs produce policy stasis, even when the efficient solution is obvious and cheap. Austin's entire AI permitting contract costs $1.1 million per year. Austin processes roughly 15,000 residential permits annually. That is $73 per permit. For context, the average building permit fee in the United States is $1,688, according to Angi's 2026 data. At $73 per permit, the tool costs 4.3% of the fee the applicant is already paying.

The liability vacuum does not help

There is a separate, legitimate obstacle that deserves acknowledgment. Standard construction contracts, specifically the AIA documents and ConsensusDocs templates that govern the vast majority of construction projects in the United States, do not mention artificial intelligence. Not once. According to Area Development's Q4 2025 analysis, exactly one standard contract template in the entire construction industry references AI in any capacity.

When Honolulu's CivCheck tool flags a code violation that a human reviewer would have caught, the liability chain is clear: the human reviewer signs off, the human is responsible. But what happens when the AI misses something? When it approves a setback calculation that violates a local overlay district the training data didn't fully capture? What if the applicant relies on the AI pre-check, skips their own code review, and builds a structure that requires demolition?

Nobody has litigated this yet. No major construction AI lawsuit exists. Legal observers, including those at Bilzin Sumberg and Hahn Loeser, describe it as a matter of when, not whether. The absence of case law makes risk-averse municipal attorneys nervous, and risk-averse municipal attorneys make city managers nervous, and nervous city managers do not sign procurement contracts for novel technology.

Every vendor in the space has responded to this anxiety the same way: human-in-the-loop. Blitz AI's Chris Prather practically begged the Business Observer to quote him on it. "Human-in-the-loop AI," he said. "The AI gives them a head start, makes them more efficient." The AI produces something Brittany Griffin, Blitz AI's principal planner, described as "a grocery list of items wrong that the human would have found in a few hours and the AI found in minutes." The human still signs. The human still decides. The AI is a faster pencil, not a replacement brain.

But explaining that distinction to a municipal attorney who has never purchased AI software and whose entire professional incentive structure rewards caution is a sales problem that technology alone cannot solve.

Where the momentum actually is

Austin's Janet Heit reported something telling: other cities are calling her. "I'm getting phone calls from other cities interested in wanting to know how we're doing this," she told Homes.com News. "I was at a conference in San Diego a week ago, and this was a big topic." Austin is now exploring expansion into commercial permits and technical reviews: plumbing, electrical, mechanical.

Pueblo County, Colorado, funded its Blitz AI deployment with a grant from the Colorado Department of Local Affairs, offering the tool free to applicants for the first year. Denver selected CivCheck in March 2026 with a target rollout for summer or early fall. Harris County, Texas, which contains Houston, approved an AI permitting program in November 2025, though no vendor has been selected and no timeline is firm.

Clariti, which acquired CivCheck, launched "AI Studio" in May 2026, essentially a consulting program offering 45-60-minute workshops to municipal teams curious about AI plan review. Clariti operates in 26 states. The fact that a workshop program for how-to-think-about-buying-AI is newsworthy in 2026 tells you where most of the market actually sits.

What this means if you are building or buying a home

If you are pulling a permit in Honolulu, Austin, or Naples, you are getting a faster, more consistent review process at no additional cost to you. The technology is mature enough that Austin ran 190 applications through it and received zero complaints.

If you are pulling a permit in the other 99.8% of American jurisdictions, your plans are sitting in a stack. A human reviewer will get to them when the queue clears, and that depends on staffing, budget cycles, vacation schedules, and whether the person who started your review finishes it or leaves midway through. A Virginia homebuilder cited by the Independent Institute, whose permit stalled because the assigned county staffer resigned and didn't hand off the file.

The technology to fix this costs $73 per permit. Your city chose not to buy it. Ask your city council why.

Limitations

The $2.4 billion estimate is my calculation based on Census permit volumes, reported AI time savings, and mortgage carrying costs. It assumes 1.2 million permits could benefit from AI pre-screening and a $2,000 average per-permit delay cost. Both figures involve simplifying assumptions: some permits face minimal delays regardless of technology, and carrying costs vary enormously by loan size and interest rate. The true figure could be higher or lower by a factor of two. The directional conclusion, that non-adoption costs billions, is robust to reasonable variation in inputs.

I could not independently verify vendor claims about accuracy rates. Clariti's "97%+ accuracy" for CivCheck is self-reported. Honolulu's 70% time reduction comes from Clariti's press release, corroborated by city official Dawn Takeuchi Apuna's quoted endorsement but not by independent audit. Austin's 50% figure comes from city staff, not third-party evaluation.

The political economy analysis relies heavily on one source, Scott Beyer's Independent Institute piece, and public choice theory generally. No municipal official I found on the record said "we don't want faster permitting because it would reduce our department's headcount." The incentive diagnosis is structural inference, not confession.