Sixty Plan Review Jobs Cost $5 Million a Year. The AI That Replaced Them Costs $920,000.
In March, Denver's Community Planning and Development Department signed a five-year, $4.6 million contract with CivCheck, an AI-powered plan review platform built by Clariti. It automatically flags missing documents, incomplete fields, and application errors before plans reach a human reviewer.
That same budget cycle, Denver cut 59 plan review positions from the department, dropping staff to 251.
Separate decisions, according to the city. But when you line up the contract and the cuts on the same spreadsheet, the arithmetic tells a different story.
What Nobody Put Side by Side
A plan reviewer in Denver's building department earns between $65,000 and $95,000 annually, depending on seniority, and once you add benefits, pension contributions, and overhead a fully loaded position runs roughly $85,000 per year. Fifty-nine of those positions cost the city approximately $5 million annually. CivCheck costs $920,000 a year, or 18.4 cents on every dollar those humans used to cost, absorbing the intake screening that occupied a significant chunk of 59 people's working hours for less than a fifth of their combined salaries.
But CivCheck does not actually review plans. It checks whether applications are complete, whether documents are attached, whether fields are filled in correctly. It is, to use the vendor's own framing, a "pre-submission" tool. Every set of drawings still gets evaluated by a human reviewer against every applicable building, zoning, and energy code. Every remaining reviewer still exercises the judgment that determines whether your addition meets setback requirements, whether your electrical panel has adequate capacity, whether your foundation design accounts for the soil conditions on your specific lot.
What changed is volume. Three hundred ten reviewers used to process Denver's permit queue. Now 251 handle the same load, a 23.5% increase in workload per person, offset by AI sorting the intake pile that used to chew through the first pass of everyone's morning, which raises the question of whether that trade is sustainable or whether it is simply the first round of cuts before the next budget cycle brings another. Early data is genuinely interesting.
Denver's First-Try Problem
Before CivCheck, Denver approved only 37% of permit applications on the first try. Nearly two out of three submissions got kicked back for missing documents, incomplete information, or errors that had nothing to do with the actual design of the building, because plans that might have been perfectly engineered still got bounced by a missing survey, a blank field on page 14 of a 22-page form, or a site plan that referenced an outdated parcel number.
Each rejection sends an applicant back to the end of the line, and each resubmission triggers another full review cycle that averages about two days for initial intake in Denver but stretches much longer for complex projects. Denver had already imposed a 180-day shot clock on permit decisions, with a promise to refund developers up to $10,000 in application fees if it missed the deadline, which was an admission that the system was buckling under the weight of its own bureaucratic friction.
CivCheck is supposed to push that 37% to 80%. That changes everything. If correction cycles that once dominated the department's workload shrink dramatically, fewer bounced applications mean fewer re-reviews, fewer staff hours burned on the same project cycling through the queue three times before anyone looks at the actual blueprints, and those 59 cut positions start looking less like a gap and more like a bet that the software will cover what the people used to.
Austin Already Has Six Months of Data
Denver is still in early deployment, but Austin, Texas, has been running an AI pre-check tool since late 2025 after a three-month pilot in 2023. Austin signed a three-year contract worth about $1.1 million annually with Archistar, an Australian company whose eCheck platform uses computer vision and machine learning to parse uploaded PDF plans against zoning regulations.
Results so far: 190 submissions, zero negative feedback, and review time for staff cut by roughly 50%. Janet Heit, Austin's chief administrative officer for Development Services, told Homes.com News that Archistar is "serving as a second set of eyes" for reviewers, "helping them double-check calculations," and Austin is now expanding to plumbing, electrical, and mechanical reviews while exploring commercial permits.
Here is the difference between Denver and Austin that should make every plan reviewer in the country pay attention. Austin deployed AI and kept its staff intact. Denver deployed AI and cut a fifth of its workforce. Same technology. Same vendor talking points about "preserving human judgment." Wildly different outcomes for the humans involved.
A Carrying Cost Nobody Calculates
NAHB published its 2026 regulation study in June, finding that government regulation at all levels now adds $131,734 to the average new home. That is 26.4% of a $499,500 sale price, up 40% from $93,870 in 2021. NAHB identifies "pure cost of delay" as $4,112 per home, split between development and construction phases.
That $4,112 figure is almost certainly too low for what permit delays actually cost a builder holding a construction loan. Here is the carrying math on the average new home: a construction loan covering 80% of the sale price at current rates around 7.5% generates a daily carrying cost of roughly $82. Initial permit review takes 22.9 days nationally, according to PermitPlace's analysis of 741 cities across 44 states, and most projects require two to three correction resubmittals, each adding a full review cycle, which means a typical residential project bouncing through two full review cycles spends 45.8 days in permit review alone.
If AI pre-check tools reduce correction cycles by 50%, as Clariti claims, total review time drops to roughly 34 days, saving 11.4 days per project at $82 per day in carrying costs, which works out to $935 per home. Multiply by roughly one million single-family housing starts per year nationally, and potential savings reach $935 million, though real adoption today covers maybe 10% of starts, putting actual savings closer to $94 million. Still enough to reshape who needs to be in the building department and who doesn't.
Twenty Cities and Counting
Deployment is spreading fast. Honolulu launched CivCheck for residential permits in December 2025, with commercial applications expected by mid-2026. Seattle's then-mayor Bruce Harrell signed an executive order directing all development applications through an AI pilot program, with full rollout expected this year. Naples, Florida, became the first city in the state to partner with Blitz AI for automated plan review. Pueblo County, Colorado, funded its own AI program through a state grant from the Colorado Department of Local Affairs. Harris County, Texas, approved an AI permitting initiative in November but has not selected a vendor.
Three companies dominate right now. Clariti's CivCheck handles intake screening and completeness checks across roughly 20 planning departments from Denver to Toronto, while Archistar's eCheck performs actual zoning compliance analysis using computer vision to parse architectural drawings, and Blitz AI targets smaller municipalities with bundled compliance and permitting integration. None do final plan review, but all three position themselves as tools that make human reviewers more efficient.
Julia Richman, Clariti's vice president of government relations, told HousingWire that cities using CivCheck see "somewhere along the lines of 70% time reduction on average." Not marginal. A 70% reduction in a process that currently averages 23 days nationally transforms a three-week wait into a one-week wait. It determines whether you hold a construction loan for an extra month. It determines whether a building department needs 310 people or can operate with 251.
What This Means for the People Who Read Blueprints
Plan review is skilled work. Not "skilled" the way people say it when they want to sound respectful before explaining why a job is about to disappear, but actually, genuinely skilled. A competent reviewer knows the International Residential Code, the local amendments that modify it in ways that sometimes contradict the base document's intent, the energy code overlay that adds another 200 pages of requirements nobody in the public reads, and the site-specific conditions that make every lot different from the one next door. They catch undersized beams and flag egress violations. They stop the retaining wall that needs engineering the applicant never bothered to commission. NAHB's own study found that building code changes over the past decade added $40,288 per home, and that complexity is exactly what makes competent reviewers hard to replace while also being exactly what current AI tools do not touch.
What AI replaces is paperwork triage: checking whether all 22 pages of the application are present, verifying the surveyor's stamp is current, confirming the lot dimensions on the site plan match the parcel record. Important work, but not work that requires someone who understands load paths and fire separation distances.
Municipal budget offices may not make that distinction. If AI handles intake and correction cycles drop by 50%, total work hours decline even as the complexity of each individual review stays the same or increases because codes keep getting harder. A budget analyst looking at hours-per-permit sees efficiency gains and nothing else. What they do not see is that remaining reviewers spend a higher percentage of their time on the hardest, highest-stakes reviews, the ones where no AI is a factor and human expertise is the only thing standing between an applicant and a structural failure that shows up three years later when the deck collapses during a birthday party.
Denver's 59 cut positions may be the first visible data point in a pattern that plays out city by city over the next several years. Builders want speed, cities want lower costs, and AI vendors are offering both. Plan reviewers who remain will be expected to do more complex work with fewer colleagues, supported by software that handles the routine screening their former coworkers used to do, which raises the question that nobody in any of these cities has answered publicly: when the next round of code changes makes every review harder, will the departments that deployed AI hire to match the complexity, or will they point to the efficiency numbers and cut again?
Whether that constitutes augmentation or replacement depends entirely on whether the departments that deploy these tools hire to maintain the hard skill base or simply pocket the savings. So far, the early indicators are split. Austin added AI and kept its team. Denver added AI and cut a fifth of it. Watch what your city does next.
Limitations
This analysis uses estimated fully loaded salary costs for Denver plan reviewers; the actual figure could vary between $75,000 and $95,000 depending on the seniority mix of eliminated positions, which Denver has not publicly disclosed. Carrying cost calculations assume an 80% loan-to-value construction loan at 7.5%, reflecting current market conditions but varying by lender and borrower. Clariti's claimed 50% reduction in review cycles and 70% time savings come from vendor data, not independent verification. PermitPlace's 22.9-day national average represents published department guidelines, not actual processing times, which the company notes can run two to five times longer. Austin's dataset of 190 submissions is small, and six months is not long enough to evaluate whether efficiency gains persist at scale or across more complex project types.
Strongest counterargument: Denver's position cuts may reflect broader municipal budget pressure unrelated to AI. Denver had been working toward permitting reform for years before CivCheck, cutting single-family and duplex processing times by 45% since 2023 through process improvements alone. CivCheck and the staffing cuts happened in the same budget cycle, but correlation is not causation. It is entirely possible that Denver would have eliminated those positions regardless, and that CivCheck was deployed specifically because the cuts made automation necessary rather than the other way around. If that is the case, the AI is not replacing workers; it is compensating for their absence. Cold comfort for the 59 who left, but a meaningfully different policy story.