San Jose issued 488 ADU permits in 2024. That number sounds reasonable until you learn how many applications it took to produce 488 approvals. According to the city's own planning department data, cited in a StateScoop report on the city's new AI pilot, more than 90 percent of ADU permit applications are returned to applicants because of missing or incorrect information. Not rejected. Returned. Sent back with a list of deficiencies that most homeowners read the way most people read IRS instructions: with growing panic and diminishing comprehension.
Into this environment, San Jose is deploying CivCheck, an AI plan review tool acquired by Clariti in 2025, which claims 97 percent accuracy on compliance checks and an 80 percent reduction in approval times. Lancaster, California, became the first U.S. city to adopt Labrynth's AI-powered permitting platform last September, with Mayor R. Rex Parris calling it a "national model for smart governance." And statewide, AB 1332 mandated that every California jurisdiction provide six pre-approved ADU floor plans by April 2025, with 30-day approval timelines for permits that use those plans.
California is attacking the ADU permitting problem from three directions simultaneously. All three directions point at the same target: making the review process faster. None of them address why 90 percent of applications never make it to review in the first place.
Where Applications Actually Die
A homeowner in San Jose who wants to build an ADU faces a sequence of requirements that begins simply and becomes opaque quickly. Zoning verification. Setback calculations. Lot coverage ratios. Utility capacity confirmation from San Jose Water and PG&E. Soil reports if the parcel is in an expansive soil zone. A Title 24 energy compliance report from a HERS rater. Structural engineering for any unit over 400 square feet. Grading and drainage plans if the lot slopes more than 10 percent. Fire sprinkler requirements if the ADU exceeds 1,000 square feet or is within a certain distance of a property line.
Most homeowners hire an architect or designer to produce the construction drawings. But architects produce construction documents, not permit applications. Between the finished drawings and a complete permit application sits a documentation gap that nobody owns: the Title 24 energy report the architect assumed the homeowner would order separately, the soils report that applies to the specific lot (not the pre-approved plan's generic assumptions), the utility will-serve letters that require separate applications to separate agencies with separate timelines.
A pre-approved plan under AB 1332 eliminates the design review step. It does not eliminate any of the site-specific requirements that generate 90 percent of the returns. You can use a pre-approved 800-square-foot, two-bedroom ADU plan from the city's gallery and still have your application returned because you didn't include a soils report, or because PG&E hasn't confirmed electrical capacity for a second dwelling on your lot, or because your property line setback is 3.8 feet instead of the required 4 feet and nobody noticed until a plan checker measured it against the parcel map.
What the AI Actually Speeds Up
CivCheck and similar tools work by parsing construction drawings against code requirements, flagging non-compliant elements, and generating compliance reports. Clariti's marketing materials cite 80 percent faster approvals, which, if accurate, compresses the plan review step from weeks to days.
On a complete application.
For the 10 percent of San Jose ADU applications that arrive complete, AI review is a genuine improvement. Shaving two weeks off a six-week review cycle saves the homeowner carrying costs on their construction loan, preserves their contractor's scheduling window, and frees the building department's limited plan reviewers to handle the applications that AI flagged for human judgment.
For the 90 percent that arrive incomplete, AI plan review is irrelevant. An application missing its Title 24 report never reaches the plan review stage. An application with incorrect setback calculations gets returned before anyone, human or algorithmic, evaluates the structural drawings. Speeding up the review of complete applications is optimizing a step that most applicants never reach.
Lancaster's Labrynth Bet
Lancaster's approach is more interesting than San Jose's because Labrynth's platform includes applicant-facing features: guided workflows that walk submitters through requirements before they file, automated checklists that flag missing documents pre-submission, and a "Red Tape Index" that Labrynth is building to benchmark permitting speed across 500-plus cities. In theory, this addresses the upstream problem. If the AI tells you before you submit that you need a soils report, a utility will-serve letter, and a HERS rater certification, you might actually get those things before filing.
In theory. Lancaster announced the partnership in September 2025. Eight months later, no public data exists on application completeness rates before and after deployment. No comparison of return rates. No measurement of whether the guided workflow actually reduced the documentation gap or whether homeowners abandoned the process entirely when the checklist revealed how much pre-work they needed to do. Labrynth's website showcases the partnership prominently. What it does not showcase is outcomes.
This is not a criticism of the technology. It is a criticism of the narrative. Calling an AI permitting tool a "national model for smart governance" before publishing a single performance metric is marketing, not governance.
AB 1332: Six Plans and a Mandate
California's pre-approved plans mandate is the most ambitious of the three interventions. AB 1332 required every local jurisdiction to establish a preapproval program by January 1, 2025, provide at least six permit-ready ADU floor plans (studio, one-bedroom, and two-bedroom) by April 1, 2025, post those plans online at no cost, and approve permits using those plans within 30 days.
Compliance is uneven. San Jose, Hayward, Santa Clara, and Placer County have functional preapproval galleries. Dozens of smaller jurisdictions have posted the minimum six plans on a webpage and called it compliance. Whether those plans are actually buildable on a given lot in a given jurisdiction depends on site conditions that the pre-approved plan, by definition, does not address because it was designed to be non-site-specific.
A pre-approved plan is a set of construction drawings that have passed code review in the abstract. Your specific lot has a specific slope, specific soil type, specific utility infrastructure, specific setback geometry, and specific drainage conditions that the pre-approved plan cannot anticipate. When the city approved the plan, it confirmed that the structure as drawn meets building code. It did not confirm that the structure as drawn fits your property.
So you buy a pre-approved plan for $3,000. You still need a site-specific soils report ($2,000 to $5,000). A Title 24 energy compliance report ($500 to $1,500). A survey to confirm setbacks ($800 to $2,000). Possibly a grading plan ($1,500 to $4,000). If the pre-approved plan's foundation design doesn't match your soil conditions, you need a structural engineer to modify it ($2,000 to $6,000), at which point it is no longer the pre-approved plan and the 30-day approval timeline no longer applies.
Who Builds Without Permits
A computer vision study published through the American Planning Association analyzed satellite imagery in San Jose between 2016 and 2020 and identified unpermitted ADU construction across the city. Of the structures identified, 78 percent were located in less affluent, more diverse neighborhoods.
That statistic should reframe every conversation about AI permitting tools. Homeowners in wealthier neighborhoods hire architects, engineers, and permit expediters who navigate the process professionally. Homeowners in less affluent neighborhoods cannot afford the $7,000 to $20,000 in pre-application professional services that separate a complete application from an incomplete one. So they build without permits, accept the legal risk, and hope nobody reports them. Or they abandon the idea entirely.
AI that makes the formal permitting process 80 percent faster serves the homeowners who were already going to navigate it successfully. It does nothing for the homeowners who can't afford to assemble a complete application, who are disproportionately the people in neighborhoods where ADUs would provide the most housing benefit.
The CalHFA Problem
California created one program that actually addressed the cost barrier: the CalHFA ADU Grant Program, which provided up to $40,000 for pre-development costs including architectural design, engineering, permitting, and site preparation. Exactly the expenses that separate a homeowner who can assemble a complete permit application from one who cannot.
That program is paused. Funding exhausted. No timeline for replenishment.
So California is simultaneously mandating pre-approved plans (which don't eliminate site-specific costs), deploying AI to review complete applications faster (which helps 10 percent of applicants), and defunding the one program that helped homeowners afford the pre-application work that determines whether their application is complete in the first place.
What Would Actually Help
San Jose's CivCheck pilot is pointed in a more useful direction than its marketing suggests. CivCheck's stated purpose is "pre-screening," not plan review. If the tool can tell a homeowner before they submit that their application is missing a soils report, a Title 24 certificate, and a utility confirmation letter, that is genuinely useful. Not because it speeds up review, but because it prevents the submit-return-resubmit cycle that wastes months of homeowner time and staff resources on both sides.
A more ambitious version would go further. An AI tool that reads a homeowner's address, pulls the parcel geometry from the assessor's database, checks the soil type against USGS data, identifies whether the lot sits in a flood zone or fire hazard severity zone, calculates setbacks from the property line survey, and generates a complete list of every document required for that specific lot before the homeowner spends a dollar on professional services. Not a generic checklist. A parcel-specific application roadmap.
Nobody has built this. Symbium's zoning pre-check comes closest, confirming what you can build on a given lot. But it stops at zoning. Between zoning confirmation and a buildable permit application, there are six to twelve additional requirements that vary by parcel, and no tool maps all of them for a homeowner before they commit $3,000 to $20,000 in professional fees.
Labrynth claims applicant-facing guidance. If it actually delivers parcel-specific pre-application roadmaps instead of generic checklists, Lancaster's pilot could be the first meaningful AI intervention in the ADU pipeline. But until performance data is published, "could be" is carrying a lot of weight in that sentence.
If You Are Planning an ADU
Start with your city's pre-approved plan gallery if one exists. Check whether your lot's soil type, slope, setbacks, and utility capacity are compatible with the plan before you purchase it. Budget $7,000 to $20,000 in pre-application professional costs (soils, survey, Title 24, engineering) on top of the plan cost. Call your city's building department and ask for the ADU-specific checklist; most jurisdictions have one, and it will tell you more about what you actually need to submit than any AI tool currently available.
If you are in San Jose, check whether the CivCheck pre-screening tool is available for public use. If it tells you what your application is missing before you file, use it. If it only reviews complete applications faster, it will not save you the months you would otherwise spend in the return cycle.
Do not assume that a pre-approved plan means a fast permit. It means a fast design review. Every site-specific requirement still applies, and site-specific requirements are what generate 90 percent of returns.
Limitations
San Jose's 90 percent return rate is cited by StateScoop and GovTech from city planning department data; the original dataset has not been independently audited and may include returns for reasons beyond missing documents (e.g., fee non-payment, incorrect zoning applications). CivCheck's 97 percent accuracy claim comes from Clariti's acquisition announcement, not third-party testing. Lancaster's Labrynth partnership has produced no published outcome data as of May 2026. CalHFA grant program pause is confirmed on the CalHFA website but the agency has not announced whether or when funding will be replenished. ADU costs vary significantly by location, lot conditions, and scope; the $150K-$350K range reflects San Jose market conditions and may not apply statewide. AB 1332 compliance rates across California's 482 cities and 58 counties have not been comprehensively tracked by any state agency.