Industries / Government, education and community / Government
Permits and licenses
Checking that a building, a business or a driver meets the rules before the work starts, and issuing the permit, license or card that says so.
5
Milestones achieved, from 1666 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1666
Surveyors to enforce rules
After the Great Fire of London, Parliament's Act for rebuilding the City set four sorts of house, ordered that the outsides of all new buildings be "of Bricke or Stone", and let the City appoint surveyors, sworn to see the rules "well and truely observed". A house built any other way was a nuisance that the Court of Aldermen could order pulled down.
Credit
Parliament of England; the Lord Mayor, Aldermen and Common Council of the City of London
2005
A company at one counter
Portugal's Decree-Law 111/2005 created Empresa na Hora (literally "company on the hour"): the founders choose a name from a pre-approved list, sign the company's articles and have it entered in the commercial register at a single counter, in a single visit. The service still runs under the Institute of Registries and Notary (IRN).
Credit
Government of Portugal; Instituto dos Registos e do Notariado
2021
A permit ranking ends
The World Bank's Doing Business report measured, country by country, the procedures, time and cost to build a warehouse, including its licenses, permits, inspections and utility connections, and ranked the results. After data irregularities in the 2018 and 2020 editions were reported inside the Bank in June 2020, it paused the report, ordered reviews and audits, and on 16 September 2021 discontinued it.
LessonA number that ranks governments needs controls on how it is produced, or it stops being trusted.
Credit
World Bank Group
2022
Driver's licenses on phones
On 23 March 2022 the US Transportation Security Administration (TSA) began accepting Arizona driver's licenses and state ID cards held in Apple Wallet at Phoenix Sky Harbor airport, at first only for travelers in its PreCheck program. TSA now accepts digital IDs from Apple Wallet, Google Wallet, Samsung Wallet or a state's own app at more than 250 airports, according to its own page. TSA sees only the details it needs, and travelers must still carry an acceptable physical ID.
Credit
Transportation Security Administration; Arizona and the other states that issue the IDs; Apple
2025
One submission, joint review
Singapore's CORENET X, soft-launched on 18 December 2023, replaced separate filings to each agency with three "gateways" (design, construction and completion): the project team coordinates the design first, the agencies review it jointly and send one consolidated reply, and large projects submit a 3D building model in the IFC-SG format. It became mandatory for new projects of 30,000 m² or more on 1 October 2025, and for all new projects on 1 October 2026. The first plan had made it mandatory on 1 April 2025; the agencies moved the dates after firms said they needed more time.
LessonChanging how a whole trade files its plans takes live projects to learn on, and longer than planned.
Credit
Building and Construction Authority and Urban Redevelopment Authority, Singapore
Watch for
AI pre-check of house plans
Honolulu's Department of Planning and Permitting says that since it launched CivCheck, an AI tool from the software company Clariti that checks residential plans against the city's code before they are filed, on 8 December 2025, average review time fell from 73 days to 32.5 and review cycles from 3.4 to 1.4. The figures come from Clariti's performance report and compare 19 permits that used the tool with 17 that did not; a city spokesperson called 19 "a relatively small sample". In a separate test, Seattle found CivCheck's completeness checks 87% accurate and its design compliance checks 92% accurate, but that a pre-check saves the applicant a review cycle only if it catches every correction a reviewer would make.
Achieved whena review run by neither the city nor the vendor, over a full year of permits with a comparable control group, confirms shorter review times.
Claimed by
Honolulu Department of Planning and Permitting and Clariti, July 2026
AI guidance before applying
Bellevue, Washington, and the startup Govstream.ai report that when 276 real 2025 residential permit applications were replayed through Govstream.ai's intake assistant, the share that would have arrived complete rose from 22% to 67%, and the assistant identified 96% of the required documents. This was a back test, not live use. Separately, city staff estimated that an internal assistant saved them 152 hours in one month of 2026.
Achieved whenapplications filed live with the assistant are compared with those filed without it, by someone other than the vendor.
Claimed by
City of Bellevue and Govstream.ai, August 2026
Open
01
A public plan-check test
An open test set for AI tools that check house plans against the building code: made-up plan sets for small homes and backyard cottages (accessory dwelling units), each with known code errors seeded in, and a score for each tool on the share of errors it finds (coverage) and the share of its flags that are right (accuracy).
The briefCredit · unclaimed02
A permit-question test
A public test of the AI assistants that residents ask about permits: 300 short project descriptions ("a 120-square-foot garden shed", "replace a water heater") for three cities, each with the answer the city's own published rules give: whether a permit is needed, and which kind.
The briefCredit · unclaimed
Open 01 · the brief
A public plan-check test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open test set for AI tools that check house plans against the building code: made-up plan sets for small homes and backyard cottages (accessory dwelling units), each with known code errors seeded in, and a score for each tool on the share of errors it finds (coverage) and the share of its flags that are right (accuracy).
Why it is still open
The results so far come from cities testing one vendor each on their own permits. Seattle measured accuracy and found that coverage decides whether an applicant saves a review cycle, and that checking a design against the "thousands of requirements" in its codes is the hard part. The nearest research test, ARCHER (July 2026), scores AI-written rule checks on 3D building models (BIM), not on the 2D drawings most house permits use. In searches of research papers and GitHub I also found code-checking prototypes and hobby projects, but no public test set of house plans with an answer key.
Built with what exists
Seattle's open Plan Comments dataset, which holds the comments reviewers made on building plan sets, to choose the errors reviewers flag most often; any drawing program to draw the plans; AI models that read drawings, such as Claude, Gemini and GPT, as the tools under test. (see the Note "Seven more flagged")
First step
Download Seattle's Plan Comments data and count the 50 most frequent corrections on single-family and accessory-dwelling permits.
Done looks like
100 made-up plan sets holding 1,000 seeded errors, an answer key that cites the code section for every error, and published coverage and accuracy for at least four tools, each run three times.
Open 02 · the brief
A permit-question test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of the AI assistants that residents ask about permits: 300 short project descriptions ("a 120-square-foot garden shed", "replace a water heater") for three cities, each with the answer the city's own published rules give: whether a permit is needed, and which kind.
Why it is still open
Portland had its experts classify about 200 made-up questions to test its own permit-appointment chatbot, and Bellevue back-tested its vendor's assistant on its own permits. I found neither test set published. Bellevue found that about one in five tree-removal applications never reached issuance, often because the work was filed under the wrong permit type or needed no permit at all.
Built with what exists
each city's published list of work that needs no permit; general AI assistants and any public city chatbot; repeated runs of each question to see whether the same answer comes back every time. (see the Note "All twenty times")
First step
Turn one city's list of work that needs no permit into 50 yes-or-no questions with answers.
Done looks like
300 questions whose answers cite the city rule, each assistant asked each question five times, and a published table of the share answered correctly on all five tries, by city and by assistant.