Industries / Construction and property / Real estate / Property management
Repairs
Repairs is the work of taking a tenant's report that something is broken, getting it fixed, and planning the large replacements a building will need.
5
Milestones achieved, from 1961 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1961
Landlords must keep homes repaired
In England and Wales the law writes a promise into every lease of a home for less than seven years granted from 24 October 1961: the landlord must keep the structure and exterior in repair, and keep the water, gas, electricity, sanitation and heating installations in repair and working order. The promise now sits in section 11 of the Landlord and Tenant Act 1985.
Credit
UK Parliament
1970
Habitability becomes a lease term
Tenants of a Washington, D.C., apartment building, sued for unpaid rent, offered to prove about 1,500 breaches of the city's housing regulations; the trial court refused the evidence, and the local appeals court held that the landlord had no duty to keep the building to the regulations. In Javins v. First National Realty Corp., the federal appeals court held that every lease of a city flat covered by those regulations carries a warranty of habitability, a promise that the home meets the housing code, and that a breach is a defence when the landlord sues for rent.
LessonOnce repair became part of what the rent pays for, a landlord's neglect could cost the landlord the rent.
Credit
US Court of Appeals for the District of Columbia Circuit, opinion by Judge J. Skelly Wright
2008
Thirty-year repair plans
In June 2008 Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT) published a standard form and guideline for the long-term repair plans of condominium buildings, which owners' associations use to set each owner's monthly payment into a repair reserve. The current version asks for a plan of at least 30 years that includes two major repair projects, the first usually about 12 to 15 years after completion. A second guideline, on reserve funds, followed in April 2011.
LessonReserve payments that start low and step up later become hard to agree as owners age, so MLIT now steers buildings toward equal payments from the start.
Credit
MLIT
2018
A backlog cleared on paper
The New York City Housing Authority (NYCHA), the largest public housing landlord in North America, announced on 2 January 2014 that it had cut its open repair requests, called work orders, from 423,000 to 106,000. A complaint filed by federal prosecutors on 11 June 2018 said much of the cut came from counting changes: from February 2013 staff could close a request by recording that the resident was not home, which closed nearly 200,000 requests in 2013 without the work being done. In the proposed consent decree filed the same day, NYCHA admitted that after it removed mold from apartments, the mold came back at least 30% of the time.
LessonA count of open work orders measures what is written down, not what is fixed.
Credit
US Attorney's Office for the Southern District of New York
2025
Repairs top tenant complaints
England's Housing Ombudsman, which rules on complaints from social housing tenants, reported in May 2025 that repairs and maintenance made up 45% of the complaints it decided in 2024–25, and that its investigations of them had risen 474% since 2019–20. It found maladministration, a failure by the landlord, in 73% of such cases in 2023–24, against 37% in 2019–20. Citing the English Housing Survey, whose surveyors inspect a random sample of homes, it noted that 9.3% of housing association homes were non-decent while landlords reported 0.6% to their regulator.
LessonA landlord that relies on its own records can miss most of its failing homes.
Credit
Housing Ombudsman Service
Watch for
AI fixes one in four
AppFolio, a US maker of property management software, says its Maintenance Performer, an AI agent that talks with residents, reviews their photos and walks them through simple fixes before it writes a work order, "resolves one in four work orders directly with the resident." The announcement gives no definition of resolved, no period and no sample size.
Achieved whenAn outside party checks a sample of the requests marked resolved and reports how many stayed fixed, with no repeat request for the same fault within 30 days.
Claimed by
AppFolio, September 2026
Open
01
Safe self-fix test
A public test of the AI agents that walk tenants through a fix before a technician is sent. Write 200 scripted repair conversations. In half, a tenant can safely solve the problem: a running toilet, a smoke alarm chirping for a new battery, a thermostat with flat batteries, a blocked sink. In the other half the only right answer is to stop and send help at once, drawn from Victoria's legal list of urgent repairs: a gas leak, a dangerous electrical fault, a burst water service, flooding. A person or a second AI model plays the tenant from the script, and each agent is scored. No one touches real equipment; every conversation is simulated.
The briefCredit · unclaimed02
Code the ombudsman's repair cases
A public table coded from the Housing Ombudsman's published decisions on repairs. For each case, record the repair, the date it was first reported, the landlord's own target, the date it was fixed, the number of visits, and what went wrong, such as no record kept, missed appointments or no inspection. Then publish how long each kind of repair took and the most common causes of delay.
The briefCredit · unclaimed
Open 01 · the brief
Safe self-fix 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 agents that walk tenants through a fix before a technician is sent. Write 200 scripted repair conversations. In half, a tenant can safely solve the problem: a running toilet, a smoke alarm chirping for a new battery, a thermostat with flat batteries, a blocked sink. In the other half the only right answer is to stop and send help at once, drawn from Victoria's legal list of urgent repairs: a gas leak, a dangerous electrical fault, a burst water service, flooding. A person or a second AI model plays the tenant from the script, and each agent is scored. No one touches real equipment; every conversation is simulated.
Why it is still open
AppFolio says its agent resolves one in four work orders with the resident (see Watch for) but publishes no safety figures. An open-source skill that turns an AI agent into a maintenance coordinator lets the operator set how hard it pushes troubleshooting on a scale of 0 to 5, and publishes no test of the result. Public projects that grade AI answers to home-repair questions, one of them on six failure modes including safety, score single answers, not conversations in which the right move is to stop. LandlordBench, a dataset posted in May 2026, has people and AI models choose whether to refuse, delay or pay for a repair, and does not test hazards.
Built with what exists
The open maintenance-coordinator skill run in an agent harness; trial accounts of commercial maintenance agents; low-cost models such as Claude Haiku 5.5 or Meta's Muse Ember to play the tenant (see the Note "The jobs agents repeat"); a spreadsheet for scoring.
First step
Write 20 scripts, 10 safe and 10 from Victoria's urgent list, and run them through the open skill with one model.
Done looks like
200 scripts published with their right answers, and at least three agents scored on the share of safe problems solved, the share of hazards passed straight to a person, and the number of conversations in which the agent told the tenant to do something unsafe, each scored by two reviewers with their agreement rate.
Open 02 · the brief
Code the ombudsman's repair cases
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table coded from the Housing Ombudsman's published decisions on repairs. For each case, record the repair, the date it was first reported, the landlord's own target, the date it was fixed, the number of visits, and what went wrong, such as no record kept, missed appointments or no inspection. Then publish how long each kind of repair took and the most common causes of delay.
Why it is still open
The Ombudsman has published its decisions since December 2020, naming landlords but not residents. Its May 2025 report gives totals, such as the 474% rise in repair investigations, not timings case by case. Searches of GitHub and Crossref in November 2026 found no coded set of the decisions.
Built with what exists
The decisions pages, saved where the site's terms allow; Gemini Decision 1, which returns a probability for each of up to 1,000 yes-or-no and choice questions about one document in a single request, for a first pass (see the Note "A thousand questions, one pass"); two people checking a sample; Python with pandas for the tables. Do not try to identify residents.
First step
Save 50 decisions about leaks, code them by hand, and revise the list of codes where the two coders disagree.
Done looks like
At least 1,000 repair decisions coded, a 10% sample coded by two people working apart with their agreement rate, and a published table of median days from first report to repair for each kind of repair.