AI read it first. Did anyone keep the original?

AI plan review fails a Florida city's six-month test

St. Petersburg, Florida, spent about six months and more than 20 vendor demonstrations testing AI that checks building plans against code. Staff gave vendors real construction plans. The software struggled with rules that need judgment, such as accessibility and door placement, and with deciding which rule wins when two conflict. It could flag dozens of supposed problems, leaving staff to sort the real ones. The city will now use software only to check that applications are complete.

For managers: an AI that flags dozens of problems gives your team a new job, sorting real findings from false ones, so time that sorting before you count any hours saved.

Source: stpetecatalyst.com

Philadelphia tenants push rules for AI that rewrites repair requests

Tenant organizers and Philadelphia City Council members are drafting legislation to limit how landlords use AI for repair requests and rent arrears, WHYY's PlanPhilly reported. Their complaint: when software rewrites or misreads a complaint, a maintenance worker can arrive without the right tools or fix the wrong thing, and the tenant may lose an accurate record for court. They want every request stored exactly as typed. Landlords say the software takes repetitive tasks off staff.

For managers: whatever tool takes in reports at your sites, keep the original words next to any AI summary, because the summary is not a record of what someone actually reported.

Source: thecooldown.com

AI assistant brings decades of nuclear repair history to engineers

Nearly all 94 nuclear reactors in the United States have been offered AI tools this year, and most have taken them up, IEEE Spectrum reports. One, NIVA, built by Atomic Canyon with industry groups, searches hundreds of thousands of operating experience reports, so an engineer with a failing pump can see how other plants fixed it, with the documents behind each answer. A separate pilot uses AI to flag weld inspection data, with a person reviewing every recommendation.

For managers: insist that any assistant shows its sources, as these do, because an answer your engineer can trace is one they can check before acting on it.

Source: spectrum.ieee.org

Service robots take over delivery runs at two Singapore hotels

Novotel and Mercure Singapore on Stevens, two hotels with 772 rooms between them, agreed on October 13 to roll out service robots from YYForce for guest deliveries, food and beverage runs, scheduled cleaning and concierge support. A model prepared before rollout projects about 10% lower monthly costs in those operations, with the robots' own cost counted. Before any robot moves, the hotels need lift integration, route mapping, network coverage and guest arrival notices.

For managers: most of that prep list belongs to engineering and IT, not housekeeping, so bring them into a robot pilot from the first meeting and check the real savings against the 10% projection.

Source: marketscale.com

One number

4%

of industrial companies surveyed have made human oversight or approval mandatory for decisions AI supports, against more than 70% that say they are confident in their safety and data policies. Industrial software maker IFS surveyed 2,275 executives and frontline workers at companies with more than 500 staff in manufacturing, energy, construction, transport and other industries. Only 11% had validated AI outputs for safety-critical work. Confidence is running well ahead of the checks.

Try this Monday

Pick one place where software now reads or rewrites what your team reports: a maintenance request, an incident note or an inspection flag. Pull three recent entries and compare the original words with what the system kept. If the original is gone, or a flag was cleared and nobody recorded who cleared it, fix that this week.

Worth a look

Fewer overtime hours, says the vendor. Restaurants using Toast's scheduling software averaged 41% fewer overtime hours per employee than those not using it, the company says.

Acting on a wrong answer. 34% of organizations had acted on an inaccurate decision or output from an AI agent, in a survey of 417 risk and audit leaders by governance software company Optro.

Homes from a robot factory. Homebuilder D.R. Horton will use Promise Robotics' robotic factories to produce more than 1,250 homes in the United States.