The Guy Who Knows Why That Pipe Is There Is Retiring. Nobody Recorded Him.

Eddie Morales has been a plumber for 34 years. He can walk into a basement he’s never seen and tell you, within two minutes, where the original installer cut corners. He doesn’t measure. He reads the room—the angle of the vent stack, the staining pattern on the concrete, the way the copper sweats near the water heater. He learned it from a guy named Phil, who learned it from his father-in-law, who learned it in Korea building barracks.

Eddie turns 62 in April. He’s not training a replacement.

Nobody asked him to.

The numbers nobody likes to say out loud

NCCER estimates 41% of the current US construction workforce will retire by 2031. That’s not a projection you can plan around. That’s a five-year countdown. Over 20% are already past 55. The median age for a carpenter is 42.3, for an electrician it’s 43.1, and for the plumbers and pipefitters keeping water flowing through every home in the country, it’s 42.6. BLS says the industry needs to add 500,000 workers this year alone just to keep pace with demand.

Meanwhile, the apprenticeship pipeline is collapsing. The UK’s NOCN data shows a 47% dropout rate for construction apprentices, with only 8,620 reaching end-point assessment in 2022/23 against an industry need of 96,000 annually. US numbers are better but not good: registered apprenticeship completion rates in construction hover around 50–56%, according to the Department of Labor’s own dashboard. For every two who start, one walks away.

The shortage everyone talks about is bodies. The shortage nobody talks about is knowledge.

What you can’t write in a manual

There’s a term in organizational psychology for this: tacit knowledge. The stuff that lives in someone’s hands and eyes, not on a spec sheet. A roofer who can feel when the substrate is wrong through his boots. A framer who knows that a particular lumber supplier’s 2x6s run a sixteenth short and adjusts without measuring. An HVAC tech who hears a compressor and knows it has eight months left.

You cannot Google this. You cannot train it in a classroom. It takes 10,000 hours of someone patient enough to let you watch, and then letting you fail, and then showing you again.

Phil didn’t write a manual for Eddie. He stood next to him for six years.

The AI play: capture before it’s gone

A handful of companies are trying to solve this, and I want to be honest about what’s working and what isn’t.

XOi Technologies out of Nashville built something genuinely useful: a video-first knowledge platform for field service techs. An HVAC technician opens the app, points the camera at a unit, and XOi’s AI identifies the equipment, pulls relevant documentation, and—crucially—surfaces video clips from other techs who’ve worked on the same model. 120,000+ pieces of indexed content. Daikin, one of the world’s largest HVAC manufacturers, rolled it out across their contractor network in 2025. The coaching feature lets a veteran tech walk a less experienced one through a repair via live video, recorded and searchable after.

That last part matters. When the veteran retires, the videos stay.

Procore’s AI takes a different angle. Their document intelligence agents can answer natural-language questions about a project by crawling submittals, RFIs, daily logs, and change orders. Ask “why was the plumbing rerouted on building C?” and instead of tracking down the superintendent who left the company in 2024, you get the RFI trail, the engineer’s response, and the photo documentation. $339 million in quarterly revenue. 95% customer retention. They’re embedding institutional memory whether their clients realize it or not.

Spot AI does video analytics for construction sites—primarily for security and safety. But the same cameras that catch someone not wearing a hard hat could, in theory, capture technique. How a skilled ironworker ties rebar at a corner. The hand position a veteran uses for overhead welding. Nobody’s productized this yet. I think someone will.

What’s not working

Most “knowledge management” in construction still means a SharePoint folder no one opens and a lessons-learned meeting no one attends. According to a UVM analysis of AI knowledge capture strategies, the most common failure mode is asking people to document what they know in addition to their actual job. They won’t. They’re already working 50-hour weeks.

The approaches that work are passive: video that records automatically, AI that indexes it without the worker doing anything extra, search that retrieves it when someone else needs it. The burden has to be zero on the expert and low on the learner.

AR-assisted training gets a lot of hype. Tilt Five, Microsoft HoloLens in construction, Trimble’s XR hardhat. But I spent a morning watching a 58-year-old steamfitter try to navigate a HoloLens interface on a job site. He took it off after four minutes. “I’m not wearing a video game to do my job.” Fair enough.

The part that keeps me up

My dad was IBEW Local 98 in Philadelphia for 31 years. When he retired, his foreman threw him a party and gave him a plaque. Nobody sat him down with a camera. Nobody asked him to explain why he routed conduit a certain way through the hospital wing he’d wired, or how he knew which junction boxes would overheat in summer, or what he’d learned in three decades about keeping apprentices alive around 480 volts.

All of that is gone now. It left in his truck on a Friday afternoon.

We’re doing this 4.8 million times over the next five years. Every one of those retirements is a library burning. Some of them are small libraries. Some of them are the Library of Alexandria.

XOi and Procore are chipping at the problem, and they deserve credit for it. But the scale of what’s being lost makes the effort look like photographing a few pages of a book while the building’s on fire.

Eddie Morales knows things about residential plumbing that no YouTube video will teach you. In two years, the only thing he’ll be teaching is his grandkids how to fish.