Twenty-five students, one trainer, a lab full of half-wired panels. That was the ratio Eric Wu walked into at the Aviation Institute of Maintenance, and it is the ratio that sold him on his next company. One instructor cannot stand behind twenty-five beginners at once. Some of those beginners will not raise a hand when they are lost. So Wu's startup put the instructor inside a pair of glasses.
Navigate.AI launched in May with $25 million and a pitch simple enough to fit on a hard hat sticker: an AI copilot that watches construction workers through their phones and Meta's smart glasses, then coaches them the way the best journeyman on the crew would: step-by-step guidance, answers pulled from specs and manuals with sources cited, and quality control that flags defects before the crew leaves the site. Its first believer is not a trade school but Lennar, the country's second-largest homebuilder, which is deploying the copilot across its entire organization, from construction managers down to skilled and semi-skilled trade workers.
I cover the people who build houses for a living, and I have learned to read a funding announcement the way a framer reads a load table: for what it is actually carrying. This one is carrying $25 million from Lennar, Invitation Homes, Tishman Speyer, and the founders of DoorDash, which means some of the biggest landlords and builders in America just bought a front-row seat to film the American tradesman at work. At the launch party, nobody asked the question every worker should ask before putting the glasses on. When the camera learns your craft, who owns what it learned?
The Arithmetic of the Exodus
Methodology first, because the numbers are doing heavy lifting. Associated Builders and Contractors estimates the industry must attract 349,000 net new workers in 2026, rising to 456,000 in 2027, and its chief economist says "a majority" of this year's demand comes from retirements rather than growth (ABC, January 2026). Majority means more than half, so 174,500 is a conservative floor for retirement-driven openings, not a point estimate. Against that, ABC counts about 40,000 completers a year from federal and state registered apprenticeship programs. Divide 40,000 by 174,500 and the pipeline replaces less than a quarter of one year's retirements, before a single allowance for growth, quits, or injuries. A second dataset lands in the same place by a different road: Lightcast's August 2026 analysis of 135 skilled-trades occupations puts retirements at roughly 40 percent of all annual openings, with more than a quarter of trades workers already 55 or older (Lightcast).
Residential construction feels this harder than most, and the Home Builders Institute's Fall 2025 report explains why: a housing-specific shortfall of about 723,000 workers, costing the industry $10.8 billion a year in lost construction and delays, with one in five electricians already over 55 (HBI via HousingWire). Every year the exodus runs ahead of the replacements, the knowledge walking out the gate is not generic but thirty years of feel accumulated in hands and ears: how tight is tight enough on a lug, what a failing bearing sounds like through a wall, which corner the inspector always checks, and that is the gap Navigate.AI proposes to film.
What the Glasses Actually Do
The product has four features, per Wu's interview with HousingWire's Builder's Daily (HousingWire, May 2026). AI upskilling and coaching talks trainees through tasks step by step. Knowledge on-demand answers questions from specs, manuals, and historical records. AI quality control watches video of finished work and flags defects. Project scoping turns a walkthrough video into an itemized scope with pricing estimates. Tishman Speyer is rolling it out to maintenance crews across its multifamily, office, and retail properties. In the trade-school pilots the pitch is a patience machine: a coach that never gets tired of the same question, for the student too proud to ask it twice.
Wu's line for the skeptics is that construction could be "one of the bigger, if not the biggest, net job gainers due to AI." Ben Coffman, an engineering VP at STACK Construction Technologies, makes the steelman version of the case in Construction Executive: the real value is not faster takeoffs but "capturing what departing professionals know before that knowledge leaves with them," and the data-center boom has delivered a wave of young hires who need that knowledge now, while the veterans are still in the building (Construction Executive, July 2026). A DEWALT study from April found 37 percent of construction professionals piloting AI but only 9 percent using it daily, with most tradespeople still learning from YouTube, which suggests the appetite is real and the delivery mechanism is still up for grabs (DEWALT, April 2026).
All of that is plausible, and none of it is proven, because Navigate.AI is months old: no independent evaluation of its coaching quality exists, no pricing is public, and Lennar's deployment results have not been published. What exists is a well-funded promise aimed at the most sympathetic problem in the industry, which is exactly when a labor reporter starts asking who benefits if the promise works.
The Case Against, at Full Strength
Now the part the launch coverage skipped, starting with ownership, because everything else follows from it. When a journeyman with three decades of conduit bending wears the glasses for a year, the model absorbs technique that took a career to build, and no trade agreement I know of assigns rights in a worker's filmed expertise the way a union contract assigns rights in overtime. Paid by the hour, on the builder's site, wearing the builder's program, the worker faces a default that holds until someone bargains otherwise: the company keeps the knowledge and the worker keeps the hourly wage. My grandfather carried a union card for forty years, and he had a saying about any deal where one side brings the skill and the other side brings the camera, a saying I will not print here.
Ownership matters because captured knowledge is leverage: a builder who owns a working model of the master's technique needs the master less urgently, which is a polite way of saying the technology that promises to preserve the craft simultaneously hands employers a tool to discount it. Note where this scales fastest: 93 percent of construction's registered apprenticeship programs are nonunion, and nonunion programs grew 18 percent since 2020 while union-affiliated programs shrank slightly (ABC). Deployment will run fastest exactly where workers have the least bargaining power over their own data.
Then there is the camera itself, because the quality-control feature that catches defects before the crew leaves is, unavoidably, a camera pointed at the crew all day, and every minute of footage becomes a record of who stood where, doing what, at what pace. Workers in warehouses already know what happens when the productivity tool and the surveillance tool are the same device. Nothing in the public material describes who can review Navigate.AI footage, for what purposes, or whether a worker can decline to be filmed and keep his job.
And the deepest objection is about what cameras cannot capture. A Hawaiian contractor tried digitizing its veteran estimators' judgment years ago and learned the honest limit: as its BIM manager put it, "you probably have to build about 10 high rise buildings before you can really say that you know what's going on" (New Civil Engineer). Judgment is scar tissue. A coach can teach the steps of bending conduit. It cannot give a twenty-year-old the memory of the time the steps were wrong, which is where the real knowledge lives. Construction tech has a deep graveyard, Katerra at $2 billion and Veev at $647 million, and every stone in it was carved by someone who confused filming the work with understanding it.
What This Analysis Could Not Prove
Limits, stated plainly. Read the 174,500 as a floor derived from ABC's qualitative "majority," not a counted number, and ABC's 349,000 projection leans on construction-spending forecasts its own economist hedged. Those 40,000 annual completers include programs outside residential construction, and some completers leave the industry, so the true residential replacement rate is worse than 23 percent, not better. I could not verify Navigate.AI's efficacy claims against any independent source; the company is months old and its pilot data is not public. Pricing is undisclosed, so no ROI math is possible yet. The footage-ownership and surveillance concerns above are extrapolated from the product's described capabilities and standard industry practice, not from any reported incident. HBI's $10.8 billion cost figure is the institute's model, not an audited loss.
What to Do Before the Glasses Go On
Four moves, depending on which side of the camera you stand on.
If you run a residential crew and you are evaluating this or anything like it, demand three answers in writing before a pilot. Who owns the video and the trained model weights derived from your workers. Whether footage can be used for discipline or productivity scoring, who decides, and what happens to the data if you cancel. A vendor that will not put those answers in the contract has told you everything about how it sees your people.
If you are the worker being asked to wear the glasses, ask the same three questions of your foreman, and ask a fourth: can you say no. If your shop has a union steward, this is exactly the kind of technology that belongs in bargaining before deployment, not in a grievance after it. Your hands learned things over decades that no dataset had, so price that accordingly.
If you are an apprentice, use every training tool they give you, including this one, but understand its limit: it is a reference, not a mentor. It will teach you the right way to do a task. It will not teach you what the wrong way costs, because it has never paid that tuition. Find the oldest person on your site and buy them coffee, because that dataset is still the best one.
If you are buying a home built in the next five years, none of this is abstract. That crew shortage is already priced into your closing costs through delays and rework. Tools that genuinely shorten the path from green hire to competent hand are worth rooting for. Just remember that the industry has been promised a technological fix for the labor shortage for fifteen years, and the shortage is worse than when the promises started. Film the master if you want, but somebody still has to train the apprentice.