The Work No One is Left to Do
It's 5 a.m. in a vineyard in Sonoma County. The grower is 67. His son moved to Denver years ago. The crew he used to count on for pruning has been thinning out for a decade — first slowly, then all at once. The vines need to be pruned before bud break in three weeks, or the season is compromised. There aren't enough hands to do it.
So an autonomous tractor — built by Agtonomy, one of our portfolio companies — is doing it instead. It's running a row that wouldn't otherwise get run. The grower is overseeing four machines from a tablet on his porch. The work is getting done.
This is the AI story almost no one is telling.
The dominant narrative — the one that shows up in every news cycle and every dinner conversation — is that AI is coming for your job. Tech CEOs warn about it. Layoffs at Meta, Alphabet, and Amazon get framed as the leading edge of a coming labor apocalypse. There is something to that story, particularly in white-collar, screen-bound work.
But the story you don't hear — the one we believe is far bigger, and far more important — is the inverse. Across the industries that actually feed us, build our homes, defend our borders, and move our goods, the binding constraint isn't job loss. It's job vacancy.
There aren't enough people left to do the work. And the work has to get done.
We don't invest in AI to take jobs from software developers. We invest in AI to do the work no one's left to do — in the industries civilization actually depends on.
The shape of the gap
Start with the numbers, because they're staggering and almost nobody is internalizing them.
Construction. By 2031, more than 40% of the U.S. construction workforce is set to retire. We're already short roughly 500,000 workers, and for every tradesperson who retires, only 0.6 new workers enter the field. Eclipse Ventures calls this the "three trillion dollar body problem" — the trillion-plus we need to spend on AI infrastructure, plus another two-to-four trillion on the 1.6 million housing units we need to build per year, all of which require physical bodies that don't exist.
Manufacturing. Projections show that 2.1 million U.S. manufacturing jobs will go unfilled by 2030, rising to nearly 4 million by 2033. Even semiconductor fabs — the crown jewels of the reshoring story — are facing 60% workforce gaps. TSMC's Arizona plant has been delayed because of skilled-labor shortages.
Defense. The U.S. is short roughly 140,000 workers needed to support submarine construction alone. The Navy estimates the broader shipbuilding industrial base needs 174,000 new workers in the next decade. More than half of new shipyard recruits are dropping out within their first year. Meanwhile, China is launching ships faster than we are.
Agriculture. The average age of an American farm laborer keeps rising. The labor force keeps shrinking. About 80% of farm labor is needed in tight pre- and post-harvest windows — exactly when reliable execution matters most, and exactly when it's hardest to secure. This is the gap Agtonomy was built to close.
Healthcare. Roughly half of U.S. registered nurses are over 50. About a million are expected to leave the workforce by 2030. Home health turnover already exceeds 80% per year, and demand is exploding as the population ages. Hospitals can't hire fast enough to keep up with the patient demand they already have.
These aren't temporary disruptions. They are demographic and structural math problems that no amount of recruitment marketing or wage inflation can solve in time.
Why this is the moment
Three forces are converging right now, and they're what make this an investable thesis rather than just an interesting macro observation.
First, the demographics are non-negotiable. Gen Z is smaller than the millennial cohort. The boomers are retiring on schedule. The U.S. needs to hire 240,000 workers a month for the next five years just to stay flat — to replace exits, before any growth. Immigration policy is tightening, not loosening. You cannot recruit your way out of a math problem.
Second, the work itself is unappealing. In robotics circles, the canonical use case for automation has always been the 3 D's: dull, dirty, and dangerous. Pruning vines in 100-degree heat. Welding inside a submarine hull. Picking groceries in a 35-degree refrigerated warehouse for ten hours a day. Operating drilling equipment in a deep mine. These are the jobs people pull away from, and the next generation isn't lining up to take them. AI doing this work isn't displacement. It's filling a vacuum.
Third, the technology has finally become reliable and affordable enough for industries where experimentation has historically been too risky. For decades, autonomy in the physical world was a research project. Compute got cheaper, sensors got better, foundation models got smart enough to handle messy real-world variation, and — critically — the unit economics finally tipped. A farmer can now buy autonomous capability through their existing tractor dealer. A grocer can deploy a micro-fulfillment center inside an existing store footprint. The infrastructure-free, embedded-in-existing-equipment versions of these technologies are shipping today, not in some hand-wavy future.
This is what "Physical AI" actually means: AI doing the work hands used to do, in the places where hands have run out.
What this looks like in our portfolio
We invest at the intersection of AI and the physical economy — agriculture, supply chain, food, and health. Here's how that plays out concretely.
Most AI investors approach these industries from the outside in. They begin with models and look for markets to apply them to. We begin with the labor bottlenecks inside physical industries — farms that need harvesting, warehouses that need staffing, manufacturers that need throughput, healthcare systems that need capacity. The limiting factor in these markets is rarely whether the technology works in a demo. It's whether it can integrate into real operational environments where downtime is costly, workflows are entrenched, and customers are deeply skeptical of unproven systems. That requires more than technical talent. It requires trust, distribution, and domain fluency.
This is where Outline has built its edge. We partner with founders who have lived these problems firsthand — growers, operators, supply-chain veterans, manufacturing insiders — and help them navigate the incumbent networks that ultimately determine adoption. In Physical AI, distribution matters as much as the model itself. A breakthrough algorithm means little if it cannot move through dealer networks, procurement systems, industrial partnerships, and existing equipment platforms. Our portfolio companies are not asking industries to rip and replace infrastructure. They are embedding intelligence into the systems that the physical economy already runs on. We believe this category will not be won by the firms with the broadest AI exposure, but by the investors who understand how technology actually gets adopted in the real world.
Agtonomy. Founded by Tim Bucher, a lifelong Sonoma County farmer who also happens to have been an executive at Apple, Microsoft, and Dell. Agtonomy doesn't try to build new tractors — growers don't trust startup hardware with their season, and rightly so. Instead, it embeds autonomous AI into the equipment growers already buy from, brands like Kubota. One operator can now manage a fleet of machines from a phone or tablet, doing the repetitive fieldwork — pruning, mowing, spraying — that the labor pool has shrunk away from. The role of the human shifts toward oversight and decision-making, not elimination. The grower stays in business.
Fulfil. U.S. online grocery is now a roughly $120 billion market, growing three times faster than in-store grocery. There's just one problem: when you fulfill a grocery order manually, you lose money before delivery costs even enter the equation. The work is brutal — picking across ambient, refrigerated, and frozen zones, dozens of times an hour, with high error rates. Fulfil's robots handle every product category and every temperature zone in a small-footprint micro-fulfillment center that bolts onto a store. Late last year, Amazon deployed Fulfil inside a Whole Foods location in Pennsylvania. The economics finally work, and the work that wasn't getting done is now getting done — quickly, accurately, and with real-time inventory tracking that cuts food waste at the same time.
Crisp. The opposite end of the same problem. Roughly 14% of global food supply is lost between farm and shelf — not for lack of food, but for lack of coordination. Crisp gives retailers, distributors, and brands real-time visibility into demand so the right product moves to the right place at the right time. Less spoilage. Fewer stockouts. More food reaching households that need it, at lower cost. The labor saved here isn't physical — it's the cognitive overhead that breaks down when humans try to coordinate millions of SKUs across thousands of locations in their heads.
Phytoform. Conventional crop breeding takes more than a decade and roughly $115 million to get a single trait from the lab to the field. That timeline doesn't survive contact with climate change. Phytoform's AI compresses it to under two years and roughly $2 million by computationally identifying the right modifications before any physical trial. The bottleneck this removes isn't human expertise — it's iteration speed. Breeders are still indispensable. They're just no longer waiting a decade per cycle.
And we're actively looking at adjacent applications of the same thesis — physical AI for construction, mining, defense manufacturing, and skilled-trade environments. Anywhere the work has to get done, and the workers aren't there.
Why this is also the most direct AI-for-Good thesis we know
"AI for Good" gets used loosely. It tends to evoke chatbots that tutor underserved students, or dashboards that monitor deforestation. Those are fine. But they aren't where AI's biggest social return on capital lives.
The biggest social return — by an enormous margin — is in the systems we all rely on every day and rarely think about. The food system. The supply chain. The construction trades. The defense industrial base. The healthcare workforce. These are systems that already operate below their potential because they don't have enough hands. Every gap in those systems shows up downstream as a higher grocery bill, an unaffordable home, a delayed surgery, essential safety systems operating below readiness, a community whose hospital is forced to scale back.
AI that fills those gaps doesn't just generate venture returns. It keeps essential things working. It's the most direct line we know between technology investment and the lives of the roughly 99% of people who don't work in tech.
AI for Good is not a separate category from this. It is this. The work that holds civilization together is also the work the labor market can no longer staff.
What we're funding
We're an early-stage fund focused on Physical AI in the physical economy. We back founders who've felt the labor problem in their bones — farmers, operators, supply-chain veterans, manufacturing engineers — paired with the technical depth to actually solve it. The pattern in our portfolio is consistent: a hard, real-world problem, a domain expert as founder, AI as the lever that finally makes the unit economics work, and an industry that's been trying and failing to hire its way out of the gap for years.
The companies that will define the next decade of AI won't be the ones that automate marketing emails. They'll be the ones that get the harvest in, the ship built, the home framed, the prescription filled, and the groceries delivered — when the people who used to do that work aren't there anymore.
If you're a founder building in that space, we'd like to hear from you.
SOURCES
Caroline Duffy, AI and the Labor Shortage Economy, Cowboy Ventures, April 2026.
Aidan Madigan-Curtis & Ryan Gibson, The Three (Trillion Dollar) Body Problem, Eclipse Ventures, July 2025.
U.S. Bureau of Labor Statistics, Job Openings and Labor Turnover Survey (JOLTS), 2026.
USDA Economic Research Service, Farm Labor data, accessed 2026.
American Hospital Association, 2025 Health Care Workforce Scan.
FAO, The State of Food and Agriculture 2019: Moving Forward on Food Loss and Waste Reduction.
Defense News, "The US Navy is at risk of losing vital shipbuilding skills," November 2024.
War on the Rocks, "A Workforce Strategy for America's Shipbuilding Future," July 2025.
Brookings Institution, "Keeping workers safe in the automation revolution," 2023.
DCVC, "Fulfil's robot revolution comes to Whole Foods," November 2025.
AgFunderNews interviews with Tim Bucher, Agtonomy, 2024–2025.
AWS Industries Blog, "Phytoform Labs processes plant genomes 100x faster with AWS HealthOmics," October 2025.
