AI Is Eating the Boring World

The biggest AI businesses of the next decade won't sell software. They'll absorb the wage bill of the physical economy.

The clearest tell is Metropolis. A nine-year-old AI parking startup raised a $1.7B Series C in October 2023 to take SP Plus private at a $1.5B price tag. The pitch wasn't "we sell SaaS to parking operators." The pitch was: we are the operator. Today Metropolis runs 4,000+ locations and 50M+ monthly transactions across North America, with computer vision handling entry/exit and AI handling everything that used to require a person in a booth.

Halter is the rural version. The Founders Fund-backed agtech put AI collars on cattle and just raised $220M in March 2026 at a $2B valuation to scale virtual fencing globally. One operator, a phone, a herd of 1,000 cows.

These are not software companies. They are operating companies wearing a software cap table.

The wedge

The most painful, labor-intensive work in physical-world businesses -- answering phones, dispatching trucks, scheduling appointments, processing claims, watching things move -- was chronically under-automated because it was too messy for SaaS and too cheap for custom software. LLMs plus cheap perception flip the unit economics overnight. The work that used to require a person in a chair now costs pennies per call.

Two flavors that are working

1. AI-native rollups -- buy the dumb infrastructure, install AI as the OS.

Metropolis is the canonical example. Buy the parking lots, run them yourself, capture both the software margin and the operating margin. Halter is doing it on ranches. We'll see this pattern in HVAC, dental practices, car washes, self-storage, laundromats, regional logistics, and senior care over the next 36 months. Whoever owns the asset and the AI captures the full stack.

2. AI workforce sold into existing operators -- don't buy the business, sell the agent.

Avoca (Kleiner Perkins, $125M total at a $1B valuation in April 2026) answers the phone for HVAC, plumbing, and roofing companies -- the kind of business that loses 30% of inbound calls and has no idea. The CEO literally got the idea after watching a Texas pool company miss seven calls in an hour.

Nitra (a16z, $187M total raised, $1B in processing volume, revenues grew 740%+ in 2025, CityMD founder on the board) runs the back office for 700+ independent medical practices. Claims, payments, scheduling -- the unsexy 80% of why doctors quit private practice.

Both companies charge for completed work, not seats.

The unit-economics shift nobody is pricing in

Last cycle's winners -- ServiceTitan, Toast, Procore, Veeva -- charged a recurring seat fee for a dashboard. The unit was "user logged in this month."

This cycle, the unit is completed work. Calls answered. Jobs booked. Claims processed. Cows moved. Cars parked. The TAM isn't software seats. It's the wage bill of every offline industry.

That's a different number entirely. US healthcare admin alone is ~$1T. US construction services is ~$2T. Add agriculture, logistics, hospitality, trades, mobility, and you're at $10T+ of payroll that AI can now nibble at on a pay-per-outcome basis.

The map

• Agriculture: Halter (cattle), Carbon Robotics (weeding), Monarch (autonomous tractors)
• Trades: Avoca (voice for HVAC/plumbing/roofing), Ressl AI, Picus Capital's trades portfolio
• Healthcare practices: Nitra (back office), Patientdesk, Dentio, Digitail (vet clinics)
• Mobility / parking: Metropolis, Hayden AI (transit enforcement)
• Manufacturing / defense: Hadrian (autonomous machine shops), Path Robotics, Anduril
• White collar back office: Proximitty, Oyyo, LexAxiom
• Logistics / warehousing: Symbotic, Locus Robotics
• Hospitality: Bear Robotics, Servi
• Senior care / facilities: emerging, watch this space

Three investable layers

1. The operators -- companies that buy the asset and run it with AI. Highest capital intensity, highest margin capture if they win.
2. The agent platforms -- sold into the long tail of independent operators who'll never get bought. Avoca and Nitra are the templates.
3. The picks and shovels -- voice infrastructure, vertical perception models, compliance tooling, observability for agents that take real-world actions. Lower-multiple but the entire stack runs on top.

The kicker

Vertical SaaS sold the dashboard. Vertical AI sells the outcome. For 20 years VCs argued about whether the trades worker would adopt new software. That argument is now moot -- the AI is the worker.

The boring businesses got skipped in the last cycle because the math didn't work. The math works now. Most of the value will accrue between now and 2028, before the category has a clean name. By the time it does, the multiples will already be priced in.

Start with the wage bill. Work backwards to the cap table.