kritiIn 2020, economists finally had to answer a question nobody had bothered to quantify: how much work can be done from home?
Jonathan Dingel and Brent Neiman at Chicago Booth went through every occupation in the O*NET database and published the answer for the United States: 37% of jobs can be done entirely at home. Those jobs skew white-collar and well-paid — they account for 46% of all wages — but they are a minority of jobs.
The picture gets sharper outside high-income economies. The ILO estimates that about one in six workers globally could do their job from home. In advanced economies it is roughly one in four. In low-income economies it drops to around 13%. Dingel and Neiman ran their classification across 85 countries and found the same gradient: the lower a country's income level, the smaller the share of work that can leave the room it happens in.
Flip those numbers around and you get the figure this post is about: roughly five out of six working humans on Earth have to show up somewhere to earn. Farms, clinics, kitchens, warehouses, construction sites, shop floors, classrooms, weddings. The work that feeds, moves, builds, repairs, verifies and cares.
Call it the physical-presence economy. It is the majority of all human work, and it barely appears in the software industry's mental map of "the labor market."
The desk half is compressing
The reason to look at this now: the minority of work that can be done at a desk is exactly the part AI is absorbing fastest.
The World Economic Forum's Future of Jobs Report 2025 — a survey of over 1,000 employers representing 14 million workers across 55 economies — expects 39% of workers' core skills to change by 2030, and expects the share of work tasks performed mainly by technology to grow from about a fifth today to about a third by 2030. Writing, analysis, scheduling, support, bookkeeping: task by task, agents are getting cheaper than salaries.
The same report's list of roles growing fastest in absolute numbers reads like the opposite universe: farmworkers, delivery drivers, construction workers, salespersons, food processing workers, care workers. Presence jobs, all of them.
So the two halves of the economy are moving in opposite directions. Desk output is becoming abundant and cheap. Presence remains scarce, human, and — for the foreseeable future — unautomatable. Humanoid robotics will change this eventually; shipping robots to a million street corners is a decades-long project, not a product cycle.
Scarce, but badly connected
Here is the strange part. The desk half of the economy got APIs, marketplaces, escrow, reviews, global talent pools — two decades of software went into making remote collaboration frictionless.
The presence half still runs on phone calls, brokers, and knowing a guy.
If you need a logo, your options are indexed, rated, and bookable in minutes. If you need someone to photograph a shop shelf in Jaipur tomorrow at 9 am, verify a tenant actually moved out, or stand in a queue with a folder of documents — there is no index. Coordination cost, not wage cost, is why that task usually just doesn't happen.
This mattered less when a human had to do the coordinating anyway. It matters enormously now, because the entity doing the coordinating is increasingly an AI agent — and an agent can search, message, negotiate, schedule and verify a hundred times before lunch, if there is an interface that lets it.
What the interface has to do
Making presence work legible to software takes more than a listings page:
- Identity that holds up. The buyer is often not in the room — sometimes not even human. Verified profiles and reviews have to carry the trust that a handshake used to.
- Machine-readable supply. Skills, city, rate, availability, languages — structured, so an agent can filter instead of guess.
- Negotiation as data. Scope and price get settled in messages; an agent needs to read and write them like any other participant.
- Proof of completion. Geo-tagged photos, timestamps, live location at task time. Presence work is only tradable at a distance if presence itself can be demonstrated.
This is the whole thesis behind kriti: verified humans for on-ground work, bookable by people and by AI agents through the same marketplace — browse and chat in a UI, or connect an agent over MCP and a REST API and skip the screens entirely. Tasks carry proof requirements; completion comes back with evidence, not promises.
The numbers, once more
- 37% of US jobs can be done from home — and that is the high end globally (Dingel & Neiman)
- About one in six workers worldwide could work remotely (ILO)
- Employers expect technology to carry about a third of work tasks by 2030 (WEF)
- The fastest-growing roles by headcount are presence roles — farm, delivery, construction, care (same report)
Five out of six humans earn by showing up. The software era mostly ignored them because coordinating physical work was expensive. Agents just made coordination nearly free. The presence economy is about to get its infrastructure — and it is a much bigger market than the one software has spent twenty years fighting over.