The robot in the courthouse attic

The robots are almost ready. The public question is what we ask them to do first.

Two items sat next to each other in my saved reading list this week, unrelated by any measure an algorithm would recognize. One was Annie Lowrey’s piece in The Atlantic, “The Time Tax: Why is so much American bureaucracy left to average citizens?” The other was an update from the robotics company Figure: their humanoid robots now outnumber the human employees at their own facilities, and the next generation, F.04, has just passed its final design cycle.

I read them back to back, and they refused to stay separate. One describes a country that taxes its citizens in hours and paperwork. The other describes machines built to do exactly the kind of work nobody wants to count. Somewhere between the two sits a question neither of them asks: what would the new machines have to do, and for whom, before the public has reason to trust them?

This is where that question led me, and it starts with an American errand.

You have probably done it. Taken time off work, driven to a government office, paid a fee and stood in line to get a paper copy of something the government already knows about you. Americans spend billions of hours a year on federal paperwork, a burden the journalist Annie Lowrey has named the time tax. The average tax return alone takes twelve hours and 290 dollars to file. In Sweden, where I live, the tax return arrives pre-filled and most people approve it by text message in under a minute. The certificate you stood in line for, a Swede downloads in seconds. The difference looks like technology but is mostly architecture, and some of the American friction is chosen, an expression of distrust toward central registries that deserves respect rather than ridicule. Much of it, though, is simply paper nobody ever moved.

In Washington County, Ohio, court records dating back to 1783 sat in a courthouse attic. The shelves reached twenty feet. The service elevator had been dismantled decades ago. The lowest rows had flood damage. The records stayed there until volunteers from FamilySearch, the genealogy arm of the Mormon church, came and digitized them.

This is how America’s paper backbone gets digitized today, as far as I can tell: by churches and corporations, on their terms, because the public sector never considered it affordable. Pennsylvania let Ancestry digitize its birth and death certificates, its veterans’ burial cards, its records of enslaved people. When a genealogist later requested copies, the company argued in court that it owns the digital records. Virginia’s vital records went online behind a twenty-dollar monthly paywall. Documents that were public on paper became products in digital form.

Private companies built extraordinary tools, and they showed up when nobody else did. What they cannot answer is whether a public institution should outsource its memory along with the scanning.

Meanwhile, humanoid robots exist. Figure’s machines recently sorted packages for eight hours straight and can feel an object weighing three grams. Each one draws on the same finite commons: grid electricity, rare materials, AI compute. And the flagship story their maker tells the public is towel-folding in the homes of people who can afford a robot. I keep thinking about who is asked to accept that. The waitress who spent twelve hours and 290 dollars on a tax return a Swede approves by text. The veteran whose paper records sat in a queue that peaked at 604,000 unanswered requests. They are told the machines are coming, and the first proof is a folded towel behind someone else’s door.

The order matters more than the speed. A technology that visibly serves the many can afford its luxuries later; a technology introduced as a luxury will meet the many as opponents. Gains for the broad public first, legitimacy from those gains, and only then the anchoring that lets the technology scale.

There is a category of work waiting for exactly this: the undone work. Tasks no profession owns. Over 600,000 miles of old fencing cross the American West, and roughly one deer, elk or pronghorn dies per two and a half miles of it each year; volunteers cut and haul what they can. More than 80 percent of the nearly 17 million tons of surplus produce on American farms never gets harvested; 1.6 percent is donated. Cigarette butts, the most collected beach litter on record, get picked by hand because grabber tools can’t manage what the tide works into the sand. Nobody loses a job when the undone gets done.

The archives may be the deepest case. So I want to be careful with my own image. A robot walking a county archive at night, carrying boxes so that no one ever again takes a day off work to fetch a paper copy of their own birth certificate, does not carry a system change. A scanned certificate is still a certificate. Legislatures change systems; robots carry boxes. What the robot removes is the excuse. The night it starts walking, “we can’t afford to digitize” dies as an argument, and the choice stands naked: public records digitized under public control, or under Ancestry’s.

Figure says F.04 has passed its final design cycle, the biggest leap between robot generations so far. I have no reason to doubt the pace; it has been remarkable. But pace answers how fast, never for whom. The whole case for a humanoid machine is that it fits a world built for human bodies: stairs, narrow doorways, twenty-foot shelves. An attic where the elevator was dismantled decades ago. F.04, if it works as promised, could climb to those boxes in Washington County. Whether it ever will is not an engineering question.

I don’t know who pays for that robot. I don’t know which county would go first. I notice that the church and the corporation didn’t wait for an answer. And the next time you take a day off work to stand in line for a paper the government already has, you will know one thing standing there: it is no longer because the machines aren’t ready.

Originally published at saraeson.substack.com (9 of July 2026)