The Janitor's Observations: An Accidental Ethnography
May 2029
The notebook was a Moleskine, pocket-sized, with a black cover and an elastic band. Samuel Osei had bought it in 2018, the year he started cleaning the fourth floor of the Meridian AI Research Institute in Cambridge, Massachusetts. The notebook cost twelve dollars. The observations in it, by the time a historian found them in 2041, were considered priceless — though Samuel, who was sixty-seven by then and still cleaning the fourth floor, disagreed with this assessment.
"They are not priceless," he said. "They are notes. A janitor's notes. I wrote them because I noticed things. That is all."
He had noticed things for eleven years. This was not unusual for a janitor — janitors are professional noticers, people whose work requires them to attend to what others leave behind, to read a room not by its stated purpose but by its residue. A janitor knows who works late by the lights. Knows who is stressed by the trash. Knows who has been crying by the tissues. Knows who has given up by the desk.
Samuel's fourth floor housed three research teams: the Language group, the Vision group, and the Systems group. Together, they employed approximately sixty researchers — PhD students, postdocs, senior scientists, a rotating cast of visiting scholars. Samuel cleaned their offices, their labs, their meeting rooms, their kitchen, and their bathrooms. He arrived at 10 PM and left at 6 AM. He had been doing this since 2018. He had watched the teams change, grow, shrink, celebrate, despair, and — in the final years of his notebook — do something he had never seen a group of people do before.
He had watched them build something that frightened them.
The lights
April 2026
The Language group leaves their lights on past midnight now. Not all of them — Dr. Rivera and Dr. Kuo and the new postdoc whose name I cannot remember. They used to leave by 9. Now they are here at 1 AM, 2 AM. I clean around them. They do not notice me. They are looking at screens. The screens have text on them — not the kind of text I read. Code. Numbers. Graphs. I do not understand the graphs. I understand the lights.
When people leave their lights on past midnight, something has changed. Either they are excited or they are afraid. The difference is in the body. Excited people lean forward. Afraid people lean back. Dr. Rivera leans forward. Dr. Kuo leans back. The postdoc sits very still, which means she has not decided yet whether to be excited or afraid.
The whiteboards
September 2026
Someone is erasing the whiteboards in Meeting Room 4B every morning before the other teams arrive. I know this because I clean 4B at 11 PM and the boards are covered with equations and diagrams, and when I clean 4B again at 5 AM the boards are blank. I am not the one erasing them. Someone comes between 5 AM and 8 AM and wipes the boards clean.
I asked Marcus in facilities whether he knew who was doing it. He did not. I began arriving at 4B at 5:30 AM. I saw Dr. Kuo. She was alone. She was wiping the boards carefully — not quickly, not with the impatient swipe of someone clearing space. Carefully. The way you erase something you do not want other people to see.
I have cleaned offices for thirty years. I know what it looks like when people hide their work. Normally, people hide their work because it is bad — because they are ashamed of it, because it is wrong or incomplete. Dr. Kuo's work was not bad. I could see it before she erased it. The equations were dense and precise. The diagrams were elegant. She was not erasing bad work. She was erasing good work that she did not want the other teams to see.
When a researcher hides good work from colleagues, the work has implications the researcher is not ready to share. This is the first time I have seen this happen on the fourth floor.
The coffee mugs
February 2027
There are nine coffee mugs on the desk in Lab 4C. Nine. I know because I count them every night. The number has been increasing: three in January, five by mid-January, seven by the end of the month, nine now. The mugs belong to Dr. Park. He is not washing them. He is not bringing them to the kitchen. He is leaving them on his desk, each one marking a night he did not leave.
Dr. Park is sleeping in the lab. I know this because I find the couch cushions rearranged in the morning and there is a toothbrush in the bathroom that was not issued by facilities. He has been sleeping in the lab for approximately three weeks. His colleagues do not know — or they know and do not comment, which in a research lab means the same thing.
People sleep at their work for two reasons: because they cannot stop working or because they cannot go home. Dr. Park cannot stop working. I know this because the mugs accumulate near his keyboard, not near the couch. He is sleeping at his desk, not on the couch. The couch is for sleeping. The desk is for working. He is doing both at the desk because the work will not let him leave.
In thirty years of cleaning, I have seen people sleep at their desks perhaps a dozen times. It means the work has crossed a line from occupation to obsession. Occupation is something you do. Obsession is something that does you.
The trash
August 2027
The trash from the Language group has changed. For years, their trash was normal research trash: printer paper, sticky notes, coffee cups, food wrappers, the occasional torn-up draft. Since June, the trash includes something new: printed outputs — sheets and sheets of text, generated by the system they are building, printed and read and annotated by hand and then thrown away.
I read the outputs. Not because I am nosy — because they are in the trash, and the trash is my domain. The outputs are text. Some of it reads like a person wrote it. Some of it reads like a person almost wrote it. The difference is subtle. I cannot always tell which is which. This, I suspect, is the point. This is what Dr. Rivera is leaning forward about and Dr. Kuo is leaning back about. The system is producing text that a janitor with no scientific training cannot distinguish from human writing.
I am not a scientist. But I have been reading trash for thirty years, and I know the difference between trash that people discard carelessly and trash that people discard carefully. The outputs are discarded carefully — folded, not crumpled. Placed in the bin, not tossed. This is the trash of people who are disposing of something that matters. Something they need to get rid of but cannot treat casually.
They are printing the outputs, reading them, annotating them — and then destroying them. Not because the outputs are bad. Because the outputs are good. And good outputs from a system that produces human-indistinguishable text are — I do not know the word. But Dr. Kuo knows the word, because she is the one who erases the whiteboards, and Dr. Park knows the word, because he cannot go home, and Dr. Rivera knows the word, because she leans forward with the specific posture of a person who is excited and afraid at the same time and cannot separate the two.
The moment
March 8, 2028
Something happened today. I do not know what. I arrived at 10 PM and the fourth floor was silent. Not the silence of an empty floor — the silence of a floor full of people who are not speaking. Every lab was occupied. Every light was on. No one was working. They were sitting at their desks, looking at screens, and the screens — I could see as I moved through the corridor — all displayed the same thing. A result. A graph. A number. I do not know what the number was.
The silence lasted until approximately 11:30 PM, when Dr. Rivera came out of her office and walked to the kitchen and made coffee and stood at the counter drinking it and I was mopping the kitchen floor and she looked at me and she said: "Samuel, has it ever occurred to you that you might be the last person in this building doing work that a machine cannot do?"
I said: "Dr. Rivera, I have been the last person in this building doing work that a machine cannot do for several years. The difference is that now you have noticed."
She laughed. Then she cried. Then she took her coffee back to her office and closed the door. I finished mopping the kitchen floor.
The publication
Samuel continued his notes through 2029. He recorded the departure of Dr. Kuo, who left the institute in July and moved to a cabin in Vermont. He recorded the arrival of journalists, who came in increasing numbers and who left behind a different kind of trash — voice recorder batteries, press badges, the particular detritus of people passing through a place rather than inhabiting it. He recorded the gradual change in the building's atmosphere — from the focused energy of a research lab to the performative energy of an institution that has become a destination.
He stopped writing in December 2029. Not because he stopped noticing — he noticed things until the day he retired, and probably after. He stopped because the notebook was full. 184 pages, front and back. Eleven years of observations.
The notebook sat in a drawer in Samuel's apartment in Somerville for twelve years. In 2041, a historian of science named Dr. Amara Osei-Bonsu — no relation, though the shared surname produced a moment of warmth when they met — was researching the social history of the Threshold era and encountered, in an interview with a former Meridian researcher, a reference to "the janitor who noticed everything."
She contacted Samuel. He was skeptical. "These are a janitor's notes," he said. "They are not science."
"They are the only eyewitness account of the daily life of the lab during the Threshold period that was written by someone who was not involved in the research," Dr. Osei-Bonsu said. "Every other account is written by researchers — people who understood the science and were invested in its success or failure. Your account is written by someone who understood none of the science and all of the people. That is irreplaceable."
Samuel allowed her to photograph the notebook. She published extensive excerpts in her book, The Threshold: A Social History (2043), with Samuel's permission and with annotations that placed his observations in the context of the breakthroughs they unknowingly documented.
The book's most-quoted passage was Samuel's observation from August 2027 — the one about the printed outputs that were folded, not crumpled. Dr. Osei-Bonsu's annotation read: Samuel Osei, a janitor with no technical training, identified the emotional valence of the AI-generated text through the manner of its disposal. He could not read the outputs as science. He read them as trash — and the trash told him everything the researchers' publications could not: that the people building this system were handling its outputs with the careful reverence of people who know they are touching something unprecedented.
Samuel read the annotation. He disagreed with the word "reverence."
"It was not reverence," he said. "It was care. They were being careful. The way you are careful with something that might break. Or that might already be broken. Or that might break you. I have seen people handle trash that way before — when they are throwing away letters from someone they love, or medical results they are not ready to read, or drafts of something they poured themselves into and cannot keep."
"They were throwing away evidence of something they had made. Something that frightened them. And they folded the paper because folding is the body's way of saying: I am treating this with respect, even as I discard it. I know this matters. I am letting it go anyway."
This is the third entry in The Backstory Shadows. For the coder whose confession was written during the same period Samuel observed, see The Coder's Confession. For the HR director who recorded a different kind of human residue, see The Exit Interview.