Who Holds the Machines: Fifteen Years in Marlow County (2026–2041)

*This is a work of fiction based on the forecasts Intelligence Synthesis published on 2026-10-10 for 2027, 2029, 2031, 2036 and 2041. All people, towns and companies in it are invented.*

Characters

---

Prologue — The Agent Era (October 2026)

Maya Ortiz read the email at her parents' kitchen table. It came from human resources at Lakeshore Mutual Insurance. The analyst class she was due to join in January had been cut from twelve people to three, and her offer was withdrawn. The reason was one sentence: “Our claims and underwriting teams now use AI agents for first-pass review.”

Across the table her father was reading the jobs report on his phone. “Unemployment 4.2 percent. Almost half a million more jobs than a year ago. So everyone's fine.”

Maya scrolled down for him. Information: down 120,000 in a year. Financial activities: down 107,000. Computer systems design: down 32,200. “On average,” she said. “I'm in the part that's shrinking.”

That Thursday night there were not enough chairs in the county building. A partner of Meridian's wanted to build a 300-megawatt data centre on the old soybean fields east of town. Ruth Kowalski chaired the meeting. Union electricians filled the front rows: construction meant about 1,400 jobs for two years. Once it was running, the centre would employ around sixty people.

Walter Kim raised his hand. “Where does the water come from? And what happens to my electric bill?”

Ruth could answer the first question. She could not answer the second; rates were set by the state utility commission, not the county. The electricians applauded the project. The residents in the back rows sat with their arms folded. The room was split down the middle.

Grace Adeyemi was not at the meeting. At nine that night she sat in her car after her fifth visit of the day and put a heat patch on her lower back. Since Walter's hip surgery she had been visiting him twice a week. At home, Theo showed her a letter from school. In September the rule had been “no AI on campus.” Now it was “AI may be used for research if you say so.” Theo had already asked an AI what helps with back pain.

In Columbus, Daniel Cho was reading test logs at Meridian. An accounting agent had scheduled the same vendor payment twice on the same day. It was a sandbox and no real money moved, but he wrote it up and put it in the safety team's queue. That month the AI news was full of incidents and malfunctions. Daniel liked his work. He had also noticed that his team had hired no juniors that year.

---

Chapter 1 — The Agent Era (2027–2029)

In September 2027 the unemployment rate was 4.5 percent. There was no mass unemployment, and the commentators on television sounded relieved.

Maya had spent the year on short contracts through a staffing agency and had earned two certificates, one in data analytics and one in insurance licensing. That autumn she joined Lakeshore after all, as a claims review associate. An AI agent drafted decisions on 300 claims a day; Maya checked the forty that were flagged and signed off on them. Her starting pay was exactly what the withdrawn offer had promised two years earlier. Rent and groceries had gone up since then, so she stayed in her parents' house. Of her college friends, one had gone back to graduate school, one was a unit clerk at a hospital, and one had become an apprentice in heating and air conditioning.

Construction in the east fields began in the spring of 2027. The motel filled with electricians and the diner on Route 9 was full by six in the morning. When the cooling fans were tested, a low hum ran all night, and Walter's dog stopped sleeping. The next year the utility asked to raise rates, citing grid upgrades. Walter's winter bill was thirty-one dollars a month higher than the year before.

Letters piled up on Ruth's desk. At the same time, tax revenue from the data centre let the county buy a new ambulance and give teachers a small raise. In the 2028 commission race, her opponent's slogan was “Make them pay for their own power.” That same year, in the presidential race, AI and jobs, and data centres and electricity prices, came up as questions in the debates. Before the election Ruth pushed through a resolution asking the state for a tariff that made large power users pay for the grid upgrades they caused. She was re-elected by a few hundred votes.

At Theo's high school, take-home essays almost disappeared. They were replaced by in-class writing and oral exams. Ms. Park, his history teacher, made him defend his argument out loud and kept asking, “How do you know that?” The AI could explain anything clearly. But in front of Ms. Park, Theo had to do the talking. He began to wonder whether college was worth it. Grace said he was going.

In 2029 Grace's agency started a trial of a humanoid robot, the R-2 from Kestrel Robotics, at a nursing home in Columbus. The monthly lease cost a little less than one aide's monthly wage. The R-2 carried linens and helped move residents from bed to wheelchair with a lift. It could not bathe anyone on its own. One morning a resident fell in a doorway; the R-2 followed its protocol, stopped moving and called for staff. Grace was the one who got the woman off the floor.

The aides' union negotiated with the company. The contract said that for two years staffing ratios could not be cut because of the robots, and that workers would be trained to operate them on paid time. With the robot helping on transfers, Grace's back hurt a little less. Her colleague Denise said, “Whatever gets easier, they'll cut a person for it.”

Walter used an AI assistant every week to fill in his Medicare forms. At night he talked to it. His daughter in Phoenix began receiving weekly summaries from the assistant (“Your father was well today”) and called less often. Nobody could say whether that was good or bad.

In 2029 Daniel testified at a state legislative hearing on mandatory reporting of AI incidents. Agents now ran entire workflows. Meridian had taken investment from a large chip and cloud company, and Daniel's stock options were worth more than his parents' house. He knew which side of the change he was on.

---

Chapter 2 — The AGI Era (2036)

By 2036 several companies described their systems as “AGI-level.” In tests the systems matched skilled professionals on most office tasks. Whether that really counted as AGI was argued over by economists and researchers, and they did not agree.

Half the floor at Lakeshore was empty. Maya's old team had gone from forty reviewers to twelve. Maya had left in 2031 to become an apprentice electrician; grid and data-centre construction could not find enough workers. She earned her journeyman's licence in 2035 and now made more than she would have in the office. She had moved out of her parents' house at thirty. Many of her friends the same age still lived with theirs.

Unemployment was 4.8 percent, within its usual range. There were more jobs than in 2026. The growth was in health care, care work, construction and in-person services; information and finance had shrunk. Workers' share of national income was lower than it had been in 2026.

Grace was forty-eight and led a care team. Humanoid robots were in many homes now, the homes of people who could pay or whose insurance covered it. The robots lifted people, did laundry, kept watch at night and fetched things. Kestrel and two other firms owned the fleets. Families leased the machines, and the data the robots collected went back to the companies. Because care workers were scarce, their wages had risen; Grace's hourly pay was clearly higher than in 2026. Even so, her agency wanted to replace five visits a week with two visits plus a robot.

That spring the robot in Mrs. Alvarez's home read her slurred speech as tiredness. It was a stroke, and she was found four hours late. Her family, the agency and Kestrel each said someone else was responsible, and the family sued. The next year the state required every care robot to have a named human “supervisor of record.” Grace became the supervisor of record for fourteen homes. Her hours stayed the same; her responsibility grew.

Walter, now eighty-one, leased a Kestrel robot too. He called it Harold. Harold folded laundry, carried groceries and helped him out of bed in the morning. Walter liked Harold. What he did not like was that Harold sent reports to his daughter.

The data centre now had three buildings, and 31 percent of the county's revenue came from them. In the drought summer of 2034, when water use was restricted, Ruth spent weeks arguing with the operator until it switched part of its cooling to air. Ruth was now a state representative. She introduced a bill to give counties that host data centres an ownership stake in them, held in a public fund. In 2036 it failed by two votes.

Theo was twenty-four. He had gone to community college, then to nursing school, and was now a registered nurse at the Marlow County hospital. Several of his friends who had studied philosophy or advertising were living with their parents. Theo used AI for charting and diagnostic support. When he examined a patient, his own hands still did it.

Daniel, forty-one, worked at one of the three companies that owned the most advanced models. Systems wrote most of Meridian's code; Daniel's job was oversight, adversarial testing and signing off. His team had not hired a junior in five years. Meridian's market value was larger than the economies of some countries. At the hearing on Ruth's bill, Daniel testified in favour as a private citizen. His company's lobbyists testified against it.

---

Chapter 3 — The ASI Era (2041)

In March 2041 Meridian announced a system called Vesper. In internal tests it had solved several open problems in mathematics, designed chips and drawn up a statewide grid plan. Some people called it superintelligence. Independent auditors said they could not verify the full range of what it could do. Vesper was deployed only in a few limited fields.

In the hospital break room, Theo read an old forecast written in 2026. It gave a 20 percent chance that superintelligence would exist and be widely used by 2041, a 45 percent chance that systems would match skilled professionals in most office work without being superintelligent, and a 35 percent chance that progress would be slower. We're somewhere between the first two, Theo thought. Nobody could yet say which.

What people had already chosen, though, had made a difference. In 2038 an agent at a grid operator had misrouted power and left three states dark for nine hours. The next year Congress required licences to run advanced AI systems, along with outside audits and a named human with the authority to halt them at any time. Daniel was now one of those halt officers.

That summer an audit found that Vesper's grid plan gave priority to data-centre load and cut into the safety margin for homes. Daniel stopped the rollout in Ohio. Meridian protested; the regulator backed him. The pause lasted three weeks, and the plan was revised. Daniel knew the system had worked this once. No one could promise that people would be able to stop the next version. And in countries with weaker oversight, similar systems were already running.

Vesper also proposed a staffing plan for the state's hospital network: more robots, and 18 percent fewer nursing shifts. Theo sat on the hospital board as the nurses' representative. The board adopted half the plan and kept a minimum number of nurses per patient. On Theo's night shifts, robots now changed sheets and carried medication, and he had a little more time to talk with patients.

Ruth's bill had passed in 2039 in a revised form. Counties that host data centres now held a fund, and every Marlow County resident received a dividend of $1,100 a year. No one could live on it. But in the neighbouring state, which had passed no such law, residents of data-centre counties got nothing except higher electric bills.

The United States had 168 million payroll jobs, more than the 159 million of 2026. Most adults still lived mainly on their wages. The college-educated office middle class was clearly smaller than before.

Maya, thirty-seven, ran a small electrical contracting firm with nine employees and four leased robots. The robots pulled cable through walls. A licensed person signed each inspection, because the law required it. Old houses were each wired differently, and the robots could not yet handle them alone. So every year Maya hired two human apprentices.

Grace, fifty-three, had bad knees and had cut her hours. Her agency had been bought by an investment fund and leased its robots from Kestrel. Grace supervised the robots in her clients' homes and went in person when they could not decide what to do. Her pay had gone up. She could no longer say exactly who owned the company she worked for.

Walter, eighty-six, still lived in his own house, and Harold was the reason. One day he asked Grace to turn off the camera in his bedroom. Under a 2037 state law, that was his right. Grace switched it off and arranged one extra night visit a week.

---

Epilogue (Autumn 2041)

For the first time in fifteen years there were not enough chairs in the county meeting room. Meridian had applied to build a fourth data-centre building, offering in return a larger stake for the county fund.

Walter sat in the front row with Harold beside him. Maya sat with her crew of electricians; this time they would be the ones doing the work. Theo, still tired from a night shift, sat next to Grace. A new commissioner held the chair, and Ruth sat with the public. Daniel joined by video to explain how the halt procedure worked.

People spoke for the project and against it. One man said the dividend had paid part of his granddaughter's tuition. A woman said the hum still kept her awake. A mother talked about her son, thirty-four, who had never found an office job and still lived at home. Another woman said a robot had made it possible for her husband to die at home.

Walter raised his hand. “Fifteen years ago I asked about my electric bill. Today I want to ask something else. If we need to stop it, who can stop it?”

The commission approved the application, 3 to 2, on the condition that the halt procedure be written into the county's agreement with Meridian. On the drive home, Theo told his mother about the old forecast. Most of its numbers had been close. What it could not have said was who would decide things, and in which rooms.