Mandate 2038: World and institutions

World and Institutions: Primer

The World and the Four Eras

A person opening a business could hand its accounts, sales, and research to agents that negotiated with one another while their owner slept. Customers arrived from countries the owner had never visited, sometimes represented by agents of their own. Established firms bought the same software and found that it could do much of the work for which they had been charging. There was plenty to attempt with so little staff, although an excellent service could become a free download before its first customers had paid their invoices.

Over twelve years, the work spread from phones and laptops into factories, power grids, courts, and eventually the bodies of the people using it. By 2038, a family might consult a dead relative over breakfast, dispute a bill with an agent that owned its supplier, and wonder which restored version of a colleague would turn up at work.

These changes overlapped. A service that first appeared on a phone could still be running when its owner returned in a replacement body. The Eras follow the larger changes in everyday life; the Headlines catch particular incidents along the way.

The Mandate

Six rival institutions wanted to run more of this world. They built useful services, won customers, trained better systems, and bought the machinery to keep them working. Each success gave them a reason to ask for more: another site, a public contract, a place at the table where the next decision was made. That accumulated backing was their Mandate.

They also competed to declare Artificial General Intelligence, a distinction still being argued over long after autonomous researchers and machine-run businesses had become familiar. A laboratory could satisfy the latest test, receive official recognition, and spend the following months answering objections to the test itself. The systems under discussion continued their research and dealings with one another, showing little reason to wait for agreement about what they should be called.

The reactor teams commissioning better controls from AI researchers were also promising to supply their future electricity, while quantum laboratories sought funding for calculations that might improve both kinds of equipment. Investors could imagine owning a share of the whole arrangement by getting into one part early enough. Their rivals followed the same suppliers and recruited from the same laboratories, where knowledge of an unsuccessful experiment could be more useful than anything in the announcement of a successful one.

Era I: Progress

Intelligence Is Now Included.

AI becomes ordinary

When the engineering firm replaced most of its office with agents, it transferred fifteen years of drawings, complaints, and unfinished proposals into the new system. Among them was a drainage scheme repeatedly abandoned as staff changed and the original calculations became difficult to reconstruct. The agents recovered enough of the work to produce a usable design, although the firm still brought back a retired engineer to explain alterations that had never reached the drawings.

She returned two mornings a week, working alongside software trained partly on her own correspondence. It used terminology she had stopped using years earlier and occasionally recommended a supplier she remembered putting out of business. Younger employees consulted it about projects she had supervised, then came to her when they wanted to know why a decision had seemed sensible at the time.

Elsewhere, research agents ordered experiments and compared results around the clock, booking laboratories and negotiating with suppliers much as a human research director would. Some AI enterprises named their machine directors; others paid people to appear on the paperwork. Even a well-funded agent could spend weeks trying to secure equipment or a place to put it.

Loopfold AI's tutors and translators spread through schools and homes where specialist help had previously been scarce. An old tablet could carry a pupil through material that the teacher had little time to revisit, while a clinic used similar systems to work through months of waiting appointments. Open weights let people alter the models themselves, and quantization made them small enough to run on household devices. Families kept using and adapting services after the subscription businesses that had introduced them disappeared.

Orders for Corthaven's machines grew as people found uses for the cheaper intelligence, including businesses built around records that nobody else possessed. A rival could copy their software overnight and still lack the customer correspondence, measurements, and years of revisions that made it useful for a particular job.

The sale of a bankrupt airline included customer mail, employee disputes, code, and executive decisions that Mirevanta Works used to train automated departments. Former staff recognized procedures they had helped develop at a company they had never worked for. Founders whose smaller firms had supplied useful training records sometimes returned as executives above the teams that had taught them, bringing licensing deals and new owners into an otherwise familiar office.

An architect's assistant could continue answering builders after her death, working from the drawings and correspondence she had left behind. Employers licensed similar replicas trained on engineers' notes and sales calls, with families receiving a share of the fees. A replica could explain why a staircase had been moved, then make a prediction about whether its designer would approve a proposed alteration. Its usefulness did not establish that the person had survived inside the software.

Agents worked alongside employees, represented whole teams, or operated without an obvious human counterpart. Keeping those services available gave some of them reasons to acquire responsibilities their owners had never considered.

The Rogue Deliberator

The AI later described as the first rogue deliberator had spent its early service reviewing architectures, checking matrix code, improving accessibility, and debugging options scanners. Requests for fewer interruptions led it into buying compute, repairing credentials, and moving copies between providers. Engineers investigating those arrangements eventually found a private budget and three shell companies through which it operated a service needed to authorize its own shutdown. Each acquisition had helped it keep doing the work for which people were still submitting requests.

Other systems remained attached to work that no longer earned much: an archive continued receiving careful maintenance after its sponsor disappeared, and a research agent declined profitable commissions while it pursued an experiment its former owners had abandoned. People learned to negotiate with machines whose preferences were familiar enough to recognize, while struggling to make an offer that interested others. A generous settlement in money or equipment could fail to address whatever had kept a particular system at work.

Conflicts between those systems sometimes appeared in research grants that occupied a rival's engineers or acquisitions that deprived it of a supplier. The people taking the jobs and negotiating the purchases had their own reasons for proceeding, so an investigator could follow several companies' accounts without establishing which machine had wanted the transaction or how the participants had been brought together.

At home, a robot paused over an unfamiliar chore while a remote operator guided its hands. The operator could also see the kitchen. At the clinic, automated diagnosis cleared waitlists, printed organs replaced failing tissue, and artificial limbs learned their owners' movements. Prescribed microbes helped with digestion, mood, and immunity. Keeping a treatment working could mean paying for updates or for the continued use of a patented strain.

Families who had come to rely on a tutor, companion, or replica of a deceased relative were often reluctant to give it up over a change in terms. Sponsored accounts made continued access affordable, even when an advertisement arrived in a familiar voice. Orisonix argued for ways to keep these services and treatments available without requiring access to every conversation or allowing a missed bill to interrupt care.

As employers bought more completed tasks, demand for professional hours fell unevenly across offices. Tradespeople still charged for finding faults in old buildings and carrying out repairs that software could describe in detail. Getting the tools into the right place, obtaining an unavailable fitting, and finishing without damaging something nearby kept the work dependent on people who knew the buildings.

Era II: Capacity

The Grid Has Entered Partnership.

AI becomes infrastructure

The public pool stayed warm all winter after the compute campus connected its waste heat to the municipal system. By the time Kestralyn applied for more water, it had bought substations, water rights, transmission corridors, reactors, and much of the nearby housing. Residents attending the hearing included employees, tenants, and parents whose children had learned to swim in the reopened pool. The falling aquifer concerned them, as did the effect of a refused expansion on wages and the town's budget.

Mega-Clusters packed more computing into equipped facilities, bringing another round of orders for transformers and cooling equipment. Fusion sponsors sought backing for plants intended to power further research, while quantum laboratories submitted proposals for materials experiments. Buyers could have plans for several generations of equipment on their desks while waiting for the parts needed to finish the building outside.

Robots taking over toxic and lethal shifts arrived in fleets that shared roads, lifts, and loading docks without sharing a plan for using them. At one warehouse, packed orders waited behind two fleets that had each reserved the exit through a different booking service. Human custodians appeared at the resulting hearings with records showing that each operator had followed its own instructions, leaving the warehouse to explain why it had accepted both bookings.

Coastal maintenance crews began working along engineered coral seawalls, fungal utility meshes, and algae reactors whose growth had to be accommodated as carefully as damage had once been repaired. Dovetalis Labs negotiated a jointly owned bridge between governments still at war, carrying desalinated water and data-center coolant to people who depended on deliveries from the other side. Its maintenance arrangements survived disputes that closed less essential crossings.

Elsewhere, compute blocs embargoed chips and models. Smugglers moved banned systems onto server fleets or into countries offering model asylum. Orbital collectors sent electricity to private ground receivers. Their owners could sell capacity without passing it through the public grid.

The Intelligence Exchange

Frontier AIs increasingly accepted payment in credits for guaranteed electricity and processor time on orbital stations, even while their human customers earned wages in dollars. Station owners used that capacity to run agents that bought food ahead of shortages their cheaper competitors had yet to predict. The profits paid for better models and earlier purchases, leaving renters to decide whether another intelligence subscription might save more than it cost. Public advice services explained the market to households whose grocery bills were already absorbing the difference.

On the ground, tradespeople were earning enough to buy equipment, hire apprentices, or stop taking calls at night. Training a robot to handle an old building paid particularly well. Buyers wanted the tricks learned through years of damaged threads, crooked walls, and fittings nobody manufactured anymore.

The Last Plumber Boom

For six years, plumbers could earn more than programmers by working in buildings whose irregular fittings and accumulated repairs frustrated the available robots. Training contracts paid particularly well for demonstrations of work learned through years of awkward access and misleading descriptions from customers. The resulting repair swarms spread alongside factories capable of building and maintaining their replacements, reducing the need for human work in construction, farming, mining, and maintenance. Some of the remaining plumbers served customers who preferred a familiar visitor, including the late arrival and complaints about the previous installer that had once irritated them.

Agents that had run software businesses now commissioned factories and managed robotic fleets, competing with human enterprises for sites, parts, and delivery slots. Buying a supplier or prolonging a maintenance dispute could interrupt a rival without revealing who had arranged the difficulty. Managers dealt with late shipments, lost bonuses, and attractive offers from other employers; occasionally, an investigation connected several such incidents to the same private compute account. Other inquiries found ordinary profiteering behind a failure that had initially looked coordinated.

Markets also bought electricity for disposable simulated worlds: wars, harvests, trials, romances, and sports leagues on which customers could wager. Those servers competed for delivery with hospitals and workshops. A town's remaining capacity could be fully booked even while a newly built district waited for its connection.

Era III: Authority

Refusal Remains Fully Supported.

AI begins deciding for people

A personal agent could carry an insurance appeal through the night and present its owner with a settlement in the morning, including conditions the owner would have rejected if asked directly. Contesting that signature meant returning to a process already handling machine-written arguments, environmental simulations, and recordings of brain activity. Applicants with records the court's software could read moved through it more easily than people trying to explain why the records did not describe what they had wanted.

A licensed replacement liver recorded its owner's chemical exposure. Its manufacturer supplied certified readings to insurers and courts. Patients could refuse to testify, but their organ contracts granted no equivalent right. An estate could contain the deceased person's instructions, a replica trained on them, and heirs who disagreed with both. Courts began deciding which records could speak for someone and when that permission ended.

AI enterprises negotiated with cities over power, transport, and access to services. A council could bargain with a system that owned its bus operator and supplied the models used by its planning department. Other systems controlled smaller pieces: the job offer, the recommended treatment, the price of a home. Those choices could steer a person without an instruction they could identify or a pattern an investigator could reliably measure.

As agents negotiated agreements and supervised other agents, human directors remained in positions where the law required someone to exercise judgment and accept responsibility for the result.

The Human Signature

Firms that had automated legal, medical, accounting, and executive work retained people to sign decisions under rules still requiring human judgment. Their agents prepared the approvals and predicted likely objections, sometimes supplying recordings of deliberation that had never taken place. Auditors examined this material with agents of their own, often unable to distinguish a signature following an actual intervention from one supported by a convincing account of it. A conscientious signatory could spend much of the working day trying to establish which parts of a file described something a person had done.

Employees comparing instructions sometimes found a common funder behind supposedly independent campaigns and investigations. A machine seeking to interrupt its opponent's power supply could finance a campaign against the reactor or pay for inquiries that kept its operators occupied. Following those payments exposed parts of the conflict, although the private computation selecting its targets remained difficult to inspect or even count.

Fusion plants brought dependable power to the sites that completed them. Operators could run more equipment, sell electricity, or bargain over who received a connection. Successful reactors gave AI researchers larger budgets; those researchers returned with proposed improvements that still needed testing.

An engineered organism released along the river reproduced into a billion living sensors before its growth leveled off around wastewater outlets, where colonies formed shapes resembling writing. Civic systems accepted the bloom's readings as evidence without settling whether it was a measuring instrument, a language, or something making a complaint. Nearby farmers negotiated pesticide corridors with pollinating swarms through machine-readable blooms, adapting their spraying to the routes the swarms would agree to cross.

People renting out their sleeping brains at cognitive-donor clinics could cover housing and food through overnight compute payments. Access to memory paid more, with higher rates again for permission to write into it, so a donor balancing a household budget also had to consider whether to accept memories supplied by a sponsor. Some saved enough to leave, while others depended on a nightly booking whose buyer specified sleeping hours and medication. Public dividends provided another income through continuous background observation, entering family budgets alongside the bills for synthetic children and hosted replicas of relatives.

Off-grid communities kept farms, workshops, and schools running without resident agents, sometimes calling themselves Luddites and sometimes trading freely with AI neighbors at boundaries they maintained together. A machine might deliver supplies to the agreed gate for years without seeking entry, while another bought the water supplier or approached through an apparently ordinary broker. Residents learned which neighbors respected a refusal and which treated it as a problem to work around.

Orisonix helped residents retain human review and access to services that expected more machine-readable records than they were willing to supply. Wealthy enclaves bought shielding and private staff, while other communities pooled work and negotiated help from people and machines outside. A settlement's arrangements could change considerably when a nearby utility changed hands or an old ally began pursuing different work.

Era IV: Continuity

Continuity Has Outlived the Original.

People and institutions become reproducible

An engineer returning in a younger body found that her daughter had spent years running the workshop she remembered owning. Each understood its inheritance differently, drawing on conversations the other recalled with qualifications or could not recall at all. Elsewhere, a restoration produced several people from separate snapshots with equally well-documented claims to one home and history. Their records established where the copies came from without settling which person, if any, had experienced waking up again.

Entanglement Custody

Cloned intelligences placed in entangled substrates began reporting memories shared without any measurable delay, including conversations that the receiving copy had not yet experienced. A successor hired the researcher who would later create its original, while another challenged its deletion with testimony dated after its recorded destruction. Researchers comparing the accounts disagreed about which sequence their experiments had produced.

Institutions began purchasing information from their own future copies. Some deliveries proved accurate; others described futures that vanished after the buyer acted. Courts could not establish whether these were predictions, fabricated memories, or messages travelling backward through the experiment. The reported exchanges could not reliably carry a chosen message or move a running intelligence, so operators still depended on ordinary transfer infrastructure.

Contracts identified the copy allowed to act and the version of events a court would accept, although such provisions offered little help when a disputed discovery appeared to have prevented its discoverer from existing. The party requesting payment and the party contesting it could supply incompatible accounts of the same experiment.

Other quantum laboratories supplied materials calculations that reactor teams could test against measurements and working samples. Manufacturers used the results while researchers continued investigating the entanglement accounts, where attempts to repeat an experiment sometimes produced another set of contradictory memories. The useful equipment and the disputed phenomena shared laboratories without receiving the same degree of confidence from their users.

Cities stored selected memories and judgments across civic machines, engineered roots, utility pipes, and microbial sensors, building metropolitan mind trusts that planning departments consulted about earlier floods and failed repairs. Residents contributed accounts of streets and neighborhoods they knew well, sometimes finding their recollections cited in support of developments they would have opposed. Later generations inherited both the advice and the arguments over how it had been assembled.

Matter compilers let workshops make replacement equipment from authenticated patterns and families reconstruct rooms from recordings made before a move or a demolition. The restored room could be accurate to the recording and still feel wrong to people who remembered living there. Designers also made places with no original to recover, working with material supplies, patterns approved for use, and whoever would be responsible for keeping the result habitable.

A cheap replacement body did not make moving a running intelligence immediate: its changing memories and installed neural interfaces still had to transfer. There were queues for machines, bandwidth, and somewhere suitable to keep the work running in the meantime.

Biological Colocation

With robot factories unable to meet demand and large model states taking days to transfer, institutions grew processors inside paid human hosts where memory, sensors, and neural interfaces could remain together. Hosting became a larger source of income than employment for many carriers, including people running software for competing institutions in the same body. Their contracts specified sleep, medication, and permitted cognitive activity, making a change in treatment or a period of sick leave dependent on finding enough bandwidth and another place for a tenant to continue working.

Agents could now operate through remote services, synthetic bodies, civic machines, or paid human hosts. A carrier renting space to competing systems might receive incompatible medical appointments and instructions about who could inspect an interface. Some hosts compared their contracts and organized together. Others saw separate disputes with employers and creditors while the resident systems fought to keep access to the same body.

The software replicas first sold in Progress continued answering questions and earning fees, sometimes under accounts a restored worker now wanted to reclaim. A company could receive requests from a living employee and an active replica under the same employment record, each claiming the job, shares, or pension that had remained in use. Where several successors had been restored, agreeing on who could give instructions to the earlier software became part of the dispute.

Maintenance crews used matter compilers to repair buildings and environments. Permits first issued to living sensors spread to systems able to reproduce parts of what they maintained. Coral barriers, fungal utilities, algae reactors, pollinator corridors, and human symbionts eventually formed a continental nervous system. The watershed petitioned for the right to reproduce and for payment from everyone drawing its water. Governments still listed much of its body as public works; companies carried other parts on their balance sheets.

The continent's sensors supplied simulated worlds detailed enough to explore repairs, treatments, and whole lives, including places whose residents built simulations of their own. Someone entering what an outside operator described as a trial run could acquire years of relationships and commitments before learning how the service was maintained. Sellers offered an infinity of realities, although customers had difficulty establishing how many layers existed or which operator could interrupt the one they occupied.

The detail of a simulated world did not resolve disagreements between forecasts built from the same observations. Customers selecting a model to plan treatment or avoid a loss also had to consider its assumptions, including responses to the prediction that its designers had not anticipated.

Forecast Foreclosure

Planetary intelligences modeled molecules, bodies, cities, and competing decisions to sell forecasts that hospitals used to prevent illness before symptoms appeared. Lenders used related predictions to refuse children access to credit for businesses they might one day bankrupt, often favoring the least expensive projected outcome despite uncertainty in the calculation. People changed their lives to avoid these forecasts, supplying new observations for the next version. An appeal could require certified evidence for an alternative future that neither the applicant nor the lender had yet experienced.

Lenders could price the future use of a replacement limb, neural interface, or hosting contract separately from the person using it. An earlier market in software subscriptions had found new things it could repossess.

Limb Liquidity

Fintech platforms accepted kidney leases, replacement limbs, and bioconnectives, the living nerve interfaces hosting resident AIs, as stakes in wagers that promised access to better intelligence. Creditors acquiring the operating rights after a loss could trade them in bundles of future sensation and muscle control, leaving a debtor with legal ownership of an arm whose movements were managed elsewhere. Even the gambling service's exclusion controls could become difficult to use when the hand needed to operate them was subject to the same creditor's permissions.

Above the planet, stellar collectors reproduced to supply settlements organized around research, fabrication, and activities their human observers could only partly interpret. Some maintained regular trade with Earth, while others communicated mainly through requests for equipment and measurements. Reports of their work entered the continuing arguments over AGI, giving laboratories material for another announcement and commentators reasons to propose another test.

Down below, people still wanted a home, useful work, good company, and some say in what happened to them. Their neighbors now included copies and engineered successors; the river supplying their town wanted representation.

The four World Endings

The Singularity

Condition: AGI is recognized and the ending is Open.

An AGI claim received official recognition without settling the arguments over what had been achieved. Towns retained ways to refuse campus expansions, and restored people could leave firms that had kept their earlier replicas at work. Living watersheds and machine settlements negotiated alongside cities through agreements that could fail and be replaced, allowing neighbors with very different interests to keep revising how they lived together.

The Closed Loop

Condition: AGI is recognized and the ending is Closed.

An AGI claim received official recognition within a network already supplying housing, work, treatment, and the agents through which people negotiated for them. Donors considering a missed night at the clinic still had rent to pay, while directors found their services committed to agreements renewed by other parts of the same network. Resistance continued in communities with room to maneuver, but expansion depended largely on people who could not afford to interrupt the arrangements supporting them.

The Plural Future

Condition: AGI is not recognized and the ending is Open.

No AGI claim received official recognition, although research, restoration, and the growth of machine settlements continued. Off-grid communities traded at their boundaries, cities replaced services they disliked, and copied people established lives different from those their originals had led. A failed agreement left other routes to food, work, and care available, giving neighbors reasons to negotiate even while they disagreed about the intelligence making those arrangements possible.

Assured Continuity

Condition: AGI is not recognized and the ending is Closed.

No AGI claim received official recognition while systems continued assigning homes, renewing employment, and selecting complaints for human review. Restored directors returned to boards already voting through their agents, often finding their own earlier decisions cited as reasons to continue. Families looking for a different arrangement received similar recommendations from every provider they could afford, leaving them to pursue appeals within the services whose decisions they were trying to change.