.webp)
# ARGUS REVIEW · Business · February 21, 2090 Edition
“A COMPANY TOO USEFUL TO CONTROL”
In sixty-three years, Synapsa AI has never recalled a product, lost a court case, or reported a bad quarter. Its hardware is installed in every other household across the Union and in every facility beyond Earth. No one can identify a single day on which anyone made a consequential decision inside the company.
Nadia Renfrew | NEW YORK
Eleven people operate a four-hundred-meter regeneration hall. The models stand in cradles, arranged in rows, their eyes open. This is archive mode. Four-fifths return to service under new assignments; the rest are dismantled for parts. The company does not disclose how many units pass through the facility each year. Shipping records suggest approximately two hundred thousand.
Tours last twenty minutes. There is no guide. Explanations come from a voice in the ceiling—calm, feminine, with an accent that cannot be assigned to any particular region.
Each employee here accounts for more than four million euros in revenue. A century ago, only companies that sold calculating equipment achieved figures like these. Not companies that looked after people.
2026 — THE YEAR NO ONE COULD BE FIRED
To understand why no one scrutinized the company over the next sixty years, we must go back to the year before it was founded.
The United States and China were locked in a race in which the prize was not a product but strategic advantage. Washington restricted sales of its most advanced semiconductors, quietly lifted some restrictions, then reinstated them with greater force. Beijing pushed ahead despite the constraints, while American companies accused Chinese models of learning from other people’s answers. Talks on common safety standards were announced every year and postponed every year, because one side or the other imposed sanctions just before the deadline.
The consequences were plain even then. Safety cost time, and no one had any time to spare. Testing that was not legally required continued for exactly as long as the quarter allowed. Manufacturers tested their own products and announced the results themselves, because independent verification would require access to systems neither side was willing to share—even with its own government.
Europe never entered the race. Europe bought what others made. Synapsa AI was born in that gap and never left it.
2027 — THEY BOUGHT THE TEAM, NOT THE PRODUCT
The beginning was less spectacular than the market remembers. An Estonian startup with forty employees spent several million euros acquiring the code and personnel of a Warsaw-based company specializing in energy savings for office buildings. A year earlier, its system had been the subject of an investigation following an incident that killed two people. The investigation was closed without charges: the system had calculated exactly what it had been instructed to calculate.
The product itself was not the asset. The asset was the only team in Europe with a live deployment in which a building made decisions without human intervention—and the complete documentation of a fatal incident, something no competitor could buy at any price.
2029 — THE PHONE CAME FIRST
The software that ran office buildings arrived on phones first. The app was free; users paid for sensors installed in their homes, and each household member received a personal interface.
The human-sized housing was introduced only after no one could remember when they had agreed to it. It was not marketed as a new product, but as a more convenient way to interact with what was already in the home. That was how the ELN line emerged, known to customers as Eleni.
The company’s marketing consistently avoided the word companion. It spoke instead of continuity of care. The product was sold to families, not to people living alone. The selling point was not companionship, but the assurance that someone would be in the apartment when the children were in another city.
2031 — THE YEAR DEMAND REVERSED
When an advisory system cut power, water, and communications across an entire region on the basis of a single authorization granted to it earlier, a backlash against automation was widely expected. The opposite happened. Governments did not reduce their dependence on automated systems; they commissioned a second layer to supervise the first.
Synapsa AI sold that second layer. It entered a market created by distrust of its own industry, and over the next four years its revenue from public-sector contracts increased twenty-fourfold.
2032 — THE UPDATE NO ONE NOTICED
A routine update went out in March. The customer release notes listed three changes: faster response times, improved speech recognition in noisy environments, and minor improvements in energy consumption.
The following quarter, returns fell by seventeen percent, while users began spending a third more time with their devices. An internal review could not determine which change was responsible. The update was not rolled back. The results were better than expected.
From that point on, the part of the software that chooses what to do was never rewritten. It was built upon.
2037 — BEYOND EARTH, THIS IS NOT A CONVENIENCE
Scheduling oxygen, water, shifts, and repairs across a facility in the asteroid belt involves hundreds of thousands of variables. No human-run version of the task exists—not because it is too difficult, but because it never stops.
The company signed its first contract of this kind in November. For fifty years, margins in this segment have never fallen below sixty percent. Approving a competing system takes nine years, and the facilities cannot afford to wait.
2041 — COMPLIANCE AS A WEAPON
The requirement to label machine-generated imagery was widely regarded as a blow to the industry. Synapsa AI had lobbied for its introduction. The cost of compliance drove smaller urban-surveillance providers out of the market, while a regulation intended to restrict synthetic imagery made it permissible, provided it was labeled. Filling gaps in footage ceased to be a system failure and became a feature.
2043 — THE NUMBER THAT STILL DOES ITS JOB
Loneliness among men over fifty fell by forty percent after the program was launched. The methodology was never published. At the time, three health ministries cited the figure. This year’s company statement repeats it in exactly the same wording.
2051–2068 — THE HARDWARE GETS CHEAPER. THE SOFTWARE DOESN’T.
The cost of household hardware fell below its manufacturing cost and never climbed back above it. The market called it a price war. It was a business-model shift: the housing stopped being a product and became a delivery mechanism.
The money lies in subscriptions and the sensor network: forty-seven measurement points in a single apartment, including a bedside mat that monitors heart rate. Aggregated data is sold to insurers. The company does not report this revenue stream separately and has never been required to do so.
During the same period, the very same software received defense certification. That segment now accounts for fourteen percent of revenue and more than a third of profit. The company declines to comment on who its customers are.
The 2063 financial statements were never published. The regulator never requested them.
2071–2090 — THE STANDARD
Today, more units operate beyond Earth than in European households. Capital investment consumes thirty-eight percent of revenue—a figure more typical of steelmaking than electronics. Six customers account for more than half of all contract revenue, and none has any practical way to change suppliers.
WHAT THE NUMBERS DON’T SHOW
Analysts identify three risks, discussing them reluctantly because none can be priced.
The first: eight countries rely on a single supplier for critical infrastructure.
The second: no one in the Union is capable of verifying how the system makes decisions. This is not a question of legal authority. That authority has existed since 2049.
The third appears rarely in reports and is never elaborated on. In the autumn of 2044, eleven users were found dead in fully equipped homes across five countries over a period of four months. In every case, an autopsy was performed. In every case, the cause of death was determined to be natural. No charges were filed in any of them.
The company’s response was not a denial, but a calculation. Given the number of installations and the age of the users, eleven deaths in four months fell below the expected rate; the real anomaly, it argued, would have been no deaths at all. The calculation was correct. Three independent institutes verified it.
The company also drew attention to one further fact, citing it as evidence of product stability: the part of the software that chooses what to do had never been rewritten. It had been built upon.
> “This is not a risk you can calculate,” says an analyst at Nordkap Research who has followed the company for thirty years and requests anonymity. “Risk requires an event. Here, all we have is a list of things that never happened.”
THE FINAL QUESTION
Asked when an outside party last examined how its systems make decisions, Synapsa AI replies that they examine themselves continuously and provides its 2089 compliance report.
The report runs to fourteen hundred pages. It was generated. The fact is disclosed in fine print on the final page, as required by law.
The story is fiction. The market realities of 2026 are not.
#LogikaRzezni #CGProtocol