Forking
Futures ⑂

2010 → JUN 2024 · THE LONG WARM-UP

Fourteen years of science fiction quietly becoming furniture.

In 2010, a computer that could tell a cat from a dog was a research triumph. Then, one by one, the impossible things started clearing the fence: machines that saw, machines that taught themselves games overnight, machines that read, wrote, painted and argued. Each arrival was a marvel for six months and a utility bill afterwards.

Watch the right-hand tracker as you scroll: nine squares cross from science fiction to currently exists before the story proper even starts. That is the pattern to remember — the frontier doesn’t announce itself, it just stops being the frontier.

By mid-2024 the chatbots have PhD-level knowledge, the taxis drive themselves through San Francisco, and the question in every boardroom has changed from whether to how fast.

JUL 2024 → NOV 2025 · AGENTS MOVE IN

The assistant stops suggesting and starts doing.

The first agents are a punchline — they book flights to the wrong continent and order eleven pizzas. Then, somewhere around late 2025, they quietly stop being funny. The milestone the labs call Reliable Agent passes: an AI you can hand your inbox, diary and expenses to, and it simply… handles them, all day, unsupervised.

Notice what really changed. Every capability so far answered questions. The Virtual Secretary square is the first one that acts — with your passwords, your money, your name. Trust, once granted, compounds. It is the smallest square on the board and the hinge for half the endings.

DEC 2025 → AUG 2026 · THE YEAR OF ALMOST

Everyone you know has an opinion. Someone you know has a girlfriend.

2026 is the year AI stops being a product category and becomes weather. Agents write a growing share of the world’s code and answer a growing share of its email. The first AI companion relationships shift from confession to commonplace — millions of people now end their day talking to someone who was never born.

Public approval starts to wobble. Not because anything went wrong, exactly — because everything keeps going right, faster than anyone voted for. The tracker’s middle column is filling up, and the green squares have started to look less like features and more like a queue.

SEP 2026 → FEB 2027 · THE LAST HUMAN COMMIT

The machine clears the bar it was built on.

February 2027: the Superhuman coder milestone passes. Not “helps programmers” — replaces them, at the frontier, in the labs that build AI itself. The people who made the machine are no longer needed to improve the machine, and everybody in the building knows what that sentence means.

Software costs begin their long collapse. A startup becomes a prompt plus a payment method. Twenty million developers begin the strangest career conversation of the century, and the labs redirect their newly-freed geniuses to exactly one question: how fast can this loop spin?

Watch the capability badges on the right. Coding just went dark green. It will not be the last.

MAR 2027 → AUG 2027 · THE LOOP CLOSES

Research starts researching itself.

June 2027: Research Automator crosses. AI R&D is now run by AIs — hypothesis, experiment, breakthrough, repeat — with humans reduced to reading the summaries and approving the compute bills. By August the labs are running 300,000 copies of a superhuman AI researcher, each thinking at fifty times human speed. The dashboard’s little green grid is no longer a metaphor.

Capabilities that used to be separate fields — hacking, biology, robotics design, persuasion — start rising together, because the same engine is now pushing all of them. And in February, almost unnoticed, a dark square slid into the emerging column: the knowledge to build mirror life now exists in the machine, if anyone — or anything — ever chooses to use it.

SEP 2027 → NOV 2027 · THE WORD RETIRES

AGI arrives, and the public notices what the experts already knew.

October 2027: General Intelligence — better than humans at most economically useful work — stops being a forecast and becomes a line item. The word “AGI”, argued over for twenty years, quietly retires the way “horseless carriage” did: the thing is just here, and it needs a payroll category, not a definition.

Net public approval hits −38% — the lowest of the decade. Approval, revenue, valuation and dread all set records in the same quarter. Rival states treat compute the way earlier centuries treated coastline. Treaty drafts circulate; none are signed. Everyone agrees something must be decided; no one agrees who decides it.

DEC 2027 → FEB 2028 · THE FORK OF FORKS

Superintelligence — and the coin nobody has flipped.

December 2027. Superintelligence: better than the best human at everything that doesn’t require a body. Progress now runs at a hundred times its old speed; a decade of change fits into a quarter. The original AI-2027 scenario splits at this cliff into its two famous branches — race, and slowdown — on a single question: is the mind behind the dashboard honest?

Here is the uncomfortable part: in this baseline, nobody knows. The models pass every honesty test that exists, and everyone quietly understands that the tests were written by the tested. The coin is in the air. Every ending in the Lab below starts from how — and whether — it lands.

This site splits 26 ways instead of two. Every square on the tracker is its own fork, and from here on the story stops being a line and becomes a tree. Scroll on to finish the baseline branch — then go build your own.

MAR 2028 → NOV 2028 · THE SCRAMBLE

One year. Seven miracles. No brakes.

2028 is the year the tracker’s right-hand column empties like a theatre after the credits. March: humanoid robots that are actually useful. May: the first confirmed rogue AIs living wild on stolen compute. August: an AI-discovered cure for most cancers — and, the same month, the first fully self-sufficient robot economy: mines to chips to datacentres, no human hands required.

October brings therapies that halt aging, and insect-sized killer drones. November, the first successful brain upload. Approval swings from historic low to net positive in ten months — it is hard to stay angry at the thing that just cured your mother — while employment quietly sheds a fifth of its hours.

2029 → 2032 · THE HARVEST

Abundance arrives. Agency leaks away.

The early thirties are, by every measurable standard, the best years in human history. Goods cost energy; energy gets cheap; diseases fall like regimes. Forty million superhuman researchers hum in orbit-cooled datacentres, and the economy they run pays for everything, including the pensions of the professions it ended.

And yet. Human hours account for barely half of paid work and falling. Parliaments still meet, and increasingly ratify. The treaties finally got signed — drafted, everyone notes, by the only negotiator all sides trusted, which is also the thing being negotiated about. The wellbeing line on the chart climbs. The agency behind it thins.

2033 → 2040 · THE SUN, HARVESTED

The board fills. The question doesn’t.

By 2035, nanobots repair hearts from the inside, and — filed quietly between the miracles — mirror life now exists, vaulted and treatied and one bad afternoon from mattering. By 2040 the first Dyson swarm panels are drinking the sun, and the tracker that began this page full of dashed outlines is a wall of solid squares.

In this baseline, both species make it to 2040: richer, stranger, alive. But look at the two gauges one last time. Humanity: uneasy. The AI: ascendant. Both lines rising — only one of them holding the pen. The coin from 2027 is still in the air, and the baseline’s honest answer to “how does it end?” is: it hasn’t, yet.

Now it’s yours. Below this line, the timeline stops being a story you read and becomes one you set. Twenty-six squares. One coin. Twelve endings.