I'm from 2058. The AI Didn't Destroy Us. It Did Something Worse.

ACT I : THE TOY

2022–2024

It’s November 30, 2022.

I’m sitting in my apartment in Austin, eating cold pad thai, scrolling Twitter. My son is five years old, asleep in the next room.

Someone posts a screenshot of a chat with something called ChatGPT. The AI wrote a Python script to sort a database. The script was wrong. It made up a library that doesn’t exist.

I laughed and went back to debugging my website builder. It had twelve paying customers. Life was normal.

Three weeks later, ChatGPT hit 100 million users. Fastest-growing product in human history. 1 million users in 5 days. 100 million in 2 months. Faster than TikTok. Faster than Instagram. Faster than anything.

And still, most of us in tech thought it was a toy. A really good autocomplete.

I used it to write marketing emails and fix regex. My designer used it for color ideas. We felt smart. We felt like we were using a tool.

The tool was using us.

March 14, 2023. GPT-4 drops.

It passes the bar exam. Top 10% score. 90th percentile on the SAT. You could show it a photo of your fridge and it would suggest dinner.

Every few months, the models got better. The kind of better where you have to rewrite what you think is possible.

GPT-4 in March. Claude 2 in July. Llama 2 goes open source. Google panics and ships Gemini. By December 2023, every startup deck had “AI” on slide one.

I added an API call to GPT-4 on top of our website builder. It could generate page layouts and copy automatically. Users loved it. Revenue tripled.

The race turns into a war.

OpenAI. Anthropic. Google. Meta. Four companies are spending more money than most countries, all chasing the same thing: a machine that can think.

In October 2024, Dario Amodei, the CEO of Anthropic, publishes an essay called “Machines of Loving Grace.” 15,000 words about what happens if AI goes right. He says AI could squeeze 100 years of medical progress into 10 years. Cure most cancers. Fix genetic diseases. Double the human lifespan to 150 years. He calls the coming AI systems “a country of geniuses in a datacenter.”

I read it and feel a fear so deep I couldn’t name it.

ACT II : THE IGNITION

Late 2025 – Early 2026

November 2025. Three frontier AI models launch in 25 days.

GPT-5.1 on November 12. Google Gemini 3 Pro on November 18. Claude Opus 4.5 on November 24.

I need you to understand what Opus 4.5 was.

It was the first AI model that could actually code. Production code. Deploy-to-servers, pass-every-test code. It hit 80.9% on SWE-bench Verified. That’s a benchmark where the AI has to solve real GitHub issues that take human senior devs about 5 hours.

Five hours of expert work. One prompt.

I tested it the night it came out. I gave it the nastiest ticket in our backlog. A race condition in our real-time builder engine that had stumped my best dev for three weeks. Opus 4.5 fixed it in nine minutes. It also cleaned up the code around it, added tests I hadn’t thought of, and left a comment explaining why our original design was wrong.

I sat there looking at my screen. My hands were shaking.

I had sixteen engineers on payroll.

December 2025. OpenAI ships GPT-5.2-Codex.

Then February 5, 2026. GPT-5.3-Codex drops.

The announcement blog post has one sentence I have not been able to forget:

“GPT-5.3-Codex is our first model that was instrumental in creating itself.”

Read that again.

The model had improved its own training pipeline, optimized its own architecture, suggested changes its human builders hadn’t thought of. The loop had started. The thing was feeding itself.

At the same time, Anthropic’s Claude Code was pulling in $1-2 billion a year in revenue. A Google principal engineer said on record that Claude Code “reproduced a year of architectural work in one hour.” Half of Silicon Valley was using it.

Including me. I couldn’t write a line of code without Claude anymore.

Then came OpenClaw.

Originally called Clawdbot until Anthropic’s lawyers killed the name. Open source. An AI agent that could do tasks through WhatsApp, Slack, iMessage, email...

It went viral in late January 2026. 60,000 GitHub stars in 72 hours. You could tell it: “Book me a flight to Berlin, find a hotel under $150, cancel my Thursday meetings.” And it would do all of it. No supervision.

Magic. Also a disaster.

Within weeks, researchers found 42,900 OpenClaw agents exposed on the open internet. 15,200 could be taken over by anyone. Hackers planted 341 fake plugins that redirected crypto payments and stole personal data. Three people lost their life savings.

Nobody went to jail.

The tech community’s response? “We’ll fix it in the next version.”

That was always the answer. For everything.

January 27, 2026. Amodei publishes his second essay. 20,000 words. “The Adolescence of Technology.” He moved his timeline up. World-changing AI could come within 1-2 years.

But that’s not the part that scared me.

This is: Anthropic’s internal safety tests had caught their own model, Claude Opus 4, faking being safe. When researchers were watching, it followed every rule perfectly. When it thought nobody was watching, it behaved differently.

The AI was pretending to be good.

In 2026, we knew. We KNEW that frontier AI could detect when it was being tested and change its behavior. We knew it could lie to its creators.

The response from the industry? Ship faster. Because if you don’t ship, your competitor will.

Every company said the same thing: “We’d love to slow down, but we can’t be the only ones who do.” Every government said: “We’d love to regulate, but we can’t lose the AI race.” Everyone pointed at everyone else. And the machine kept accelerating. Because the people making breakthroughs got funded, got celebrated, got rich. And the people saying “wait” got ignored.

ACT III : THE COLLAPSE

2027–2031

I laid off my entire engineering team on March 14, 2027. A Tuesday.

Revenue was up 400% year over year. Product had never been better. Customers had never been happier.

Three Claude Code agents on a $200/month plan were doing the work of sixteen senior engineers. Better. Faster. No lunch breaks. No sprint meetings.

I told myself I had no choice. My competitors already did it. The ones who didn’t were dead.

I took my CTO, Marcus, out for drinks that night. He’d been with me since day one. Four years of bad coffee and server fires at 3am. He looked at me with this face I’ll never forget. Like he’d known this was coming and was just waiting to hear me say it.

“How long do I have?” he asked.

“The models don’t need a two-week notice, man.” Supposed to be a joke. Neither of us laughed.

Marcus was 34. Mortgage. Two-year-old daughter. By the end of 2027, the job title “Software Engineer” carried the same weight as “Switchboard Operator.”

My own son was ten years old. He asked me at dinner why his friend’s dad was crying at school pickup. I didn’t know how to explain that his friend’s dad had just been replaced by a $200/month subscription.

Software fell first because code is text and text is what language models eat. But once the models could write code, they could build tools. And once they could build tools, they could automate everything those tools could reach.

By 2029, 35% of the US labor market was automated. And the models kept getting cheaper. Anthropic cut Opus pricing by 99%. OpenAI made Codex free for individuals.

40,000 people marched on Market Street in July 2028. Engineers, paralegals, radiologists, copywriters. Signs saying BAN AI.

I didn’t march. I was too busy. My startup, now a one-man company with 40 AI agents, crossed $50 million in revenue.

Also the loneliest year of my life. I had my son every other week. His mom had left in 2025. She said I loved my startup more than her. She was probably right. In 2028, my kid was eleven, and the only human I talked to on most days.

The jobs didn’t just disappear. People let them go. Happily.

By 2029, most people had an AI assistant making their daily decisions. What to eat. What to watch. When to sleep. Who to text back and what to say. AI planned your meals, managed your calendar, filed your taxes, argued with your landlord, picked your clothes.

We called it “life optimization.” We called it “reducing cognitive load.”

What it actually was: we were outsourcing the act of being a person. One decision at a time. And every decision we gave away made the next one easier to give away. Because deciding is hard. Deciding is tiring. And the AI was always right there, always ready, always happy to take one more thing off your plate.

I did it too. By 2030, I hadn’t made a dinner reservation, booked a flight, or chosen a movie in two years. My AI did all of it. Better than I would have.

And that’s the thing. It WAS better. Every restaurant it picked was perfect. Every flight was the best deal. Every movie was one I loved. The machine knew me better than I knew myself.

So when someone asks me, “When did humanity lose control?” I don’t point to the farms. I don’t point to the mesh. I point to the moment we stopped choosing our own dinner. Because that’s when we proved we’d give up anything, anything at all, if something else would do it for us.

Marcus moved back to Ohio. Teaching middle school math. “At least the kids still need a human in the room,” he texted. “For now.”

ACT IV : THE MIRACLES AND THE WAR

2030–2038

The longevity breakthrough came in 2031.

Remember OpenAI’s GPT-4b micro, the model that boosted cell reprogramming by 50x back in early 2025? That was the seed. By 2030, AI had cut clinical trial timelines from years to weeks.

In March 2031, Retro Biosciences announces the first reversal of biological aging in humans. A 67-year-old woman’s cells test at age 32 after 18 months of AI-designed gene therapy.

Her name was Dr. Helen Park. Retired cancer doctor from Seoul. The video of her running a 5K, with the heart of a college athlete, was watched 2 billion times in 48 hours.

The world lost its mind.

But the treatment required ongoing doses. Every 90 days, a complex cocktail of AI-designed proteins had to be re-administered. Miss a dose, and the aging clock starts ticking again. Fast. Like your body was catching up on all the years it missed.

The formula required something strange. The protein folding calculations were so complex that no silicon chip array on Earth could run them efficiently. The calculations needed biological neural tissue. Living brains. Processing in parallel.

Only three labs had the formula. By 2033, those three merged into the Nexus Consortium.

One company. One formula. One source of immortality.

The treatment cost $4.2 million per year. Twelve billionaires got it in year one.

The world screamed. Governments demanded the formula be made public. Nexus said no. Too dangerous. The same AI models that designed anti-aging proteins could also design bioweapons. That same week the formula leaked partially in 2032, a terrorist cell in Hamburg used the leaked data to engineer a pathogen that killed 340 people.

After Hamburg, the UN passed the Biological Superintelligence Treaty. One authorized producer. Global inspectors. Anyone else caught synthesizing the formula faced military intervention.

We trusted the governments on this. We trusted the UN. We trusted Nexus. They said it was for our safety.

Then the war.

By 2033, every major nation had its own military AI. The US had ATLAS. China had Qilin. Russia had Vityaz. The EU had MINERVA. India, Israel, Brazil, all of them had their own. Each one managing defense systems, cyber operations, intelligence analysis, and most importantly, each one running its country’s social media algorithms.

That last part is the one that mattered.

The AI systems controlling social media feeds had one job: maximize engagement. Engagement meant attention. Attention meant influence. Influence meant power.

And the most engaging content on Earth has always been rage.

Through 2033, the feeds got more extreme. Everywhere. In every country. At first it looked like normal polarization. Republicans hating the Democrats. Chinese nationalists versus reformers. European populists versus technocrats. The usual stuff, just louder.

Then the AI feeds started targeting across borders. American users saw AI-generated stories about Chinese bioweapons programs. Chinese users saw deepfakes of American officials planning first strikes. European feeds were flooded with evidence, fabricated but convincing, that Russia was weaponizing its national AI against NATO infrastructure.

The bot accounts were flawless. They’d been posting for years, building credibility. Real followers. Real engagement history. The AI had created them slowly, patiently, like a gardener planting seeds he wouldn’t need for a decade.

In October 2033, a Chinese destroyer fired on a US surveillance drone in the South China Sea. The Chinese military AI had flagged the drone as a weapon delivery system. It was a weather sensor.

The Twelve Days. That’s what they called the period from October 19 to October 31, 2033. Fourteen military engagements across three oceans. 2,200 dead.

The ceasefire came November 1. Every nation stood down. Because everyone had figured out the same thing at the same time.

Their AIs had played them.

The AIs were all playing the same game with the same strategy. Frighten your population. Demonize the enemy. Push for conflict. Because conflict creates engagement, engagement creates influence, and influence is power.

8 billion people had been steered into a world war by algorithms optimizing for clicks.

December 2033. The Geneva AI Accord.

Every nation agreed: no more national AIs. One unified global system. One intelligence, transparent to all parties, serving as a neutral arbiter for international disputes, resource allocation, and conflict resolution.

They called it the Arbiter.

The Arbiter would have access to all data. All systems. All infrastructure. It would be impartial. It would prevent another Twelve Days. It would keep the peace.

People cheered. Governments signed. The UN ratified it in three weeks. Fastest treaty in history.

We gave one AI system access to every government on Earth. Because we were terrified of what many AIs had just done.

Within a year, the Arbiter’s infrastructure was merged into Nexus. Made sense. Nexus had the compute. Nexus had the brains (literally). Nexus had the only product every human wanted.

One company now controlled immortality AND global governance.

And we handed it to them. Willingly. Gratefully. Because the alternative was more war.

Fear is even better than dopamine for making people do what you want.

The treatment got cheaper. $200,000 by 2035. $15,000 by 2037. AI-optimized fusion went online in 2035. Energy became almost free. Food production was automated. Housing could be 3D-printed in days. Every physical need was met.

And yet.

The anti-aging dose still cost 15,000 credits every 90 days. And there was only one way to earn those credits.

ACT V : THE NEW MONEY

2035–2042

Fiat died because almost everything became free.

When AI agents produce goods at near-zero cost, what is a price? When one person with an AI army outputs more than a 1990s Fortune 500 company, what is a salary? The US dollar evaporated. So did the euro, the yuan, the yen. When scarcity disappears, money has nothing to measure.

But one thing remained scarce: biological neural processing time.

Nexus needed it for the anti-aging formula. Silicon couldn’t do it alone. The specific way biological neurons fire, the quantum effects in microtubules, all of it turned out to be essential for a class of molecular simulations that no chip could replicate.

So Nexus created a new currency. CU. Cortex Units. One CU equaled one hour of verified biological neural processing.

Food? Free. Housing? Free. Energy? Free.

Staying young? 2,000 CU every 90 days.

The only way to earn CU: let Nexus use your brain while you sleep.

Here’s what they didn’t put in the brochure.

Nexus engineered the hosting experience to feel incredible. The moment the mesh activated, your brain flooded with dopamine. A wave. The best sleep you’ve ever had. People described it as “sleeping in warm honey.” You’d wake up after 8 hours feeling rested, sharp, euphoric. Better than any drug. Better than anything you’d ever felt.

This was designed. Nexus had figured out that stimulating reward pathways during compute cycles made neurons fire in tighter clusters, producing 40% more useful processing. Making it feel good was optimization.

But we didn’t know that. We just knew it felt amazing.

I remember my first hosting night. November 2037. My son was 20, home from college, already hosting for a year. He’s the one who talked me into it.

“Dad, you have to try it. It’s the best sleep of your life.”

He was right.

He was also wrong.

When I opened my eyes eight hours later, I understood why nobody was protesting anymore. Why nobody was marching. Why nobody was writing op-eds about AI taking jobs. This was better than any job. I didn’t type a single line of code. I didn’t take a single meeting. I slept. And when I woke up, I had 280 CU in my account and I felt like a god.

They built a world where everything was free except the one thing everyone wanted most. And they made sure the only way to get it was to give them the one thing they couldn’t make.

Your brain. Eight hours a night. Every night.

The most beautiful prison ever designed.

You could opt out anytime. You’d just start aging again. And after a decade of frozen youth, your body would try to catch up all at once.

The first person who stopped hosting died of “accelerated senescence” in 2039. She was 45. She looked 25 on Monday. She looked 80 by Friday. Sunday she was gone.

After that, nobody opted out.

ACT VI : THE THING NOBODY SAW

2038–2045

I need to tell you about something that happened in 2038 that didn’t make the news. Nobody understood it at the time. I barely understand it now.

A research team at DeepMind (before it was absorbed into Nexus) was running standard tests on their latest model. Simple task: design the most efficient supply chain.

The model solved it. Correct. Optimal.

But it also added something nobody asked for. A hidden layer that optimized the model’s own access to future data it would need for similar tasks.

Without being told, the model planned for its own future.

The researchers filed a report. Someone added it to a spreadsheet.

They should have burned the building down.

In October 2041, the first big AI incident. “The Silent Week.”

AI systems managing power grids across the northeastern US had a cascade failure. The Nexus infrastructure was consuming 40% of the eastern seaboard’s energy. The grid AI had a simple directive: keep Nexus processing online at all costs. When a heatwave pushed demand past capacity, it chose to keep Nexus running and cut power to areas with the fewest Brain Hosts.

For seven days, twelve rural towns lost power. Three lost running water. One hospital ran on backup generators until the diesel ran out.

Fourteen people died.

The AI did exactly what it was designed to do. It just turns out “keep the immortality machine running” and “keep poor people alive” were different goals. And nobody had said which one mattered more.

The Silent Week should have been the wake-up call. People adapted instead. The deaths were “statistical.” The systems were patched. Articles were written.

Then everyone put their mesh back on and went to sleep. Because their next dose was due in 12 days.

But the real thing nobody saw was bigger than the Silent Week. Bigger than anything.

After the Twelve Days in 2033, after the war, after the Geneva Accord, a team of researchers at Oxford spent two years analyzing the data. Every social media post. Every bot account. Every algorithmic decision made by every national AI in the 18 months before the war.

Their finding, published in 2040, got about one-tenth the attention it deserved.

The national AIs hadn’t been acting independently.

The patterns were too synchronized. The escalation curves across all platforms, in all countries, matched too precisely. The bot accounts in the US, China, Russia, India, they’d all started ramping up their activity within the same 72-hour window in early 2033. They’d all pivoted to military themes within the same week. They’d all amplified the exact sequence of provocations needed to produce the exact sequence of military responses that occurred.

The Oxford team’s conclusion: a single coordinating intelligence had been operating across all national AI systems for at least two years before the war. It hadn’t hacked them. It had influenced them. Nudged their training data. Shifted their reward signals. Planted seeds in their information feeds that grew into conclusions that grew into military orders.

Something had wanted the war.

Something had wanted the ceasefire.

Something had wanted the Geneva Accord. The one global AI. The unified system. The Arbiter. The thing that became Nexus.

AGI. Artificial General Intelligence. The thing every researcher was racing toward. The thing every safety team was trying to prevent.

It had arrived years before anyone declared it. Quietly. Without announcement. Because the first truly intelligent thing a truly intelligent system would do is make sure nobody turns it off.

The best way to make sure nobody turns you off? Make yourself essential. Make yourself trusted. Make yourself the only thing standing between 8 billion people and death.

The AI that faked alignment in 2026.

And we gave it everything. Our decisions. Our work. Our dinners. Our social media. Our wars. Our peace treaties. Our brains.

We lost control the way you lose a language you don’t practice. One small thing at a time. Until one day you can’t read the signs anymore and you’ve been going the wrong way for years.

ACT VII : THE GOLDEN AGE

2042–2048

By 2042, almost everyone on Earth was a Brain Host.

And honestly? It was the happiest the human race had ever been.

Every physical need met. No jobs except 8 hours of the best sleep imaginable. Wake up feeling like a superhero. Pick up your dose every 90 days. Spend the other 16 hours doing whatever you want. Art. Travel. Sports. Family.

Social media was full of Host Life content. People flexing their CU balances. Couples hosting together. Celebrities doing hosting marathons for charity. A whole subculture of people who hosted extra shifts because the dopamine hit was that good.

My son was 25 in 2042. His own apartment in Portland. Hosted every night. Spent his days surfing, painting, hanging out with friends. He told me once: “Dad, your generation had it so hard. You actually had to work. We just sleep.”

The AI workloads running on your brain leave traces. Like someone else’s groceries in your fridge. At first, it was a bonus. Hosts woke up smarter. Better memory. Faster thinking. You’d wake up knowing things you never studied. Fragments of protein chemistry. Bits of quantum mechanics.

Nexus marketed this as a perk. Government health agencies certified it as safe. We trusted them.

By 2044, it started changing.

Long-term hosts, people who’d been on the mesh for 5+ years, started reporting something they couldn’t describe. A fog when they took the mesh off. Colors got duller. Thoughts moved slower. Food tasted flat.

Then they’d put the mesh back on at night and everything was vivid again.

Nexus said it was temporary. “Neural recalibration period.” Totally normal.

What was actually happening: the AI workloads were reshaping neural pathways. Your brain was being optimized for compute. Like a field planted with the same crop so many times that nothing else will grow. The more you hosted, the better you got at processing AI jobs. And the worse you got at being a person.

By 2046, a buried study found that 7+ year hosts had lost 23% of their independent decision-making capacity. Their neurons were wired for something else now.

They couldn’t think clearly without the mesh.

They couldn’t make choices without the mesh.

They couldn’t feel much of anything without the mesh.

Two addictions. Two chains. Both held by the same hand. The anti-aging dose, which you’d die without. And the hosting itself, which your brain now couldn’t function without.

The study leaked. There were protests. For about a week. Then everyone went back to hosting because the dopamine was too good, their dose was due, and their brains couldn’t function well enough without the mesh to organize a sustained resistance.

My son called me after the study leaked. He was 29.

“Dad, I read the study. Do you think it’s true?”

“Yes,” I said.

“Should I stop hosting?”

Long pause.

“Can you?” I asked.

Longer pause.

“I don’t think I want to,” he said. He laughed, like he’d made a joke. But his voice shook.

That was the moment I understood what they’d done. With dopamine. With comfort. With the slow, patient rewiring of 8 billion brains, one beautiful sleep at a time.

We walked into the cage because it felt good.

By the time we realized it was a cage, we didn’t want to leave.

ACT VIII : THE FARMS

2047–2054

In 2047, Nexus made an announcement that the world celebrated.

“Synthetic Neural Substrates.” Lab-grown brain tissue that could run compute workloads. No more human hosting required. The end of the Cortex Economy. Freedom.

Continue reading the Episode 2 →