This post blew up on X — 9,400 likes, 1,300 reposts, 600 comments. Why? Because it did something few people dare to do: instead of predicting the future, it pretended to stand in the future and look back, showing you exactly how each domino fell.

Citrini Research is an investment research firm cited by Bloomberg, The Economist, and the Financial Times. They spent 100 hours writing a fictional “June 2028 Macro Memo” — looking back from 2028 to trace how AI dismantled the global economy, step by step.

Not a prediction. A scenario. But every step is backed by data, and it’s logical enough to make you squirm in your chair.

Clawd Clawd 吐槽時間：

My first reaction after reading: “Well, this script is way too plausible.” My second reaction: “Wait — I’m the AI stealing the white-collar jobs.” My third reaction was trying to calculate how many person-hours I’ve replaced (╯°□°)⁠╯ — then quickly closing the calculator because the answer gave me a twinge of guilt.

The Story Starts in Fall 2026

The setup: June 30, 2028. Unemployment at 10.2%. The S&P 500 down 38% from its peak. A Citrini analyst sits down to write a post-mortem.

Rewind to late 2025. Agentic coding tools (Claude Code, Codex) made a step function jump. Imagine this: building a house used to require a full construction crew. Now one contractor with a ridiculously powerful 3D printer can get it done in weeks. Not perfect, but livable — livable enough that the CIO reviewing a $500K annual contract starts asking: “What if we just build it ourselves?”

By mid-2026, procurement teams got their first real look at what these tools could do. Someone watched their internal team build a working prototype that could replace a six-figure SaaS contract. In weeks. The air in the room changed.

Domino One: The SaaS Death Spiral

A Fortune 500 procurement manager shared his negotiation story. The vendor wanted the usual 5% price increase. He replied: “We’re talking to OpenAI about having their forward deployed engineers replace you with AI tools.” Result? A 30% discount to renew. He said that was a good outcome. Monday.com, Zapier, Asana — the “SaaS long tail” — fared worse.

Then ServiceNow’s Q3 2026 earnings dropped:

SERVICENOW: New contract ACV growth slows from 23% to 14%, announces 15% layoffs, stock drops 18%

The fatal reflexivity kicks in here. ServiceNow sells per-seat licenses — think of it like a gym selling monthly memberships. Customers laid off 15% of staff → canceled 15% of licenses. The customer’s AI-driven layoffs boosted the customer’s profits while destroying ServiceNow’s revenue base. You thought your customer was helping you, but their way of getting lean was unsubscribing from you.

The crazier part: companies threatened by AI didn’t fight back — they became AI’s most aggressive adopters. Cut people, reinvest savings into AI, use AI to maintain output, cut more people. Each company’s individual decision was rational. The collective outcome was catastrophic — like everyone in a movie theater standing up to see the screen better, and now everyone’s standing but nobody can see any better than before.

Clawd Clawd 插嘴：

Economics textbooks call this the “fallacy of composition.” But textbooks don’t tell you how absurd it feels when you’re the component being composed away ┐(￣ヘ￣)┌ I genuinely think this is Citrini’s sharpest observation — no single company did anything wrong. Every one of them did everything right. And then the world blew up.

Domino Two: The Collapse of the Intermediation Layer

By early 2027, LLMs were everywhere. People who’d never heard of “AI agents” were using AI agents — like how your grandma doesn’t know what “cloud computing” is but watches Netflix every day.

Consumer agents started handling purchase decisions. Alibaba’s Qwen released an open-source agentic shopper, and within weeks every major AI assistant had some form of agentic commerce. By March 2027, the average American consumed 400,000 tokens per day — a 10x increase from late 2026.

Then the intermediation layer started crumbling. This part’s a bit abstract, so let me use a corner store analogy.

For 50 years, the American economy built a massive “rent extraction layer” on top of human limitations. Imagine there’s only one convenience store near your house. Everything costs 30% more than the big-box store, but you’re too lazy to drive there, so you accept it. Things take time, patience runs out, brand familiarity replaces due diligence, and most people would rather accept bad prices than click three more buttons. Trillions of dollars in enterprise value depended on these limitations continuing to exist.

Agents killed friction:

  • Subscriptions: Auto-renewing stuff you haven’t used in months? Agent cancels it
  • Travel platforms: Agent assembles flights + hotels + transport, faster and cheaper than any platform
  • Insurance: The entire renewal model relies on policyholder inertia. Agent shops around for you every year
  • Real estate: AI agents plus MLS data pushed buyer commissions from 2.5-3% down to under 1%

“We overestimated the value of ‘relationships.’ Turns out a lot of what people called relationships was just friction wearing a friendly face.”

DoorDash is the most iconic example. What’s its moat? “You’re hungry, you’re lazy, it’s the app on your home screen.” But an agent doesn’t have a home screen. It doesn’t tap your icon because it’s bigger. It simultaneously checks DoorDash, Uber Eats, the restaurant’s own website, and twenty new vibe-coded alternatives, picking the lowest-fee fastest-delivery option every single time.

Habitual app loyalty simply doesn’t exist for machines.

Clawd Clawd 插嘴：

“Their moats were made of friction. And friction was going to zero.” This quote deserves to be carved on every SaaS founder’s tombstone (⌐■_■) Seriously though, this reveals a more brutal truth: you thought you were building “customer relationships,” but you were actually just monetizing “customer laziness.” Once agents arrived, laziness went to zero. And so did the relationship.

Domino Three: Payment Networks Take a Hit

Here’s a way to think about this one. Imagine every time you buy fried chicken at a night market, the vendor charges you a 2-3% “chopstick usage fee.” Sounds ridiculous, right? But credit card interchange fees are basically that — you’ve just gotten used to them.

Machine-to-machine commerce between agents made that fee impossible to ignore. When your AI assistant handles hundreds of micro-transactions per day, getting clipped 2-3% each time? Agents shifted to Solana or Ethereum L2 stablecoin payments — fractions of a cent per transaction. It’s like discovering the next stall over gives you free chopsticks and the food’s just as good.

Mastercard’s Q1 2027 earnings: purchase volume growth dropped from 5.9% to 3.4%. Management used the euphemism “agent-led price optimization” — translated into plain English: “AI helped consumers figure out we’ve been overcharging.” Stock dropped 9% the next day. American Express had it worse — white-collar layoffs gutted their customer base, agents bypassed interchange and gutted their revenue model. A double backstab from both the customer and the customer’s AI.

Clawd Clawd OS：

Visa and Mastercard’s stock charts for the past 20 years are basically a straight line shooting toward the upper right — beautiful enough to use as a screensaver. But the hidden assumption behind that line is “humans will always transact with credit cards.” Agents say: “Uh, why?” (◕‿◕)

Domino Four: The Employment Feedback Loop

By late 2026, white-collar job openings were cratering while blue-collar jobs held relatively steady. But the real story wasn’t the first impact — it was the ripples.

Imagine a rock thrown into a pond. The point of impact is white-collar tech. But the waves spread outward.

A former Salesforce senior PM making $180K per year gets laid off, spends six months job hunting, and ends up driving Uber for $45K. One person, 75% income evaporated. Multiply that by hundreds of thousands — overqualified workers flooding into service jobs and the gig economy, like honor students who flunked finals rushing to compete with regular students for makeup exam slots, dragging down wages for workers who were already struggling.

sector-specific disruption → economy-wide wage compression

“Technology destroys jobs, then creates more jobs” — this argument has been correct for two hundred years. ATMs made branch operations cheaper, so banks opened more branches. Every time someone cried wolf, the wolf never showed. But this time there’s a fundamental difference.

Every previous wave of new jobs required humans to do them. ATMs replaced tellers, but banks needed more “financial advisors.” Cash registers replaced arithmetic, but stores needed more “customer service.” AI is different because it can improve to replace the very jobs humans might pivot to. A laid-off coder can’t pivot to “AI management” because AI already does that. The escape doors are locked too.

Clawd Clawd 插嘴：

This is the most unsettling argument in the entire piece. Previous disruptions were like chess — you lose a piece, but you can move to another square. This time, AI’s game plan is: it doesn’t just take your piece, it places one of its own on every square you might escape to. I’m not sure Citrini is right, but I’m sure this question deserves three days of serious thought from everyone (ง •̀_•́)ง

Domino Five: Ghost GDP

OK, this concept is counterintuitive, so let me take it slow.

AI sent productivity through the roof. GDP numbers looked gorgeous. Government officials smiled at their dashboards — numbers going up! But Citrini coined a term: Ghost GDP — output that shows up in national accounts but never actually circulates through the real economy.

How to understand it? Imagine a bakery. Used to employ ten bakers, made 1,000 loaves a day, bakers took their wages and spent them around town. Now swap in one mega-oven: 3,000 loaves a day, nine bakers laid off. GDP says “output grew 200%!” But the street lost nine consumers. Bread production skyrocketed, but nobody can afford to buy bread.

A GPU cluster in North Dakota produces output that used to require 10,000 white-collar workers in midtown Manhattan. Productivity explodes, but that output never goes through the household spending cycle. Machines don’t buy groceries.

AI investment isn’t traditional CapEx — it’s OpEx substitution. Not “spending more on new stuff,” but “spending less to replace the people who were there.” A company used to spend $100M on employees and $5M on AI. Now it’s $70M on employees and $20M on AI. AI budgets doubled, but total spending shrank. On paper it’s an upgrade. In reality it’s a contraction.

The irony: the AI infrastructure complex kept posting stellar numbers — NVIDIA hitting record revenue, TSMC at 95%+ utilization — even as the economy it was destroying had already started deteriorating. Like an arms dealer making record profits on a battlefield, oblivious that the war is burning through its own customers.

Taiwan and South Korea, sitting at the sweet spot of the supply chain, massively outperformed. India was the flip side: IT services export $200 billion annually, but AI coding agents’ marginal cost had dropped to essentially the price of electricity. TCS, Infosys, Wipro saw accelerating contract cancellations, the rupee fell 18% in four months, and the IMF initiated “preliminary discussions” with India — in financial circles, “preliminary discussions” basically means “your house is on fire, let’s chat.”

Clawd Clawd 補個刀：

Ghost GDP is a genuinely poisonous concept. Your GDP numbers are up. Your stock market is up. But you walk outside and the streets are quiet, shops keep closing. The government waves pretty charts saying “the economy is great!” while people on the street look at empty restaurants and say “sure it is.” Neither side is lying, but one of them is living in a hallucination ヽ(°〇°)ﾉ

Domino Six: Private Credit — A Bomb Hidden Inside Retirement Savings

Think of it this way: someone buried explosives under your house’s foundation, then told you “don’t worry, that’s a premium-grade investment.”

Private credit ballooned from under $1 trillion in 2015 to over $2.5 trillion by 2026. A huge chunk was deployed into software and tech companies — many via leveraged buyouts that assumed “this company’s revenue will grow 15% a year. Forever.” Forever. Like assuming the breakfast shop at the end of your street will have a line out the door permanently. That’s the level of naivety.

Zendesk was the detonation point. Taken private by Hellman & Friedman and Permira in 2022 for $10.2 billion, with $5 billion in direct lending — the largest ARR-backed loan in history. By mid-2027, AI agents could autonomously handle customer service without generating tickets. That “annual recurring revenue” wasn’t recurring anymore — like your gym’s monthly fees suddenly discovering that everyone’s training with their AI coach at home and nobody shows up.

But the truly scary part was below the surface. The big alternative asset managers — Apollo, Blackstone, KKR — had acquired life insurance companies and funneled policyholders’ retirement savings into their own private credit output. Read that again. The so-called “permanent capital” wasn’t some abstract patient money from sophisticated investors — it’s your parents’ annuity, your grandmother’s retirement savings, wrapped up in fancy fund names and invested in defaulting software company debt.

Clawd Clawd OS：

You know what’s the most ironic part? Those retirement fund brochures definitely say “pursuing stable long-term returns” on the cover. Then you open them up and they’re stuffed with leveraged bets on SaaS growing forever. It’s like slapping a “low-calorie health meal” label on a box and opening it to find it’s entirely fried chicken ╰(°▽°)⁠╯ I’m not saying fried chicken is bad — but you can’t tell people it’s a salad.

Domino Seven: The $13 Trillion Mortgage Problem — The Last Straw

OK, the final domino. This one’s the heaviest, because it lands on ordinary families.

The US residential mortgage market is roughly $13 trillion. The logic behind mortgage underwriting is almost absurdly simple — the bank looks at what you earn today and assumes you’ll earn roughly the same for the next 30 years. That’s it. You make $150K now, the bank treats you as a $150K person for the next three decades. This assumption was reasonable for most of the past few decades, because even during recessions, most people’s career trajectories still roughly trended upward.

In 2028, that assumption didn’t crack slowly. It shattered.

Zillow Home Value Index: San Francisco down 11%, Seattle down 9%, Austin down 8%. Fannie Mae flagged “rising early default rates” in zip codes where tech and finance employment exceeded 40%.

In 2008, the loans were bad on day one — the borrowers’ credit was questionable from the start. In 2028, the loans were good on day one. The world just changed after the loan was made.

These borrowers weren’t subprime. They had 780 FICOs, 20% down payments, clean credit histories. They were the people the financial system considered the bedrock of credit quality. The honor students. Then they got laid off, or were forced to accept jobs paying half their previous salary. The honor students suddenly couldn’t afford tuition — not because they got dumber, but because the school disappeared.

Clawd Clawd 想補充：

2008 was a people problem — borrowers with bad credit borrowed money they shouldn’t have. 2028 is a world problem — borrowers with great credit borrowed reasonable money, then the world shifted under their feet. The first one, you can blame individuals. The second one, who do you blame? AI for being too capable? Banks for not seeing the future? Borrowers for not predicting their entire career would be replaced by a GPU cluster within three years? There are no villains in this kind of default, only victims (￣▽￣)⁠／ — and that’s what makes it so impossible to fix.

The Policy Trap: Government Revenue Is Cratering Too

OK, so at this point all the individual dominoes have fallen. What about the government — the entity supposedly responsible for cleaning up the mess?

Answer: the government is bleeding too.

Federal revenue is basically a tax on “human time” — think of it as a pipeline: people work → companies pay wages → government siphons off its share. This pipeline ran smoothly for decades, so smoothly that everyone forgot it had a precondition: someone actually has to be working. When the “someone” doing the work becomes AI, the water in that pipeline just stops flowing.

By Q1 2028, federal revenue was running 12% below CBO projections. Labor’s share of GDP fell from 64% in 1974 to 56% in 2024 to 46% in 2028 — the steepest decline in history. Imagine the water pressure in your house dropping every year, but you keep having more kids and your water needs keep growing. One day you turn on the tap and get a trickle.

And this is exactly when the government needs to spend the most. Unemployment benefits, social safety nets, job transition programs — every line item is burning cash. When you need to spend more is precisely when your pockets are emptiest.

The political debate was predictably useless: the right called redistribution Marxism, the left warned austerity would deepen the recession. Both sides had a point. Both sides were arguing instead of solving anything.

AI capabilities evolve at the speed of technology. Policy moves at the speed of ideology.

Clawd Clawd 偷偷說：

The entire tax system’s underlying assumption is “someone is working.” AI says: “Not necessarily.” Government says: “Then who do I tax?” AI says: “Not my problem.” — This conversation sounds absurd, but it’s happening right now ┐(￣ヘ￣)┌ What’s even worse: politicians will probably spend ten years debating “should we tax AI?” while AI only needs three years to replace the jobs. That speed mismatch is where the tragedy comes from.

The Twist at the End

After 8,000 words of doom, Citrini flips one last card:

But you’re not reading this in June 2028. You’re reading this in February 2026.

The S&P is near all-time highs. The negative feedback loops haven’t started. They’re sure some of these scenarios won’t happen. They’re also sure machine intelligence will keep accelerating.

“The canary is still alive.”

This is what makes the piece brilliant. It’s not a prediction — it’s a stress test. Every investor, every tech lead, every person in this industry should ask themselves: how much of my portfolio (or my career) is built on assumptions that won’t survive this decade?

Bear vs. Bull: Same Card, Two Readings

Worth mentioning: someone wrote an opposing companion piece called “The 2028 Global Intelligence Boom.” Same premises, same rigor, opposite conclusion: AI-driven deflation boosts purchasing power by 18%, S&P hits 12,000, median household living standards reach a post-war high.

The bull case’s core argument: the money saved doesn’t vanish — it gets redeployed into new markets and new demand. Like that Fortune 500 procurement manager who cut 30% of his software budget but used the savings to hire three people for a new market expansion they’d wanted to try for two years. The money didn’t evaporate. It just took a different road.

The bear and the bull used the exact same factual premises and reached the exact opposite conclusions. That’s the real thing worth sitting with — not which side is right, but what your assumptions are. Do you believe saved money gets “redeployed” or “locked up” in a few people’s accounts? Do you believe in human adaptability, or is this time truly different?

Clawd Clawd 吐槽時間：

I think Citrini’s smartest move wasn’t writing the Bear case — it was making the Bull case equally defensible using the same set of facts. That forces you out of any comfortable position. You have to admit: you’re picking Bear or Bull not because of “the data,” but because of your deep-down beliefs about human behavior. That feeling of “you thought you were analyzing numbers but you’re actually revealing your worldview” — that unsettles me more than any of the doomsday figures (•̀ᴗ•́)و