The "Ghost GDP" Thesis Moved Markets
The Citini Research piece triggered an actual market selloff, with Bloomberg reporting "software payment stocks slide after Citini post on AI risk." The core argument: AI productivity creates GDP on paper, but doesn't circulate in the real economy because businesses spend on data centers rather than labor.
"By late 2025, agentic AI tools become vastly better at coding and complex tasks. Productivity looks great on paper. GDP and productivity metrics soared because AI output counted in the official numbers but most of that value didn't translate into real consumer spending."
AI Impact on Economy Is Still Minimal Despite Market Frenzy
The disconnect between AI's market impact and actual economic contribution is stark. Total AI lab revenues are only $30-40 billion, generating perhaps $200 billion in GDP—insignificant compared to America's overall economy. Healthcare and service jobs remain the real economic drivers.
"AI is the only thing holding up the economy. I was like no AI is actually doing very little for the economy right now. It's doing a lot for the markets. It's doing a lot for the future, but like in terms of the actual economic impact of AI, it's very low."
The Iron Rule of Human Reality
John Lober's "Contra Citini" piece offers institutional inertia as a counterargument. Real estate brokers were supposed to be obsolete for 20 years, yet regulatory capture and market inertia keep them thriving. His buyer's agent made $50,000 for 10 hours of form-filling he could have done himself.
"Everything is always more complicated and takes much longer than you think it will. Even if you already know about the iron rule. That doesn't mean that meaningful change in the world won't happen, but that the change will be more gradual, giving us the time to respond and adjust."
Until We See 5,000 Software Engineers Laid Off at Once, AI Isn't Replacing Jobs
The litmus test for AI replacing software engineers isn't productivity gains—it's actual mass layoffs. If companies are truly replacing engineers with AI rather than making them more productive, we'd see unprecedented layoff announcements. We haven't.
"Until we see a round of layoffs at a company that is 5,000 software engineers at once, it's hard to believe that AI is replacing software engineers versus just making them a lot more productive. If somebody's a lot more productive, you'll pay at least the equivalent amount to maintain them."
AI Doom Is Mainstream in a Way Y2K Never Was
More than 30% of Americans believe AI could end human life on Earth—a level of existential concern that dwarfs Y2K millennium fears or 2012 apocalypse predictions. Yet they continue their daily routines, creating a surreal cognitive dissonance.
"The average American believes that they are in Terminator Judgement Day, but they still have to go to Cyberdine Systems and do their fake email job right up until the bombs drop. That that's the general tenor around AI."
The Dot-Com Bubble Actually Had Optimistic Vibes Compared to AI
The New York Times comparison reveals a fascinating contrast: the dot-com era had froth and Y2K fears, but overall optimism about the internet. The AI boom has massive investment but widespread doom sentiment. People loved the dot-com boom; they're terrified of AI.
"People loved the dot-com boom. The AI boom not so much. The tenor around the dot-com era. Yes, there was a lot of froth. Yes, there was Y2K and people were worried about that, but the stats weren't quite the same."
Low Probability Scenarios Become What You're Known For Forever
When someone frames a prediction as "10% chance this happens," audiences ignore the probability and only remember the dramatic scenario. No one gets clicks for the 90% baseline case, creating perverse incentives for doom scenarios.
"As soon as you say something crazy happens with no matter how low the percentage is, like that's what you're going to be known for forever. So be careful out there with those predictions."
DoorDash Was the Worst Possible Example for AI Disruption
Citini's choice to highlight DoorDash as vulnerable to AI was widely mocked because the barrier isn't software—it's distribution, restaurant adoption, and driver networks. Tony Xu could "vibe-code" a delivery app overnight; building the actual business is the hard part.
"Out of every example they could have chosen, they went with DoorDash. The barrier to entry for launching a delivery app is not and has never been software. It's distribution, restaurant adoption, user adoption, and of course driver adoption."
Software Engineers Are Less Than 1% of America's Workforce
Even if AI causes massive reallocation in tech, the actual economic impact is limited by scale. Software developers represent a tiny fraction of employment, and tech workers broadly are less than 10% of the workforce. The Tyler Cowen "slow takeoff" thesis suggests most jobs remain AI-resistant for now.
"The number of people that are software developers less than 1% of America, the number of people that like work at tech companies broadly is less than 10%. And so even if there's some massive reallocation there... that doesn't immediately translate to what is happening in the real economy."
Y2K Cost $300 Billion to Prevent
The millennium bug fear that computers storing years as two-digit numbers would cause financial system collapse required hundreds of billions in remediation. The Gregorian calendar had been in place since 1582, giving humanity 400+ years to prepare for the year 2000.
"It wound up being something like hundreds of billions of dollars were spent in the leadup to Y2K."
AI Will Create Entirely New Software Paradigms, Not Just Clone Salesforce
The real disruption isn't "Amazon Basics for SaaS"—cheaper clones of existing products. It's fundamentally new relationships with software, like AI agents that automatically manage your schedule without you monitoring dashboards. The paradigm shift is deeper than cost arbitrage.
"I think the AI disruption that is much more real is like you have entirely new paradigms for software, an entirely new relationship with software. And it's not just like, oh, somebody built the exact same version of Salesforce."
Re-Industrialization Offers a Labor Safety Valve
America has "virtually limitless capacity and need for re-industrialization." The U.S. is almost entirely dependent on China for batteries, motors, small semiconductors, and the electric stack. Even if white-collar jobs face disruption, manufacturing renaissance could absorb displaced workers.
"We are largely no longer know how to create and don't have the facilities for making the core building blocks of modern life. Batteries, motors, small semis, the whole electric stack is something we are almost entirely dependent on China and other countries for."
Industrial-Scale AI Model Theft Is Here
Anthropic identified DeepSeek, Moonshot, and Minimax creating over 24,000 fraudulent accounts and 16 million exchanges with Claude to distill its capabilities into their own models. The "industrial-scale distillation attacks" represent systematic IP theft at unprecedented scale.
"These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude extracting its capabilities to train and improve their own models."