Scott you’re a genius and my great admiration is primarily your seeing reality behind the BS of wealth inequality which is a moral thing as much as a comic thing…a societal decision made by individuals. Thanks for this post
Former Microsoft Sr. Director here: half of Google and Amazon's most recent profits on their SEC filing came from real cash flow from their core businesses. THE OTHER HALF CAME FROM NON-CASH, THEORETICAL GAINS IN THEIR STAKE IN ANTHROPIC.
Imagine you invest in a friend’s lemonade stand. Later, you give him even more money, and you both agree the stand is now worth a billion dollars. Thanks to a quirky accounting rule, you can claim your share of that billion dollars as your own personal "profit" for the year.
People who see blowout AI profits from Big Tech aren't looking carefully into those profits.
Yeah, operating profit is just the first half. When the news says "Amazon's profit surged!!!," and then mom and pop investor goes and buys Amazon stock, they are talking about net income, not operating profit.
Scott, your argument seems to underweight the trajectory risk. It reads as if today’s AI systems, with their current jaggedness and failure modes, are roughly the systems we’ll be dealing with going forward. But people have been saying LLMs are about to hit a wall since the GPT-3 era, and the wall keeps moving.
Capabilities are not just improving, several measures suggest the pace of improvement has accelerated, especially with reasoning models. A year ago, many of my friends in tech thought AI was mostly hype. Then they were forced to use it in real workflows, and now many of them are openly worried about getting replaced in the next year or two.
The latest frontier models are also getting powerful enough that even the current administration, despite its deregulatory instincts, is moving toward pre-deployment government evaluations and reportedly considering formal review of new models before release. The U.S. and China are also reportedly exploring AI guardrail talks to prevent the rivalry from spiraling into crisis.
So I agree that “AI apocalypse” can be used as marketing. But dismissing the risk as mostly narrative seems like wishful thinking, and like a refusal to come to terms with the reality of the situation. It seems unwise to blindly assume today’s gaps are durable. No one has a crystal ball, but it is far from guaranteed that AI systems will remain this jagged while capability, reliability, scaffolding, and adoption keep advancing this quickly.
Deployment of new technologies generally lags their capability development curve. Models might be improving quickly but significant organizational, cultural, financial and regulatory hurdles remain to comprehensive, rapid adoption of Gen AI. Most firms are dabbling in it or deploying it within a couple functions. Even among firms rapidly adopting AI in software development or similar roles, the data are mixed: some recent research shows that AI-adopting firms experience faster headcount growth, other studies find the opposite; the point is, the evidence is far from conclusive that AI advancement and deployment erase jobs.
Scott, with respect, I think the historical analogy is misfiring. The issue isn't AI replacing human labor the way prior technology waves did. It's that AI is replacing human cognition — the most valuable of all human activities, and the financial and psychological foundation of the entire professional class. Every prior wave substituted for muscle, memory, or calculation while leaving judgment and interpretation intact. This one is being explicitly built to perform judgment and interpretation. I hope you're right and I'm wrong. But I doubt it.
Cognition refers to the mental processes involved in acquiring knowledge and understanding, encompassing thinking, knowing, remembering, judging, and problem-solving
"Fear is the product. Capital is the outcome." thats the single best sentence written about AI economics this year and it deserves the extension you didnt quite give it.
The apocalypse narrative isnt just a capital attraction strategy. its a talent attraction strategy. if every software engineer believes AI will replace all jobs within five years, the rational career move is to join the company building the AI rather than the company being disrupted by it. Amodei, Altman, and Musk arent just scaring investors into funding them. theyre scaring the talent pool into working for them. the prediction doubles as a recruitment pitch because the implied message is "the safest place to be when the flood comes is on the ark, and we're building the ark."
Your employment data is the strongest section of this piece and it needs to be cited more widely because it demolishes the narrative with actual numbers. Meta returning to 2021 headcount. Microsoft at 47% above pre-pandemic levels even after the layoff announcement. net tech employment flat at 9.6 million rather than collapsing. the gap between what the CEOs say in press interviews and what their own HR departments are actually doing is the most revealing data set in the entire debate.
The Oracle example is particularly telling. 18% workforce reduction with negative cash flow projected through 2030. thats not AI efficiency. thats a company struggling to fund a capital expenditure cycle that may not generate returns for years. framing cost-cutting as AI transformation is the corporate equivalent of putting a GPU sticker on a budget memo.
The Schiller narrative feedback loop observation is the part of this piece that should worry policymakers more than it currently does. if the apocalypse narrative causes companies to pre-emptively freeze hiring because they believe AI will replace the roles anyway, and workers pre-emptively exit sectors they believe are doomed, you get a labour market contraction that has nothing to do with AIs actual capabilities and everything to do with the story being told about them. the recession that follows gets attributed to AI when it was actually caused by the narrative about AI. thats the self-fulfilling prophecy in its most precise form.
Your three scenarios are cleanly structured. Id put rough probabilities on them: bubble bursts (20%), slower-than-expected timeline (55%), disruption faster than adaptation (25%). the middle scenario is the one that gets the least attention because its boring and hard to monetise as content. but its where most of the evidence currently points, and its the scenario where thoughtful policy on transition infrastructure could make the biggest difference.
Seems to me that call center personnel will be the among the first to lose their jobs. There are about 3 million of these persons working in the USA. It is pollyanna to claim that they will find good jobs in the AI economy. They will be retrained to do what?
This is going to age about as well as a Dianna Russini denial.
I’ve enjoyed your “outspoken contrarianism” for years (I mean, how else does a middle-aged bald man get likes on the internets), but this one just seems, well, lazy.
And I typically wouldn’t care to reply, but it’s these sorts of pieces that our lazy congresspeople will glom onto to avoid or delay much-needed policy making.
1) This time IS different for at least the reason that — as you point out — AI’s attack surface is so expansive that displaced workers will find few places of refuge. What’s also different this time, is that the folks building the disruptive tech aren’t gaslighting us like the titans of automation, enterprise SW, and offshoring did before (and still do) with the, “we enable workers to do more high-value work.” Bullshit.
2) Any commentary on this subject that uses unemployment as a (“the”) KPI should be dismissed out of turn. What a useless metric and anyone serious knows this. (“I’d like to solve the puzzle, Pat: Mass Underemployment”)
3) it’s cute to think the companies that will reap great profits from all this new found labor efficiency will re-invest it in fun, new jobs producing projects. Hope springs eternal I suppose. The cynic, however, sees more stock buybacks.
I’ve been using both ChatGpt and Claude along with some others for 2 years. It does not replace a human. It replaces the gap in our ancient brains capabilities today and the advanced technology we keep creating. It doesn’t have the human emotion and empathy that makes us superior. I have thought of things my AI miss. It compliments me. Maybe it learns like I did to always challenge its responses or not. I was at the beginning of computing in the workplace in 1982, internet in 1992 and next is AI.
"Fear is the product" is the truest line here, and I'd like to back it up from the inside. I spent three decades building engagement products, and the open secret of that work is that fear and outrage simply perform better than calm. They're stickier, they travel further, they keep you scrolling. So a fear narrative doesn't just spread because it's useful to the AI incumbents. It spreads because the entire information system is tuned to reward it.
One gentle caveat: I'd be careful telling a frightened 24-year-old that the panic is "engineered," even when it is. The narrative may be manufactured, but the anxiety in their body is real, and being told it's marketing rarely makes anyone feel better. The fix isn't debunking the fear — it's giving people something solid to stand on.
Years ago, my brother, a computer scientist and engineer, shared an observation: programmers are really blue-collar workers. What they manufacture is software, but their highly skilled jobs aren't that different from the manufacturing jobs of the past. Their workstations are seats on a 21st-century assembly line.
Now, knowing that programmers created AI in their own image, we can assume that it is fairly adept at doing their job. But my guess is their belief that AI will soon be doing all our jobs is the result of an epidemic of Dunning-Kruger infecting the tech world. Someone needs to tell the technocrats that just because they understand how AI will do their job doesn't mean they understand how it will do mine.
But if for nothing else, maybe this debate can help explain why we find ourselves on a cultural precipice. For some reason, we thought we could walk away from the way things have always worked. In the past, engineers and artists had to do a dance, together building what was not only possible, but what was inspiring and beautiful. Only recently have we let the nerd run free, and it seems they’ve optimized us into a cultural cul-de-sac where feeding on one's own tales is celebrated with wide-eyed wonder. Much of AI is propped up by a plagiarism economy where the Cliff Notes version of centuries of accomplishments has been rolled up and resold with little to no thought given to origin or ownership. Worse, they’ve enticed us to strengthen their hold by collectively optimizing their tools. Let’s be clear, when folks talk about AI and say, “not yet” or “it’s just getting better,” its improvement is based on our action. It is odd; somehow that got us to pay to work for them, training our replacements. Still, I'm not quite ready to believe that we're the beta testers for societal collapse. A bit of a tech collapse, sure, the bureaucratic bloat coupled with inflated compensation in a sector that has been passing off incremental improvement as innovation for decades, heck, a financial reckoning is probably long overdue.
So yes, I can find solace in the belief that this AI charade won’t last forever; the bubble will burst. We can only hope that in the end, tech's inflated influence on culture will be the most notable casualty of the AI apocalypse. Wouldn’t that be ironic? If the ones hit hardest by the AI reality check are its creators, while the rest of us are simply a little better-off for a thinning of the algorithmic veil. Hey, if someone has the time, you want to teach your favorite LLM to master irony, it would be nice if it could contextualize its own demise.
You should listen to your own family of podcasts more sir. Alice Han on China Decode stated last week that the unemployment rate in China among 16-24 year-olds is 16.9%. She also spoke with cohost James Kynge about robots able to make dim sum (can make 3000 dumplings/hour). China has much more robot utilization than in the US. AI/robots are coming for way more jobs than you keep saying on your other podcasts
Prof G, you the man but we still have to start a policy conversation now to accommodate the downside risk of AI replacing jobs at scale (whether self fulfilling or economically inevitable). I've written about some ideas to start the conversation now so that government policy can have a chance (probably small but worth a shot) to enact it in the next 1-3 years. Now is the time of policy moonshots to reinvent the welfare state and divide accountability for an economy that is betting the house on this:
My wife was a freelance academic copy editor. Was. But then everybody discovered that you don’t need one, ChatGPT does the work just as well. This week at my small business I got an SEO tech fix list from Claude that I would have paid a consultant $1,000 to provide me. 10 minutes, and it’s exhaustive. Tomorrow I’ll generate a new offering contract from Claude that I would have paid a lawyer for - probably 3 or 4 hours at $300/hr. When I hear people confidently predict that the destruction in creative destruction won’t outstrip the creation I have to wonder whether they actually are integrating these tools into their workflow. It’s not like previous tech innovations.
Scott you’re a genius and my great admiration is primarily your seeing reality behind the BS of wealth inequality which is a moral thing as much as a comic thing…a societal decision made by individuals. Thanks for this post
Former Microsoft Sr. Director here: half of Google and Amazon's most recent profits on their SEC filing came from real cash flow from their core businesses. THE OTHER HALF CAME FROM NON-CASH, THEORETICAL GAINS IN THEIR STAKE IN ANTHROPIC.
Imagine you invest in a friend’s lemonade stand. Later, you give him even more money, and you both agree the stand is now worth a billion dollars. Thanks to a quirky accounting rule, you can claim your share of that billion dollars as your own personal "profit" for the year.
People who see blowout AI profits from Big Tech aren't looking carefully into those profits.
There’s something called operating profit. It excludes one time gains or losses. Right?
Yeah, operating profit is just the first half. When the news says "Amazon's profit surged!!!," and then mom and pop investor goes and buys Amazon stock, they are talking about net income, not operating profit.
Scott, your argument seems to underweight the trajectory risk. It reads as if today’s AI systems, with their current jaggedness and failure modes, are roughly the systems we’ll be dealing with going forward. But people have been saying LLMs are about to hit a wall since the GPT-3 era, and the wall keeps moving.
Capabilities are not just improving, several measures suggest the pace of improvement has accelerated, especially with reasoning models. A year ago, many of my friends in tech thought AI was mostly hype. Then they were forced to use it in real workflows, and now many of them are openly worried about getting replaced in the next year or two.
The latest frontier models are also getting powerful enough that even the current administration, despite its deregulatory instincts, is moving toward pre-deployment government evaluations and reportedly considering formal review of new models before release. The U.S. and China are also reportedly exploring AI guardrail talks to prevent the rivalry from spiraling into crisis.
So I agree that “AI apocalypse” can be used as marketing. But dismissing the risk as mostly narrative seems like wishful thinking, and like a refusal to come to terms with the reality of the situation. It seems unwise to blindly assume today’s gaps are durable. No one has a crystal ball, but it is far from guaranteed that AI systems will remain this jagged while capability, reliability, scaffolding, and adoption keep advancing this quickly.
I wrote more about why I think the usual “technology always creates new jobs” argument breaks down here: https://alont.substack.com/p/what-happens-when-we-automate-our
Wow. Your view is powerful and supersedes Mr Galloways. Kudos!!!
Deployment of new technologies generally lags their capability development curve. Models might be improving quickly but significant organizational, cultural, financial and regulatory hurdles remain to comprehensive, rapid adoption of Gen AI. Most firms are dabbling in it or deploying it within a couple functions. Even among firms rapidly adopting AI in software development or similar roles, the data are mixed: some recent research shows that AI-adopting firms experience faster headcount growth, other studies find the opposite; the point is, the evidence is far from conclusive that AI advancement and deployment erase jobs.
Scott, with respect, I think the historical analogy is misfiring. The issue isn't AI replacing human labor the way prior technology waves did. It's that AI is replacing human cognition — the most valuable of all human activities, and the financial and psychological foundation of the entire professional class. Every prior wave substituted for muscle, memory, or calculation while leaving judgment and interpretation intact. This one is being explicitly built to perform judgment and interpretation. I hope you're right and I'm wrong. But I doubt it.
Cognition refers to the mental processes involved in acquiring knowledge and understanding, encompassing thinking, knowing, remembering, judging, and problem-solving
AI doesn't do that.
"Fear is the product. Capital is the outcome." thats the single best sentence written about AI economics this year and it deserves the extension you didnt quite give it.
The apocalypse narrative isnt just a capital attraction strategy. its a talent attraction strategy. if every software engineer believes AI will replace all jobs within five years, the rational career move is to join the company building the AI rather than the company being disrupted by it. Amodei, Altman, and Musk arent just scaring investors into funding them. theyre scaring the talent pool into working for them. the prediction doubles as a recruitment pitch because the implied message is "the safest place to be when the flood comes is on the ark, and we're building the ark."
Your employment data is the strongest section of this piece and it needs to be cited more widely because it demolishes the narrative with actual numbers. Meta returning to 2021 headcount. Microsoft at 47% above pre-pandemic levels even after the layoff announcement. net tech employment flat at 9.6 million rather than collapsing. the gap between what the CEOs say in press interviews and what their own HR departments are actually doing is the most revealing data set in the entire debate.
The Oracle example is particularly telling. 18% workforce reduction with negative cash flow projected through 2030. thats not AI efficiency. thats a company struggling to fund a capital expenditure cycle that may not generate returns for years. framing cost-cutting as AI transformation is the corporate equivalent of putting a GPU sticker on a budget memo.
The Schiller narrative feedback loop observation is the part of this piece that should worry policymakers more than it currently does. if the apocalypse narrative causes companies to pre-emptively freeze hiring because they believe AI will replace the roles anyway, and workers pre-emptively exit sectors they believe are doomed, you get a labour market contraction that has nothing to do with AIs actual capabilities and everything to do with the story being told about them. the recession that follows gets attributed to AI when it was actually caused by the narrative about AI. thats the self-fulfilling prophecy in its most precise form.
Your three scenarios are cleanly structured. Id put rough probabilities on them: bubble bursts (20%), slower-than-expected timeline (55%), disruption faster than adaptation (25%). the middle scenario is the one that gets the least attention because its boring and hard to monetise as content. but its where most of the evidence currently points, and its the scenario where thoughtful policy on transition infrastructure could make the biggest difference.
Seems to me that call center personnel will be the among the first to lose their jobs. There are about 3 million of these persons working in the USA. It is pollyanna to claim that they will find good jobs in the AI economy. They will be retrained to do what?
This is going to age about as well as a Dianna Russini denial.
I’ve enjoyed your “outspoken contrarianism” for years (I mean, how else does a middle-aged bald man get likes on the internets), but this one just seems, well, lazy.
And I typically wouldn’t care to reply, but it’s these sorts of pieces that our lazy congresspeople will glom onto to avoid or delay much-needed policy making.
1) This time IS different for at least the reason that — as you point out — AI’s attack surface is so expansive that displaced workers will find few places of refuge. What’s also different this time, is that the folks building the disruptive tech aren’t gaslighting us like the titans of automation, enterprise SW, and offshoring did before (and still do) with the, “we enable workers to do more high-value work.” Bullshit.
2) Any commentary on this subject that uses unemployment as a (“the”) KPI should be dismissed out of turn. What a useless metric and anyone serious knows this. (“I’d like to solve the puzzle, Pat: Mass Underemployment”)
3) it’s cute to think the companies that will reap great profits from all this new found labor efficiency will re-invest it in fun, new jobs producing projects. Hope springs eternal I suppose. The cynic, however, sees more stock buybacks.
I’ve been using both ChatGpt and Claude along with some others for 2 years. It does not replace a human. It replaces the gap in our ancient brains capabilities today and the advanced technology we keep creating. It doesn’t have the human emotion and empathy that makes us superior. I have thought of things my AI miss. It compliments me. Maybe it learns like I did to always challenge its responses or not. I was at the beginning of computing in the workplace in 1982, internet in 1992 and next is AI.
There will be a tipping point. Suddenly almost everyone will prefer driverless taxis. It's coming soon.
Have your people figure out how many humans currently derive a lot of their income from driving taxis.
"Fear is the product" is the truest line here, and I'd like to back it up from the inside. I spent three decades building engagement products, and the open secret of that work is that fear and outrage simply perform better than calm. They're stickier, they travel further, they keep you scrolling. So a fear narrative doesn't just spread because it's useful to the AI incumbents. It spreads because the entire information system is tuned to reward it.
One gentle caveat: I'd be careful telling a frightened 24-year-old that the panic is "engineered," even when it is. The narrative may be manufactured, but the anxiety in their body is real, and being told it's marketing rarely makes anyone feel better. The fix isn't debunking the fear — it's giving people something solid to stand on.
Years ago, my brother, a computer scientist and engineer, shared an observation: programmers are really blue-collar workers. What they manufacture is software, but their highly skilled jobs aren't that different from the manufacturing jobs of the past. Their workstations are seats on a 21st-century assembly line.
Now, knowing that programmers created AI in their own image, we can assume that it is fairly adept at doing their job. But my guess is their belief that AI will soon be doing all our jobs is the result of an epidemic of Dunning-Kruger infecting the tech world. Someone needs to tell the technocrats that just because they understand how AI will do their job doesn't mean they understand how it will do mine.
But if for nothing else, maybe this debate can help explain why we find ourselves on a cultural precipice. For some reason, we thought we could walk away from the way things have always worked. In the past, engineers and artists had to do a dance, together building what was not only possible, but what was inspiring and beautiful. Only recently have we let the nerd run free, and it seems they’ve optimized us into a cultural cul-de-sac where feeding on one's own tales is celebrated with wide-eyed wonder. Much of AI is propped up by a plagiarism economy where the Cliff Notes version of centuries of accomplishments has been rolled up and resold with little to no thought given to origin or ownership. Worse, they’ve enticed us to strengthen their hold by collectively optimizing their tools. Let’s be clear, when folks talk about AI and say, “not yet” or “it’s just getting better,” its improvement is based on our action. It is odd; somehow that got us to pay to work for them, training our replacements. Still, I'm not quite ready to believe that we're the beta testers for societal collapse. A bit of a tech collapse, sure, the bureaucratic bloat coupled with inflated compensation in a sector that has been passing off incremental improvement as innovation for decades, heck, a financial reckoning is probably long overdue.
So yes, I can find solace in the belief that this AI charade won’t last forever; the bubble will burst. We can only hope that in the end, tech's inflated influence on culture will be the most notable casualty of the AI apocalypse. Wouldn’t that be ironic? If the ones hit hardest by the AI reality check are its creators, while the rest of us are simply a little better-off for a thinning of the algorithmic veil. Hey, if someone has the time, you want to teach your favorite LLM to master irony, it would be nice if it could contextualize its own demise.
You should listen to your own family of podcasts more sir. Alice Han on China Decode stated last week that the unemployment rate in China among 16-24 year-olds is 16.9%. She also spoke with cohost James Kynge about robots able to make dim sum (can make 3000 dumplings/hour). China has much more robot utilization than in the US. AI/robots are coming for way more jobs than you keep saying on your other podcasts
The AI job apocalypse may be overstated as a sales strategy, but overstated does not mean false
a Fortune article reported a 16% relative decline in employment for early career workers aged 22 to 25 in the most AI exposed occupations
Prof G, you the man but we still have to start a policy conversation now to accommodate the downside risk of AI replacing jobs at scale (whether self fulfilling or economically inevitable). I've written about some ideas to start the conversation now so that government policy can have a chance (probably small but worth a shot) to enact it in the next 1-3 years. Now is the time of policy moonshots to reinvent the welfare state and divide accountability for an economy that is betting the house on this:
https://skylerscoggan.substack.com/p/the-menu-no-one-is-ordering-from
My wife was a freelance academic copy editor. Was. But then everybody discovered that you don’t need one, ChatGPT does the work just as well. This week at my small business I got an SEO tech fix list from Claude that I would have paid a consultant $1,000 to provide me. 10 minutes, and it’s exhaustive. Tomorrow I’ll generate a new offering contract from Claude that I would have paid a lawyer for - probably 3 or 4 hours at $300/hr. When I hear people confidently predict that the destruction in creative destruction won’t outstrip the creation I have to wonder whether they actually are integrating these tools into their workflow. It’s not like previous tech innovations.
You just don't like change, do you?