Why the real political backlash to AI hasn’t started yet, what the politics of jobless prosperity might look like in an AGI world, and how the labs should prepare.
One thing worth adding to the 'it's the economy, stupid' section: we have solid evidence that LLMs are a deflationary tool, but no evidence yet that they're a platform—in the sense that electrification or containerisation were platforms, restructuring what was possible to organise economically rather than just reducing the cost of existing tasks. Agentic AI doesn't obviously change this; orchestrated workflows with an unreliable reasoner at each step aren't agency in any meaningful sense, and the failure modes appear at the exception-handling that defines most knowledge work roles.
The more interesting dynamic may be execution compression rather than displacement. When the cost of execution drops, firms don't typically optimise to a smaller team delivering the same output—they reach further down the project backlog. The nail gun didn't produce mass unemployment among framers; it made framing cheaper, which made more and more complicated construction viable. Most organisations have more ambition than capacity, and a productivity technology gets absorbed there first.
Compression also relocates value rather than eliminating it. When execution was expensive, procedural compliance was the dominant quality mechanism. When execution is cheap, intent and verification become the load-bearing elements—which is roughly the Jeff Koons dynamic applied to knowledge work. The roles that survive aren't diminished versions of what came before; they're the judgment-intensive parts that were always there, previously obscured by the volume of execution around them.
None of this forecloses displacement in specific sectors, or rules out a platform moment if the technology develops further (who knows). But the jobs apocalypse scenario is one tail of a wide distribution. The future, as usual, likely won't live there—and a scenario analysis across the spectrum would be fascinating, since the political economy of a deflationary-tool outcome looks quite different from a platform transition.
I wonder to what extent a potential AI backlash will have a psychological dimension as much as an economic one. If knowledge workers are indeed affected by AI automation at scale, the issue may not just be income loss or unemployment, but status collapse. A lot of politics is driven by relative status not absolute material conditions. For many knowledge workers, work is as much about identity and social proof as about money.
A “post-AGI social contract” needs a good answer so that people can still tell themselves a good story about their place in society.
You've accurately pointed out that 2026 progressive policies are already more radical than what Anthropic and OpenAI policy propose (rent control, state-run grocery stores, etc.), and that when the crisis actually hits, policy will likely escalate beyond the conception of a 32-hour work week. But if these policies do come to pass on a larger scale -- especially in urban areas -- could they potentially curtail some of the negative impacts of AI job displacement?
For example, if we had embraced Andrew Yang's UBI proposals ten years ago, would that have lessened public anxieties about AI? While I agree that "AI hasn’t yet driven Americans from their jobs and into hunger," I hold some optimism that the policies we can implement to address the current cost-of-living crisis will help us address the potential AI job displacement crisis. From my view, anxiety about AI seems to stem partially from an understanding that there are few social safety nets that the American public can depend on -- and a belief that both Democrats and Republicans are incapable of passing legislation to provide effective solutions.
Yes, I think there’s a lot to this! Low trust in government and a feeling that the economy is rigged are definitely animating a lot of the fears, and understandably so. I’m not sure what the right policy is (UBI seems like it has some tricky downsides), but the general point that the labs and government should find some way to promise people that, should the worst kind of jobs wipeout happen, there is a clear mechanism to make sure people are well off is vital.
Especially because, should that kind of jobs wipeout happen, it should come along with immense abundance. So we should be able to use it to make everyone richer and better off. But that is entirely dependent on the politics and governance of it.
another preparation might be addressing the previous wave of nascent automation that has rendered advanced countries mired in industrial obsolescence for 3 decades on, with depressed heartlands and younger precariat. that could be taken as a dress rehearsal for the bigger productivity increase coming. the transition to automation will have similarities at some level to the challenges of transition to industry and before that to agriculture. that might suggest balancing investment in domains other than confident software systems, reengineer offering of advanced countries to local and global markets, provisioning infrastructure and commercial environment to capital to unfetter it to test and develop latent comparative advantage, elevate procreation, families and classrooms to raise the mass conceptual and ethical level to meet the needs of labor to automate and operate automated society, and develop agency structure and capacity for engaging novel hazards of intelligent software development. in a word - employ people in the components of transition to automation, something the private sector and well socialized workers are likely to do fine with, it is the agencies / the state capacity / the externalities to commercial return / the superstructure, where the challenge lies.
I just published a capstone essay to bring together my previous work, which I feel has been converging on what you point out in this piece. I'm glad to see a Stanford professor is also putting historical context into what could amount to a labor disruption that goes beyond typical concentrated impacts of the past. I have subscribed, and would love to get your input on my piece, if you have the time
Agree on the individual point. The engineers leaning into AI tooling are shipping faster and with sharper judgment, and the ones treating it as a fad are visibly falling behind. That bit of the argument is solid.
The genre keeps missing the headcount maths though. If AI makes one engineer five times more productive, the company doesn't suddenly need five times the output. It needs fewer engineers. Productivity gains flow to capital, not labour, and the engineer who 'mastered AI' still loses their seat when the org chart shrinks.
We've watched this film before. Spreadsheets didn't replace accountants, but the ratio of accountants to revenue collapsed across two decades. Same mechanism, faster timeline.
The individual advice (learn the tools, stay sharp, build judgment) is fine. It just doesn't solve a market-level problem. You can master the tool that replaces you.
"In the short run, we can imagine crafting policies that commit the labs to sharing profits with society and compensating people for their job losses, only if a certain amount of measured unemployment occurs. This way, instead of the labs making the public an offer, they are offering a commitment that only activates if needed.
In the longer run, we can imagine building from our basic measurement tools to a full-blown, automated auditing system that constantly monitors data flows from government and from the labs, credibly communicating to society exactly what is going on inside the frontier labs and how it’s affecting society."
What mechanisms can we build to make sure AI labs don't try to obfuscate the true scale of AI-driven job loss to avoid paying fair compensation to affected workers? Past reporting showed that OpenAI (specifically Chris Lehane) pushed internal staff away from researching the negative impacts of AI on labor markets. We'd need to build durable, transparent ways to stop OpenAI or Anthropic from deliberately misreporting/fudging numbers to escape a tax cliff.
OK I've been thinking a lot about this -- it's a great point. I think there are probably two important parts. First, we need the government to produce better data on job loss that doesn't rely on internal lab data. There is interesting work going on now to try to improve on the surveys that BLS and others run because the response rates are terrible. We should throw a lot more into that effort. That's independent of the labs which is good.
Second, for the lab data, we need to develop internal auditing technology that provably reflects the reality of their model usage. I'm not a tech expert but my impression is that there are good cryptographic techniques that could be used to prove that the live feeds the labs provide to this hypothetical auditor are true and complete.
Thanks for the response. Certainly independent govt surveys should get more resources -- but of course there's the chance that self-reported job losses may be misattributed to AI for political reasons, and thus skewed upward. E.g. "I definitely got fired because of AI, not because of poor performance or an unrelated economic slowdown/shock."
Definitely -- I wouldn't place any stock in the self-reported cause. But data on job-related tasks paired with data from the labs and elsewhere on which tasks AI is taking on could help to triangulate from self-reported job loss.
In a similar vein to academic readiness, we should build expertise and capacity within the government itself. This may involve building agencies that can oversee AI development (even if in the immediate term they don't do much), but more broadly it would mean ensuring there are plenty of AI experts across the government in the White House, key government agencies like the Fed and the military, and on Congressional policy teams. Past experience shows that at least in the initial moments of fast-moving crises, politicians tend to be very deferential to experts within the government itself. Think about the initial responses to the financial crisis and covid, for instance. It would be good if the people they were deferring to were either actually experts on AI or had experts on AI working for them.
Yes this is a great point. Both bringing in experts and also equipping the government with AI tools and data flows that let them see what’s going on in the economy and in the labs
One thing worth adding to the 'it's the economy, stupid' section: we have solid evidence that LLMs are a deflationary tool, but no evidence yet that they're a platform—in the sense that electrification or containerisation were platforms, restructuring what was possible to organise economically rather than just reducing the cost of existing tasks. Agentic AI doesn't obviously change this; orchestrated workflows with an unreliable reasoner at each step aren't agency in any meaningful sense, and the failure modes appear at the exception-handling that defines most knowledge work roles.
The more interesting dynamic may be execution compression rather than displacement. When the cost of execution drops, firms don't typically optimise to a smaller team delivering the same output—they reach further down the project backlog. The nail gun didn't produce mass unemployment among framers; it made framing cheaper, which made more and more complicated construction viable. Most organisations have more ambition than capacity, and a productivity technology gets absorbed there first.
Compression also relocates value rather than eliminating it. When execution was expensive, procedural compliance was the dominant quality mechanism. When execution is cheap, intent and verification become the load-bearing elements—which is roughly the Jeff Koons dynamic applied to knowledge work. The roles that survive aren't diminished versions of what came before; they're the judgment-intensive parts that were always there, previously obscured by the volume of execution around them.
None of this forecloses displacement in specific sectors, or rules out a platform moment if the technology develops further (who knows). But the jobs apocalypse scenario is one tail of a wide distribution. The future, as usual, likely won't live there—and a scenario analysis across the spectrum would be fascinating, since the political economy of a deflationary-tool outcome looks quite different from a platform transition.
Enjoyed this read.
I wonder to what extent a potential AI backlash will have a psychological dimension as much as an economic one. If knowledge workers are indeed affected by AI automation at scale, the issue may not just be income loss or unemployment, but status collapse. A lot of politics is driven by relative status not absolute material conditions. For many knowledge workers, work is as much about identity and social proof as about money.
A “post-AGI social contract” needs a good answer so that people can still tell themselves a good story about their place in society.
You've accurately pointed out that 2026 progressive policies are already more radical than what Anthropic and OpenAI policy propose (rent control, state-run grocery stores, etc.), and that when the crisis actually hits, policy will likely escalate beyond the conception of a 32-hour work week. But if these policies do come to pass on a larger scale -- especially in urban areas -- could they potentially curtail some of the negative impacts of AI job displacement?
For example, if we had embraced Andrew Yang's UBI proposals ten years ago, would that have lessened public anxieties about AI? While I agree that "AI hasn’t yet driven Americans from their jobs and into hunger," I hold some optimism that the policies we can implement to address the current cost-of-living crisis will help us address the potential AI job displacement crisis. From my view, anxiety about AI seems to stem partially from an understanding that there are few social safety nets that the American public can depend on -- and a belief that both Democrats and Republicans are incapable of passing legislation to provide effective solutions.
Yes, I think there’s a lot to this! Low trust in government and a feeling that the economy is rigged are definitely animating a lot of the fears, and understandably so. I’m not sure what the right policy is (UBI seems like it has some tricky downsides), but the general point that the labs and government should find some way to promise people that, should the worst kind of jobs wipeout happen, there is a clear mechanism to make sure people are well off is vital.
Especially because, should that kind of jobs wipeout happen, it should come along with immense abundance. So we should be able to use it to make everyone richer and better off. But that is entirely dependent on the politics and governance of it.
another preparation might be addressing the previous wave of nascent automation that has rendered advanced countries mired in industrial obsolescence for 3 decades on, with depressed heartlands and younger precariat. that could be taken as a dress rehearsal for the bigger productivity increase coming. the transition to automation will have similarities at some level to the challenges of transition to industry and before that to agriculture. that might suggest balancing investment in domains other than confident software systems, reengineer offering of advanced countries to local and global markets, provisioning infrastructure and commercial environment to capital to unfetter it to test and develop latent comparative advantage, elevate procreation, families and classrooms to raise the mass conceptual and ethical level to meet the needs of labor to automate and operate automated society, and develop agency structure and capacity for engaging novel hazards of intelligent software development. in a word - employ people in the components of transition to automation, something the private sector and well socialized workers are likely to do fine with, it is the agencies / the state capacity / the externalities to commercial return / the superstructure, where the challenge lies.
I just published a capstone essay to bring together my previous work, which I feel has been converging on what you point out in this piece. I'm glad to see a Stanford professor is also putting historical context into what could amount to a labor disruption that goes beyond typical concentrated impacts of the past. I have subscribed, and would love to get your input on my piece, if you have the time
https://aprostheticmind.substack.com/p/the-cycle-technology-displacement?r=7yfpji&utm_campaign=post&utm_medium=web
Agree on the individual point. The engineers leaning into AI tooling are shipping faster and with sharper judgment, and the ones treating it as a fad are visibly falling behind. That bit of the argument is solid.
The genre keeps missing the headcount maths though. If AI makes one engineer five times more productive, the company doesn't suddenly need five times the output. It needs fewer engineers. Productivity gains flow to capital, not labour, and the engineer who 'mastered AI' still loses their seat when the org chart shrinks.
We've watched this film before. Spreadsheets didn't replace accountants, but the ratio of accountants to revenue collapsed across two decades. Same mechanism, faster timeline.
The individual advice (learn the tools, stay sharp, build judgment) is fine. It just doesn't solve a market-level problem. You can master the tool that replaces you.
"In the short run, we can imagine crafting policies that commit the labs to sharing profits with society and compensating people for their job losses, only if a certain amount of measured unemployment occurs. This way, instead of the labs making the public an offer, they are offering a commitment that only activates if needed.
In the longer run, we can imagine building from our basic measurement tools to a full-blown, automated auditing system that constantly monitors data flows from government and from the labs, credibly communicating to society exactly what is going on inside the frontier labs and how it’s affecting society."
What mechanisms can we build to make sure AI labs don't try to obfuscate the true scale of AI-driven job loss to avoid paying fair compensation to affected workers? Past reporting showed that OpenAI (specifically Chris Lehane) pushed internal staff away from researching the negative impacts of AI on labor markets. We'd need to build durable, transparent ways to stop OpenAI or Anthropic from deliberately misreporting/fudging numbers to escape a tax cliff.
OK I've been thinking a lot about this -- it's a great point. I think there are probably two important parts. First, we need the government to produce better data on job loss that doesn't rely on internal lab data. There is interesting work going on now to try to improve on the surveys that BLS and others run because the response rates are terrible. We should throw a lot more into that effort. That's independent of the labs which is good.
Second, for the lab data, we need to develop internal auditing technology that provably reflects the reality of their model usage. I'm not a tech expert but my impression is that there are good cryptographic techniques that could be used to prove that the live feeds the labs provide to this hypothetical auditor are true and complete.
Thanks for the response. Certainly independent govt surveys should get more resources -- but of course there's the chance that self-reported job losses may be misattributed to AI for political reasons, and thus skewed upward. E.g. "I definitely got fired because of AI, not because of poor performance or an unrelated economic slowdown/shock."
Definitely -- I wouldn't place any stock in the self-reported cause. But data on job-related tasks paired with data from the labs and elsewhere on which tasks AI is taking on could help to triangulate from self-reported job loss.
In a similar vein to academic readiness, we should build expertise and capacity within the government itself. This may involve building agencies that can oversee AI development (even if in the immediate term they don't do much), but more broadly it would mean ensuring there are plenty of AI experts across the government in the White House, key government agencies like the Fed and the military, and on Congressional policy teams. Past experience shows that at least in the initial moments of fast-moving crises, politicians tend to be very deferential to experts within the government itself. Think about the initial responses to the financial crisis and covid, for instance. It would be good if the people they were deferring to were either actually experts on AI or had experts on AI working for them.
Yes this is a great point. Both bringing in experts and also equipping the government with AI tools and data flows that let them see what’s going on in the economy and in the labs