The deflation-driven purchasing power gains are real and underweighted — agreed. But the boom case has a timing problem. Engels' pause --industrialisation took sixty years before median wages caught up to productivity.
The gains were real in aggregate; the distribution was brutal for decades, think of Charles Dickens' novels, Peterloo, or Marx. Political instability doesn't come from aggregate misery, it comes from visible relative deprivation during the catchup. The stability of the boom scenario depends on how long the pause lasts, not whether the gains eventually arrive.
It's amazing how you missed the bit about how if a huge chunk of the economy is laid off, they AREN'T GOING TO HAVE MONEY TO SPEND. It doesn't matter if things are cheaper. They aren't going to buy it. Material costs are going to remain the same because AI can't create that, so food, housing, and material goods are going to have a bottom cost, which will eat whatever income the laid-off contingent brings in. Not to mention: Deflation across the board is going to be catastrophic to the year-over-year gains our investment-based economy requires.
Smart move writing the mirror version. But honestly, reading both back to back left me more uneasy than either one on its own. Two equally detailed, equally coherent futures that completely contradict each other. That should bother people more than it seems to.
Maybe the future is unknowable, or we'll get different places living in opposite futures. I'm left wondering what happens in countries further removed from the epicentres of the revolution, countries with different levels of adaptability and economies more or less dependent on rent extraction. All possible combinations could be a blessing or a curse!
I also read them back to back, and what struck me first is actually what they have in common. Having done a bit of scenario development, these things often start with identifying 2 variables that are important and uncertain, and then plotting them against each other to get quadrants. In this and the Citrini piece, one continuum is of course if AI will be a benefit or a harm to the economy.
The other continuum is unnamed, but I suggest it should be will AI be economically successful or not? Both this and the Citrini piece assume AI will be wildly, fantastically successful, economically speaking. I'd also be curious to consider scenarios for the AI is a technically interesting capability with limited general economic utility because it costs so much to develop and operate at the grand scale the industry is attempting.
Bosed on the thesis of both papers, I created The Intelligence Index (TheIntelligenceIndex.com) We can now watch outcomes in real time. I hope that your scenario wins but it is slightly trailing for now... I'd love your thoughts.
Thank you. It's a good alternative scenario we should take into account. I understand where the you are coming from. Being a successful entrepreneur, your optimism, adaptability to change and ability to learning / evolve with technology change would probably put you at the top 1% of the world.
Whether there will be many job losses or many entrepreneurs / business starting new business lines imo depends on how quickly the MAJORITY of the population can retrain, adapt and find new income streams in areas where AI don't perform better than most of us. If this happens, the ability to consume will stay, lower incomes can support similar lifestyles due to increased supply of goods and services.
Without intervention (say laws capping layoffs, AI tax, effective re-training systems). A rough estimate is perhaps 20% of the population that are displaced can effectively adapt, become better off. 40% finding some way to muddle through, 40% having significant lifestyle adjustment and in need of external assistance. No science to this estimate.
One thing I do take comfort in is 2027/28 would likely be too early, given power, resource, production constraints and many companies likely moving slower to adopt AI (at the expense of market share loss) so we probably have more time to adjust.
Pure opinion, for the sake of my kids, I am happy to be wrong
One key insight is that a transaction is involved between the parties, where one party loses (the SaaS vendor) and the other gains (e.g., clients). The winner realises savings and then immediately reinvests in alternatives that extend benefits. An example of a constraint on value creation is being released. Excellent article.
Michael, this is the most rigorous articulation of the abundance bull case I've read. The distinction between demand-driven deflation and technology-driven deflation is the key insight that the bear case consistently misses.
I'd push back on one implicit assumption: the scenario treats the deflation as primarily digital -- SaaS repricing, service intermediation compression, agent-driven cost optimization. That's the easy deflation. The hard deflation -- the one that determines whether this is a real abundance boom or just a digital efficiency gain -- happens in physical goods and infrastructure.
Here's where I see it from the edge AI and IoT side: the cost of intelligence is collapsing, but the cost of deploying intelligence into the physical world is not collapsing nearly as fast. Running an AI agent that replaces a financial advisor costs pennies. Running an AI system that replaces a warehouse worker requires robots, sensors, edge compute hardware, industrial connectivity, power infrastructure, and maintenance teams. The capex curve for physical AI is steep, and it's governed by atoms, not bits.
The services deflation you describe (8-12% annualized) is credible because services are information-heavy and distribution is digital. But the sectors where most household spending actually goes -- housing, healthcare delivery, food, energy, transportation -- are physical. They require infrastructure buildouts that take years, not months. The robot that folds laundry needs connectivity that works in every room. The autonomous delivery vehicle needs edge inference and real-time mapping infrastructure. The AI-optimized energy grid needs millions of IoT sensors deployed and maintained.
Azeem Azhar's comment about Engels' pause is the right historical frame. I'd add that the length of the pause is directly proportional to how fast we can build the physical infrastructure layer -- the connectivity, the edge compute, the sensor networks -- that translates digital intelligence into real-world deflation. The digital abundance boom could arrive by 2028. The physical abundance boom depends on infrastructure deployment timelines that are measured in decades, not quarters.
The investable insight: the bottleneck isn't intelligence anymore. It's the physical infrastructure to deliver that intelligence where it matters -- factory floors, hospital rooms, farm fields, distribution centers. The companies building that connective tissue between digital AI and physical deployment will capture an enormous share of the value in the abundance transition.
The deflation-driven purchasing power gains are real and underweighted — agreed. But the boom case has a timing problem. Engels' pause --industrialisation took sixty years before median wages caught up to productivity.
The gains were real in aggregate; the distribution was brutal for decades, think of Charles Dickens' novels, Peterloo, or Marx. Political instability doesn't come from aggregate misery, it comes from visible relative deprivation during the catchup. The stability of the boom scenario depends on how long the pause lasts, not whether the gains eventually arrive.
This is excellent
It's amazing how you missed the bit about how if a huge chunk of the economy is laid off, they AREN'T GOING TO HAVE MONEY TO SPEND. It doesn't matter if things are cheaper. They aren't going to buy it. Material costs are going to remain the same because AI can't create that, so food, housing, and material goods are going to have a bottom cost, which will eat whatever income the laid-off contingent brings in. Not to mention: Deflation across the board is going to be catastrophic to the year-over-year gains our investment-based economy requires.
Smart move writing the mirror version. But honestly, reading both back to back left me more uneasy than either one on its own. Two equally detailed, equally coherent futures that completely contradict each other. That should bother people more than it seems to.
Maybe the future is unknowable, or we'll get different places living in opposite futures. I'm left wondering what happens in countries further removed from the epicentres of the revolution, countries with different levels of adaptability and economies more or less dependent on rent extraction. All possible combinations could be a blessing or a curse!
I also read them back to back, and what struck me first is actually what they have in common. Having done a bit of scenario development, these things often start with identifying 2 variables that are important and uncertain, and then plotting them against each other to get quadrants. In this and the Citrini piece, one continuum is of course if AI will be a benefit or a harm to the economy.
The other continuum is unnamed, but I suggest it should be will AI be economically successful or not? Both this and the Citrini piece assume AI will be wildly, fantastically successful, economically speaking. I'd also be curious to consider scenarios for the AI is a technically interesting capability with limited general economic utility because it costs so much to develop and operate at the grand scale the industry is attempting.
This is incredibly unique
Great article, really well written. Google negative feedback loop though
Bosed on the thesis of both papers, I created The Intelligence Index (TheIntelligenceIndex.com) We can now watch outcomes in real time. I hope that your scenario wins but it is slightly trailing for now... I'd love your thoughts.
I love it. Great work.
You asked for @michael bloch's thoughts and not mine, but here are mine, which I hope are helpful:
I think that the US is going to experience one side and China the other.
AI is going to displace many white collar workers and AGI will likely be here before 2033.
In a society that values capital over people I think it rational to expect the gains from AI to not be widely shared, but hoarded.
In a society that subjugates capital, on behalf of people, I think it rational to expect the gains of AI to be more widely distributed.
Just my thoughts.
Wow! That’s is a mind shifting perspective. Uni-dimensional but very interesting.
Thank you. It's a good alternative scenario we should take into account. I understand where the you are coming from. Being a successful entrepreneur, your optimism, adaptability to change and ability to learning / evolve with technology change would probably put you at the top 1% of the world.
Whether there will be many job losses or many entrepreneurs / business starting new business lines imo depends on how quickly the MAJORITY of the population can retrain, adapt and find new income streams in areas where AI don't perform better than most of us. If this happens, the ability to consume will stay, lower incomes can support similar lifestyles due to increased supply of goods and services.
Without intervention (say laws capping layoffs, AI tax, effective re-training systems). A rough estimate is perhaps 20% of the population that are displaced can effectively adapt, become better off. 40% finding some way to muddle through, 40% having significant lifestyle adjustment and in need of external assistance. No science to this estimate.
One thing I do take comfort in is 2027/28 would likely be too early, given power, resource, production constraints and many companies likely moving slower to adopt AI (at the expense of market share loss) so we probably have more time to adjust.
Pure opinion, for the sake of my kids, I am happy to be wrong
One key insight is that a transaction is involved between the parties, where one party loses (the SaaS vendor) and the other gains (e.g., clients). The winner realises savings and then immediately reinvests in alternatives that extend benefits. An example of a constraint on value creation is being released. Excellent article.
Michael, this is the most rigorous articulation of the abundance bull case I've read. The distinction between demand-driven deflation and technology-driven deflation is the key insight that the bear case consistently misses.
I'd push back on one implicit assumption: the scenario treats the deflation as primarily digital -- SaaS repricing, service intermediation compression, agent-driven cost optimization. That's the easy deflation. The hard deflation -- the one that determines whether this is a real abundance boom or just a digital efficiency gain -- happens in physical goods and infrastructure.
Here's where I see it from the edge AI and IoT side: the cost of intelligence is collapsing, but the cost of deploying intelligence into the physical world is not collapsing nearly as fast. Running an AI agent that replaces a financial advisor costs pennies. Running an AI system that replaces a warehouse worker requires robots, sensors, edge compute hardware, industrial connectivity, power infrastructure, and maintenance teams. The capex curve for physical AI is steep, and it's governed by atoms, not bits.
The services deflation you describe (8-12% annualized) is credible because services are information-heavy and distribution is digital. But the sectors where most household spending actually goes -- housing, healthcare delivery, food, energy, transportation -- are physical. They require infrastructure buildouts that take years, not months. The robot that folds laundry needs connectivity that works in every room. The autonomous delivery vehicle needs edge inference and real-time mapping infrastructure. The AI-optimized energy grid needs millions of IoT sensors deployed and maintained.
Azeem Azhar's comment about Engels' pause is the right historical frame. I'd add that the length of the pause is directly proportional to how fast we can build the physical infrastructure layer -- the connectivity, the edge compute, the sensor networks -- that translates digital intelligence into real-world deflation. The digital abundance boom could arrive by 2028. The physical abundance boom depends on infrastructure deployment timelines that are measured in decades, not quarters.
The investable insight: the bottleneck isn't intelligence anymore. It's the physical infrastructure to deliver that intelligence where it matters -- factory floors, hospital rooms, farm fields, distribution centers. The companies building that connective tissue between digital AI and physical deployment will capture an enormous share of the value in the abundance transition.