There is quite a bit of chatter about whether AI is the best thing or the worst thing that has ever happened to us. Fair enough, I’m still figuring this out as well, it could be either. AI is impressive, it is a force multiplier in ways we have never seen before. As I’ve been tackling this question, I keep coming back to how we are in the middle of what Strauss and Howe called a Fourth Turning.
This framework of the history of civilizations says at the resolution of the fourth turning actually ends when the people in charge stop agreeing on what to do about it. It is generally agreed the fourth turning started around the 2008/9 financial crisis. That didn’t resolve. Covid didn’t. The wars haven’t.
Because of the massively disruptive potential of AI, what I want to know is whether AI is the thing that finally resolves this cycle.
AI showed up as a bet, not a crisis
Every week there seems to be a new take on AI. It is going to cure cancer. It is going to take every job. It is going to save us or kill us, sometimes both in the same article. Recently a researcher at one of the frontier labs resigned and said the labs are gambling with our lives racing toward systems that can improve themselves. That got a lot of attention, of course it should have.
What caught my attention was something else, not the warning, rather who inside the buildings were agreeing with him.
I buy into the concept that Western society moves in cycles, most everything I see is cyclical. Strauss and Howe laid it out in their book The Fourth Turning (1997) with a follow up, The Fourth Turning is Here (2026): roughly eighty to ninety years per cycle, four turnings in each one, and the last turning they call the Crisis.
Most people who use the framework say our current Crisis started ~2008/9. I subscribe to the basics of the framework because when I look at the last eighteen years the pattern fits what I see. I don’t take it as a law of history. Plenty of serious historians think it is pattern matching dressed up as theory, and they might be right.
For me it is a useful way to organize what has happened since the end of WWII. The pattern matches.
While the mid 20th to early 21st century haven’t been all peaches, let’s look at the past ~20 years. The financial crisis. Covid. Two wars running at the same time. Inflation that ate a whole generation’s ability to buy a house. And now AI.
AI doesn’t fit the list the way the others do. 2008 was a financial system failing. Covid was a virus. The wars, too political for me to address, we can agree war is bad. All were bad things that happened to people. Nobody set out to build them (arguable). Not including AI just yet, these events are cracks in a society.
AI is simply in a league of its own. People built AI because they believe in the upside, and a lot of that upside is real. I use it every day. I’ve built a business on it.
So I don’t think of AI as a crack in the dam the way the others were. I think of it more as a crack generator. Whether it actually produces a crack, and how big, depends on where it goes from here. The capability could keep climbing or it could plateau. The economics could produce broad gains or mass displacement. It could stay a tool that humans point at problems, or it could stop being a tool.
Nobody knows. The people building it don’t know. That is what makes AI different from every other event in this cycle. The others were done deals by the time we noticed them. This one is still being decided.
The current Fourth Turning
I’m interested in a few things, first is what will end the current cycle, and second, everything AI. Is AI going to push us to a resolution? If we have had current societal stressors, how does AI fit the patterns?
The financial crisis ended with bailouts, near zero interest rates, and quantitative easing. The acute part ended. The structural exposure didn’t. The banks that were too big to fail got bigger. The masses got cheap credit, raised housing prices out of the range of younger generations, and helped set the stage for massive inflation overall post 2020.
Covid ended with emergency spending on a scale nobody had ever seen, a reopening, and then inflation. The acute part ended. The fiscal and generational damage didn’t. A kid who was nineteen in 2020 is twenty five now and has never known what a normal entry into adult life looks like.
The wars current wars are different. They haven’t been patched. They are still going. But they are contained by proxy structure and by the fact that they depend on specific administrations wanting to keep them going, and that containment is what keeps them from becoming the event that ends the cycle on their own. They put strain on the system, but they aren’t what ends it.
Inflation is the bill for the earlier patches showing up late. It landed hardest on people who work for wages and on the generations trying to get into the housing and job markets at exactly the wrong time.
It is the same shape every time. The acute event ends. The patch protects whoever already holds assets. Wealth concentrates a little more. Grievance builds a little more. Nothing gets resolved, and then the next event lands on top of the duct tape from the last one.
That is the pattern I see. The question that interests me is why it keeps repeating.
Why patches never fix anything
My working theory, a Fourth Turning resolves when the people in charge stop agreeing on what to do about it. How big the crisis gets has surprisingly little to do with it.
Of course patching is the default. If you are already on top, you hold the reins, a patch is the lowest risk move available to you. It preserves (or grows) your assets and your positions. A real resolution puts both of those at risk. So regardless of ideology, the people with the most to lose reach for the patch, and they agree with each other while they do it.
2008 fits. Both parties agreed on the bailout playbook. Plenty of people were furious about it, and none of that fury changed the response. Covid fits too. The people in charge agreed on the spending and the reopening timelines, and again the anger didn’t matter.
People yelled/are yelling constantly during the financial crisis, Covid, the wars. Real fracture looks like a party splitting. An institution breaking. A policy that can’t get built because the power to build it has divided too sharply. If the thing still gets built, the people in charge are still aligned, however loud the crowd gets.
Loud public disagreement is not the same thing as the elites fracturing.
This isn’t unique to our cycle either. I’m not going to teach history here, I only want to note that the earlier Crisis eras fit, as also shown by Howe and Strauss, and many others. Before the Civil War you had decades of legislative compromises deferring the slavery question until the Democratic Party split in 1860 and a unified response became impossible. The New Deal didn’t end the Depression, and isolationist and interventionist elites stayed divided until Pearl Harbor forced their hand. Before the Revolution, colonial and British elites got to a point where compromise stopped being good enough for one side.
Even with different visual causes each time, there were the aame fractures underneath.
You can actually see the early stage of this type of fracture forming right now, and interestingly it isn’t around AI.
There is an insurgent wing inside the Democratic coalition going after its own establishment leadership. At a June 2026 primary victory party in New York, the crowd chanted “You’re next!” when Hakeem Jeffries (Minority Leader of the U.S House of Representatives) came up on screen. Before that a city councilman had filed to primary him, and then the insurgent wing’s leaders including Mamdani and Ocasio-Cortez stepped in and discouraged the challenge. That is very similar to what happened before 1860, at the stage where containment is still winning. The Republican coalition is having their own issues as well.
This is worth watching, while clearly not a break yet.
When a Crisis actually resolves, the fracture is rarely elites against the masses. It is elites against each other, and the disagreement usually comes down to self preservation more than morals. Some of them calculate that relieving the pressure is what keeps them safe long term. Others calculate that holding the line does. That is what actually split the Democrats in 1860, and it is what I see as the more useful thing to watch for now.
I’m much more interested in AI, and how AI’s disruptive capabilities will affect us all.
AI cannot cause civilization scale disruption on its own.
Deploying AI at a truly disruptive scale takes institutional compute and institutional money. OpenAI’s Navier-Stokes result, impressive as it was, ran ten thousand agents for about eighty eight hours, and OpenAI’s own executives put the compute cost in the millions of dollars. Nobody does that from a basement. A company made that decision, budgeted for it, and pulled the trigger.
So even when AI is the tool, its disruptive ability, its position as a Crisis crack, runs through a human decision point, and whether the people at that decision point are aligned or fractured is what matters. The model’s benchmark score comes after that, not before, they are a distraction.
To be fair, this framework can’t be tested in the short run. You can’t confirm you are in a Crisis until you see how it resolves, so nobody can call the final crack in advance, and I’m not trying to. Second, and more important, once a pattern, such as that described in The Fourth Turning gets named and popularized, people can consciously play to it or against it. Nobody in 1860 or 1935 was reading a book about turnings. We are.
It is very possible the awareness of this pattern might be distorting the very pattern it describes.
Just because we cannot prove the pattern nor the stage, or that the pattern now described may break its ability to predict the future does not break my argument. Rather I see these points placing a fence around it.
So the question, at least for me, is whether AI breaks elite agreement the way slavery and intervention did, or stays noise the way 2008, Covid, and even the ongoing wars to a point did/do.
Where AI sits in that test right now
Congress has been trying to preempt state AI regulation for over a year and failing. A proposed ten year moratorium on state AI laws was stripped from the 2025 reconciliation bill on a 99 to 1 Senate vote. A second attempt through the defense authorization bill failed too. This year the Obernolte-Trahan discussion draft split opinion all over again. Many House Democrats said its preemption went too far. Many Republicans and industry groups said it didn’t go far enough. The House’s own AI Commission, co-chaired by Ted Lieu, who used to be Obernolte’s task force partner, put out a statement saying the draft couldn’t serve as the basis for productive dialogue.
That is a bill that can’t get built because the power to build it is genuinely divided. And the division isn’t the normal kind. Normal gridlock is one party blocking the other. This is opposition coming from both directions on the same bill at the same time. The disagreement runs right through both parties, which tells you it is about the underlying question and not about partisan positioning. That is much closer to what elite fracture actually looks like than a party line vote would be.
Of course bills stall all the time for boring procedural reasons. Stalling by itself proves nothing. What makes this one different is the cause of the failure. Nobody has been able to put together a majority from either side. That isn’t as common on big issues. Then along comes California, often seen as a leader of legislation, who’s governor just signed legislation that aims to safeguard its population from AI.
It seems that everyone in Washington agrees China is a threat in respects to AI. The preemption camp argues a patchwork of state rules is a handicap in that race. And the same camp making that argument can’t agree among themselves on whether the answer is federal control or state control. Agreement on the external enemy is simply cover for disagreement on the internal response to AI.
With this all in mind, pivot back to the resignation I mentioned in the beginning. Jacob Coxon left Anthropic on September 8 and said the frontier firm understands the danger AI poses, yet races anyway! Why? Because it believes nobody else will act responsibly! Wow! Evan Hubinger, a sitting alignment lead at Anthropic, backed him publicly and actually went further.
Take the test I laid out above, the public resignation is just one data point, not a fracture in and of itself. Nobody’s institution split. No policy became unbuildable. The lab is still shipping. It matters as an early tremor. If departures multiply, or a lab actually changes course because of internal pressure, then you have real fracture right where the technology gets made, long before it ever shows up in politics.
When the resignation hit the news, the common thread I noticed was seeing this as a story about how dangerous the models are. I read it as a story about whether the people building these systems still agree with each other. A researcher saying the models are dangerous is a Tuesday. A researcher saying it, and a senior alignment lead at the same company saying yes, he’s right, while the company keeps shipping toward an IPO, tells you the agreement inside the building is thinner than it looks from outside.
Now we are getting to my point. Stop watching the models. Start watching the people.
The AI policy fights are breaking down the same way. Accelerationists and the safety camp both have real money and real influence. Labor populists on the left and the right have landed on the same complaint from different directions: juniors aren’t getting hired, wages aren’t tracking output, and the wealth is piling up with whoever owns the infrastructure. Capital owners need the AI story to keep running. Regulators and safety people need visible guardrails once the disruption gets real. Those interests were aligned during the buildout. Nothing guarantees they stay aligned once the bill comes due.
The signal to watch from here is simple. Does the preemption fight resolve into a federal / global framework that actually passes, which would mean managed consensus reasserted itself? Or does it keep stalling across bill after bill, which would mean the fracture is real and sustained? If something structurally new passes, the fracture resolved into an outcome. If we get years of stalemate across multiple approaches, that is standing fracture. Not noise, but not resolution either.
Will I be shown I’m wrong?
If the preemption fight resolves into an accepted framework within the next few legislative sessions, and the internal dissent at the labs produces no more departures and no institutional consequences, then the fracture I’m describing was noise. The elites held, same as 2008, same as Covid, and I was wrong.
If it fractures
If elite agreement does break, the framework says the Crisis resolves. It doesn’t say gently. Every historical resolution Howe and Strauss named came with a price. What happens if AI causes the final fracture that ends the current cycle? I have my opinions and I aim to put a price on each scenario rather than let “resolution” sound like a nice word.
Revolt, or political rupture. A new governing coalition, not just a change of administration. i.e. American Revolution, French Revolution, Red October. This could be progressive left, populist right, too difficult to say. This is the scenario fed by mass displacement without offsetting gains. AI has been claimed it can cause: junior hiring collapsing, wages stagnating, a generation locked out of both housing and entry level work. A bubble burst would compound it and destroy retirement accounts and debt service. The historical cost is the Civil War. Roughly 620,000 Americans dead, more than every other U.S. war combined at the time. That is the general price tag attached to “new governing coalition,” this isn’t a peaceful realignment at the ballot box.
Authoritarian consolidation. Authoritarianism doesn’t need a coup or a leader who refuses to leave. It needs infrastructure. Surveillance, monitoring, control capacity, built during the crisis under a banner nobody can argue with. I’m thinking national security or fraud prevention or public safety, infrastructure that outlasts the administration that built it and gets used by whoever holds power next.
AI is the tool here. It stays a tool, but it becomes the mechanism power uses to build control at a scale no prior crisis, no prior society, no prior despot had available. The historical cost here comes in a different form. Cold War surveillance programs like COINTELPRO and the mass wiretapping efforts ran for decades past whatever justified them. The cost is permanently narrowed civil liberties that outlive the crisis that made them seem reasonable, not to mention the millions of people who died or were executed under such a rule.
Forced external unity. A hot conflict, most likely involving China and the US on opposite sides, that forces domestic cohesion the way WWII did. I think this is the least likely of the three. It is also the closest match for how the last two Crisises actually ended, which should give anyone pause.
AI feeds it because both sides are racing for it as a strategic asset, which turns the contest over the tool into the contest itself. The historical cost is WWII. Somewhere between 70 and 85 million dead, roughly three percent of everyone alive at the time. The scenario that most cleanly resolves a Crisis is also the most catastrophic by a wide margin.
The path of AI
We don’t know exactly where AI is going nor where it will take us. Two of AI’s possible paths don’t feed fracture at all.
If it produces broad, shared gains, the crack is minimal. If capability plateaus and juniors get hired again in a few years, the disruption was temporary.
If there is loss of control, such as self improving AI that recently hit the wire, the one path where AI stops being a tool and could independently cause the resolution directly instead of cracking the dam for humans to fight over. I personally treat it as the low probability exception to everything else, for the reason I gave earlier, as well as have written about extensively. Doing anything with AI at that scale still takes institutional resources, and institutions are run by people who can turn things off.
One thing runs underneath my findings. AI is a multiplier on whatever direction humans are already headed. The same tool that speeds up real productivity speeds up displacement. It speeds up the surveillance buildout if that is the political response. It speeds up loss of control if alignment fails. It shortens the time it takes to get wherever we were already going. That is why none of these scenarios is decided in advance.
The multiplier applies to every branch the same way, and which branch we end up on comes down to human choices, most of them made by a small number of people who currently agree with each other.
If it doesn’t fracture
If we keep the status quo, that means the people in charge manage AI the way they managed the previous fractures. Bailouts. Retraining programs. Symbolic regulation. Enough to blunt the sharp edges without touching who captures the gains.
We are watching this happen right now. Challenger, Gray & Christmas counted 87,714 AI attributed job cuts through May 2026, already past the total for all of 2025, with AI the leading stated reason for layoffs three months running. Three weeks later Amazon, Anthropic, Microsoft, and the OpenAI Foundation announced they were backing RAISE US, a retraining fund fronted by a former Commerce Secretary and a former governor from opposite parties, explicitly billed as nonpartisan, with more than $500 million committed toward a $1 billion goal.
That is the textbook move. The companies responsible for the displacement fund the visible remedy. Elites across party lines endorse it. The structural question, whether those jobs are coming back at all, never gets asked.
Let’s give the skeptics their due here. They might be right. Large scale worker retraining has roughly ninety years of policy history behind it and very little to show for it. Both parties can support skills training without ever having to say out loud that millions of jobs might simply be gone. That is exactly why retraining is the politically safe answer. It doesn’t require anybody to admit the harder possibility.
If this is the path we’re on, AI becomes one more crack in the stack instead of the final one. Grievance keeps compounding even while the visible response looks fine on paper. That space between “we announced a program” and “we addressed the problem” is exactly where the next patch’s failure gets stored up. And it is a huge space to be sure.
That outcome would fit this possibility just fine. It is the pattern since 2008 running one more time. If a year or two from now we have well funded, bipartisan branded retraining programs, layoffs still climbing, and no structural change to who captures the gains, that is what patched-not-resolved looks like while it’s happening.
And yet, I might well be wrong
Nobody can call the end of the cycle, the framework itself rules that out. What I can show is that AI sits on the same fault line as every crack since 2008, whether it wins or loses on its own economic terms, and that the thing worth watching is whether the elites stay aligned, not how capable the models get.
I’m watching for two possibilities that would prove the mechanism wrong.
The first is a resolution without a fracture. The system gets reordered at Fourth Turning scale, new institutions, a new distribution of who captures the gains, while the people at the top stay in agreement the whole way through. If the elites voluntarily and collectively restructure who benefits from AI, without being forced into it by a split among themselves, then fracture was never necessary and I have this backwards. There is no historical precedent for it, which is exactly why it would be a clean falsification.
One distinction inside that. If elites share the gains because some of them got scared of a split and moved to head it off, that is the mechanism working, not failing. Only sharing that happens with no visible internal divergence counts against the possible outcome. The problem I see, I’m not sure that is even observable from outside.
The second is the one I think is more likely, and it is the one that would take down not just my entire premise, but the whole cycle theory underneath it.
Call it the whimper.
The elites stay aligned, manage every disruption indefinitely, and the Fourth Turning ends without any climactic resolution at all. Grievance stays high. Nothing breaks. The cycle rolls over into a new one with no reordering.
If that happens, my fracture argument might still be true about dramatic resolutions, but it would be beside the point, because the framework’s assumption that Crisis eras end in a climax would have been wrong all along.
I take the whimper seriously. I trouble with discounting it even.
The elites have managed every crack since 2008 successfully by their own measure. And a class that can personally insulate itself from the fallout, with private security, off grid compounds, and second passports, has less reason to fracture, not more. They can manage from behind the wall because they aren’t the ones paying for the patches. People point at billionaire bunkers as proof the elites know a break is coming. I read it differently. Hedging tells you they see risk. Whether they expect to lose, or expect to split, is a separate question the bunkers don’t answer.
My gut says the whimper fails, and the reasons are structural rather than about what any individual power player believes.
The debt math has a limit. The patch tool since 2008 is cheap money, and every use of it adds to a debt load whose interest cost eventually competes with everything else the government does. That limit doesn’t move because the people at the top have bunkers.
The demographics have a clock. The generation locked out of housing and now locked out of entry level work is becoming the voting majority while the generation with a stake in the current arrangement ages out. That one is arithmetic, and arithmetic doesn’t pass.
And finally the population has to stay bought. Managed consensus only works while people are pacified enough not to force the issue. Wealth concentration and generational lockout are the opposite of pacification. At some point the price of keeping people bought goes past what some of the elites are willing to pay, and that is the exact moment their self preservation calculations start to diverge.
Which means the whimper, if it runs long enough, is what produces the fracture. Sustaining it requires the elites to keep paying. The first ones who decide the payment isn’t worth it are the split.
In the AI debate, watch the elites, not the models
Everybody is watching the benchmarks. Every release, every capability jump, every leaked eval, and then the same argument about whether it’s real or hype.
I’d watch something else.
Watch what the elites do. Watch whether AI legislation in Congress passes or keeps stalling. Watch whether the resignations multiply or stay singular. Watch whether the retraining money gets followed by anything structural or whether it is the whole response. Watch whether the people who currently agree with each other keep agreeing.
If I’m right, the next few years will show elite agreement fraying in places that have nothing to do with how smart the models got. The factures in society will continue to grow, most likely to a hard breaking point.
If I’m wrong, the frameworks will pass, the labs will go quiet, and the retraining programs will allow peoples’ jobs and livelihoods to rebalance, and that will be the end of the story.
Either way, the powers that decide what to do about the models are the ones worth watching.
Speak Your Mind