On the point at which the frontier stops being the interesting part

I have been losing interest in frontier models.

This has surprised me. I have spent the better part of two years building frameworks for thinking about how AI systems behave, why they fail, what the missing architectural primitives are, how organizations can develop the signal discernment to work with them. I have written about Anthropic’s structural advantage over OpenAI. I have written about the inverted bubble, the chip lifecycle economics, the chip valuation death schedule. I had a piece queued up; a coda, a synthesis, that was supposed to sit on top of all of it.

I keep not finishing it. The reason, I think, is that the frontier has stopped being where the interesting thing is happening.

The interesting thing, the load-bearing thing, the thing that will determine whether the next decade is survivable, is happening one layer up. It is happening in procurement. It is happening in the vendor ideologies that used to be inferred and are now declared. It is happening in the collapse of the distinction between software and political project. It is happening in the specific mechanisms by which technology is enabling the worst impulses of our worst leaders and, more alarming than that, producing the conditions under which the worst people can become leaders in the first place.

Mythos remains interesting to me because Mythos is a forcing function on the real world. It changes ransomware economics. It changes enterprise security calculus. It intersects systemic financial risk at load-bearing points. The rest of the frontier lab discourse; who is ahead, whose benchmark beat whose, which founder said what on which podcast, has the texture of a horse race being run inside a burning stable.

This is not a complaint about the work. The work is serious. It is a confession that my attention has moved, and an attempt to follow that movement honestly rather than pretend I am still interested in the thing I am no longer interested in.

What happened on Sunday

On Sunday, April 19, 2026, the Prime Minister of Canada told the country that our bond with the United States, for generations the pillar of Canadian security and prosperity, has become a weakness. He invoked Isaac Brock. He said he would not sugarcoat our challenges. He named the challenge: the thing that used to be the strength is now the thing to correct.

The same morning, Palantir Technologies posted a summary of its CEO’s book as a set of 22 numbered theses on social media, prefaced with “because we get asked a lot.” The document is a foreign-policy doctrine. It argues for the re-militarization of Germany and Japan. It calls for universal conscription in the United States. It asserts that “some cultures have produced vital advances; others remain dysfunctional and regressive.” It positions Palantir, a vendor with pre-approved status on the Canadian federal procurement list through 2028, as the software layer of a new deterrence regime, explicitly rejecting pluralism as a luxury the West can no longer afford.

Also on Sunday, Jake Tapper spent a segment of State of the Union methodically reading Donald Trump’s words back to Elise Stefanik. Your whole civilization will die tonight. He made her defend it. He would not let her euphemize it into “the regime.” He said, twice, that she was applying two different standards. Jake Tapper is not a lefty. Jake Tapper is a mainstream Zionist CNN anchor who spent the segment treating “the President of the United States threatening to annihilate a civilization” as the plain-English fact that it was.

Also on Sunday, Doug Ford announced he would sell the $28.9 million private jet that Ontario had bought for him three days earlier, after public backlash he had evidently not anticipated. The plane had not even been delivered yet. He had spent the weekend discovering that his own province could not be made to agree that a premier with a $485 billion debt load needed a Bombardier Challenger.

Four things, one Sunday. Each one small in isolation. Together: a tell.

Every one of these is a tech story, eventually

The Carney address is a procurement story. It is about the layer of the stack, Layer 5, institutional and procurement governance, that was never really governed at all, that was assumed to be too boring to matter, and that is now the place where the sovereignty of a G7 country is leaking out.

The Palantir manifesto is a tech story, openly. It is a software vendor publishing a political platform under its own brand and asserting that its products are the operational arm of that platform. This is not normal. Microsoft does not do this. Oracle does not do this. SAP does not do this. The only reason Palantir can do this is that enough of its revenue comes from defence and intelligence contracting that it believes it does not need to hedge, and enough of its political alignment comes from its founding that it believes declaring the hedge was never needed.

The Tapper-Stefanik segment is, at root, a story about what platforms, Truth Social, X, and the feedback loops they create inside political parties, have done to the Overton window around genocide as a permissible thing to threaten from the office of the President. Stefanik was not inventing the euphemism on the spot. She was reciting it. Somebody trained her on it. The training happened on platforms.

The Ford plane is a story about an information environment in which a third-term premier can genuinely not anticipate that his constituents will object to a $28.9 million jet during a cost-of-living crisis, because his information environment is now so thoroughly a hall of mirrors that the reality outside it arrives, when it arrives, as a surprise. The hall of mirrors is built on tech.

Every one of these is a tech story. The technology didn’t cause Alex Karp’s Frankfurt School dissertation, or Doug Ford’s self-regard, or Donald Trump’s genocidal impulses, or Elise Stefanik’s willingness to defend them on national television. Those things were there already. What the technology did was enable them. It lowered the friction. It removed the intermediate steps where somebody, somewhere, used to say no, that’s too far, that won’t play.

The Karp tell

I did not know, until this weekend, that Alex Karp’s 2002 PhD dissertation at Goethe University Frankfurt was on aggression as a mechanism of social integration. Specifically: on how jargon functions to bind communities together by giving them a shared vocabulary for the expression of aggression toward outsiders.

He wrote it under Karola Brede. Jürgen Habermas was apparently his original intended advisor; the relationship ended before the dissertation did. The thesis was an extension of Adorno’s Jargon of Authenticity, Adorno’s argument that certain kinds of philosophical language function as ideological sign-weapons, hollow at the core, whose purpose is to elevate the in-group and exclude the out-group.

Karp’s dissertation took this descriptively rather than critically. That is the part the Frankfurt School people have been writing about for five years. Where Adorno saw jargon as something to be criticized and resisted, Karp treated it as a functional mechanism to be understood. He produced, as a 35-year-old graduate student, a technical account of how aggressive language binds communities together.

Twenty-four years later, he has published a 22-point manifesto, written in saturated jargon, that deploys aggression to bind a community together against declared cultural inferiors, opponents of hard power, and the “shallow temptation of a vacant and hollow pluralism.”

He is applying his own PhD to the problem of Silicon Valley political consolidation. The dissertation is the playbook. The manifesto is the execution.

This should be more widely understood than it is. When a man who wrote his doctoral thesis on the technical mechanics of how aggressive jargon produces in-group cohesion then produces a jargon-saturated, aggressively framed manifesto explicitly arguing against pluralism and for hard power, he is not being sloppy. He is not accidentally reinventing fascism. He is doing the thing his dissertation described. He knows exactly what he is building. The jargon is the mechanism. The manifesto is not a political platform; it is a ritual text for a community being consolidated around a specific form of aggression.

That is the vendor on Canada’s Software Licensing Supply Arrangement through 2028.

The enablement thesis

Here is what I think the throughline is, and what I think the piece I should actually be writing is about.

We have built, over the last fifteen years, a technological substrate whose primary political effect is to lower the friction for the expression and coordination of the worst human impulses, while simultaneously raising the friction for the slower and more deliberative institutional processes that used to contain them.

The substrate includes social platforms, recommendation systems, large-language-model-mediated discourse, data-fusion platforms, surveillance infrastructure, and the procurement regimes that route all of it into the operational core of the state. The substrate is not neutral. It was not designed to be neutral. It was designed to extract attention, to maximize engagement, to concentrate capital, and, increasingly openly, as with Palantir, to advance specific political projects.

What it has enabled is not merely political polarization or information disorder, though it has enabled those. What it has enabled is the rise of people to leadership roles who, in the institutional environments of thirty years ago, would have been filtered out by the slow accumulation of written records, deliberate counsel, and professional check-and-balance. The filter is gone. The filter was slow and deliberative; the substrate is fast and affective. The people who rise now are the people who are best at the substrate, which is a set of skills substantially disjoint from the set of skills required to govern.

This is the enablement thesis. It is not a thesis about AI specifically. AI is an accelerant. The substrate (a term I barely used before AI gained dominance) precedes AI and will outlast whichever lab currently has the highest benchmark. But AI is the most powerful accelerant yet deployed, and the political project Palantir has now declared in writing is an attempt to couple AI’s accelerant properties directly to state power, under vendor-controlled terms, with declared ideological commitments that are incompatible with democratic pluralism. Slouching away from Bethlehem, at pace.

The consequences fan out. They fan out into the Strait of Hormuz, which could close this week and take eighteen months of global food supply with it. They fan out into the information environments that produced the 2024 American election and the 2026 Canadian one. They fan out into the procurement files where Canadian defence, policing, and soon healthcare data is quietly routed through a vendor that has just published a cultural-hierarchy manifesto. They fan out into the rooms where Jake Tapper has to read a president’s words back to a congresswoman because the congresswoman’s entire political class has lost the ability to call the words what they are.

We are not watching a car wreck. A car wreck has a clear end state, a few casualties, an insurance adjuster, a tow truck. We are watching something on the scale of the event that killed the dinosaurs. An extinction-level punctuation mark. A before-and-after with no restoration on the other side.

And here is the thing that actually makes it a meteor rather than a 747 collision or any other metaphor of human error: we built it. The substrate is on a trajectory now that is substantially independent of the humans who built it. Nobody can call it off. The platforms cannot be un-launched. The recommendation systems cannot be un-trained. The Supply Arrangements run to 2028 and the vendor ideologies have been declared and the political class has been selected through a filter that no longer filters, and the thing is coming down whether or not we are ready for it.

The dinosaurs did not know. We do.

That is the only difference, and it is not a small one. It is the whole difference. The dinosaurs could not name the meteor. We can.

What I should be writing about

I have been asking myself, for weeks, why I have lost interest in the frontier model story.

I think the answer is that the story I was telling about the frontier, about Anthropic’s structural advantage, the coming consolidation, the economics of the chip lifecycle, was a story about how people would be affected by which company would win. How life would change. Because it is changing, and maybe this time the tech moral panic is valid. That story continues to be true and continues to be well-sourced and continues to be the thing I know how to do.

But winning is no longer the matter of significance. It is whether the thing built on top of the winner is survivable. And that question cannot be answered at the frontier layer. It can only be answered at the layer where the frontier meets institutions, vendors meet governments, ideologies meet procurement files, and declared cultural hierarchies meet the Supply Arrangements that extend past the next election.

That is the layer I have been writing about, on and off, for months, without naming what the project was. The sovereignty series. The Palantir piece. The nudgment work on organizational signal discernment. The drift signatures work on whether we can even detect when the substrate is failing. It has all been one project. The project is: how technology is enabling the demise of citizenry, and what, if anything, can be done about that from inside the institutions that still exist.

I did not know this was the project until this weekend. I am writing it down now because I want to stop pretending the old project is the one I’m still doing.

Who cares about frontier models. Who cares, frankly, about Mythos, when the Strait of Hormuz might close and people cannot access enough food because the person in the White House is publicly threatening to kill a civilization and the congresswoman on CNN cannot bring herself to say that is what he said.

The meteors are in the sky. We made them. They are coming down. What will the survivors do afterwards? and more urgently what can still be done in the interval between now and impact. The institutions are still there, most of them.We are on the verge of mass food shortages, and poverty exploding, and already vulnerable communities being intentionally made more vulnerable by a guy (among others) who nearly studied under Habermas. The procurement files are still open. The sovereignty framework still applies. The vendor has declared itself; the Prime Minister has named the dependency; a senior anchor on a centrist network is no longer willing to euphemize. Those are four data points, on one Sunday, of the substrate beginning to be named rather than absorbed.

That is where attention must go. Which model wins is no longer an interesting story, nor is it that important. The irony is that what is important is staring us straight in the face, but we are collectively prevented, in multiple ways, from looking at it. By our own unwillingness, by distractions, by intention. The impending impact is over our heads, but we can’t see it.

Jennifer Evans is the founder of Pattern Pulse AI.