John Herrman

@jwherrman.bsky.social

8.0Kfollowers

568following

630posts

posting about posts at new york magazine. have me on your podcast!

Latest posts

John Herrman·5d

Chatted with @ftrain.bsky.social and Rich Ziade about my thwarted and slightly pathetic attempts to get something out of OpenClaw, the AI assistant that we are all definitely still talking about a few months later aboard.com/podcast/john...

John Herrman·5d

The AI industry is prone to massive, frequent, and intensifying narrative mood swings. Part of this is typical of a boom. But I think it's also intrinsic to the tech itself nymag.com/intelligence...

There are plenty of reasons for this pattern. We're going through a massive investment cycle in a new technology, which understandably produces questions about bubbles, polarizes the economic discourse, and invites trader-brained, boom-or-bust interpretations of new developments. The economic stakes are now unavoidably high, and whatever happens is genuinely everyone's problem. Relatedly, while there's plenty of coverage of Al that tracks and contributes to these swings, the dominant mood of the industry - up to and including the communicative styles of AI CEOs
— is substantially determined in the tech hothouse of
X, where it's happening and nothing ever happens appear to be the only available positions, or at least the only ones that get engagement. At the same time, fears of job loss, talk of automation, the visibility of obvious externalities like slop content and cheating in school, and emerging worries about data-center construction have given tech's latest investment cycle an unusually powerful ethical and political dimension and turned the potential success or failure of the broadly defined object of AI into mainstream rooting interests.

For one, despite its obsession with forecasts and timelines, this is an industry without a good predictive theory of what its product will soon be good at. It wouldn't be unfair to say that the industry's actual approach to figuring out where things are going is to train a model and see, which gives AI researchers, and their companies, a genuinely unusual relationship to their products: They're building them, sure, but they experience their actual capabilities
- like a sudden aptitude for advanced math after a long stretch of profound innumeracy — as something similar to discoveries. This was clear in the way that Anthropic talked about the cybersecurity capabilities of Mythos as a strange, emergent phenomenon of training general-purpose models. (Contrast that with the development cycle in the nearby and similarly high-stakes semiconductor industry, for example, where swings in sentiment are comparatively mild or at least further apart.)

John Herrman·Jun 26

As I wrote last week, the Trump administration's approach to AI regulation is almost uniquely undesirable. Now it's coming for superficial allies, too nymag.com/intelligence...

John Herrman·Jun 26

mostly this is just another example of zuckerberg's desire to copy or clone every growing company in the industry. but there's a grim, post-media vision of the future embedded in the concept of meta prediction markets, too nymag.com/intelligence...

John Herrman·Apr 28

the hottest trend of 2026 is feeding yourself into a machine you don't understand, and which doesn't understand you nymag.com/intelligence...

It is made clear to you, if you didn't already know, that installing OpenClaw is basically a giant, intentional self-hack: a way to see what AI models now capable of manipulating software tools, most notably in tools like Claude Code and Codex, might be able to get done if given the keys to a civilian-grade email-and-document machine.
You're asked to choose how you'd like to talk to your bot, and because it's the app mentioned in most of the online testimonies, you choose Telegram, which from experience you mostly associate with scams, spam, and, lately, the AI propaganda of the Islamic Revolutionary Guard Corps of Iran.
This is your bot, which is to say your computer, which is to say, well, ChatGPT in a costume with prosthetic hands that can touch your keyboard and move your mouse. It asks you to assign a vibe, give it your time zone, and describe what sort of personality it should have. It all reminds you of setting up a video game. You tell it to be matter-of-fact and concise. You spend a while installing plug-ins that allow OpenClaw to actually take action on your machine, and "Skills," or, essentially, written, repeatable instructions for the agent. In order to set up OpenClaw, an AI chat interface for your computer, you spent a lot of time in the command line, which is how people had to interact with computers before you were born. You note that we've moved from issuing commands to computers to clicking on computers to, now, chatting and haggling with computers, which doesn't sound as obviously empowering as you wish it did, human-agency-wise.

You search X, a political influence project run by Elon Musk where people also talk about AI, and where OpenClaw has for some time been a hot subject. An anonymous account called Big Brain AI - bio: "Learn not to get left behind when AI takes over" — is quoting an interview with OpenClaw's founder. "If you don't navigate them well, if you don't have a vision of what you're going to build, it's still going to be slop," the founder says. Big Brain editorializes in LLM voice: "The agentic trap is what happens when you remove yourself from the process too early." You're failing your own system, in other words.
Skill issue. You're NGMI. In response to Big Brain's post, the head of start-up incubator Y Combinator, who has Sam Altman's old job, chimes in. "This is 100% correct. I experience this 10 hours a day now."
I'm turning my OpenClaw into a Vannevar Bush Memex. It's just going to remember everything I care about and read and it will become my second brain.
Books, writings, research, all of it will be in my personal knowledge wiki and usable for helping me think. pic.twitter.com/lcfxMRnfTq
- Garry Tan (@garrytan) April 7, 2026
You grant that you might just be out of your depth here. But you're also curious if he has people in his life who can check to make sure he's okay.

475

John Herrman·Apr 23

I wrote about the mutual contempt that motivates our polycular AI elite nymag.com/intelligence...

That said, the AI industry resembling a multitrillion-dollar broken-up polycule can't be helping. One influential anonymous X account run by an OpenAI employee - speaking of interesting communications strategies! — worries that it might have some downsides:

tweet from Roon: the ai labs, in competing with each other, are burning huge amounts of the commons on public trust in ai to win minor points against the others. their lobbyists, pr machines, lawsuits. it's the very opposite of what marxist class struggle analysis would tell you
end tweet.

From inside the industry, or even if you spend enough time steaming in the Al hothouses of X or LinkedIn, this map of intra-AI rivalries and vendettas is legible and, for some of these guys, ideologically coherent, rooted in old and substantive disagreements about how to build intelligent machines. From the outside, though, old, festering disagreements about alignment, AI safety, and novel corporate governance structures tend to lose a lot of texture, and the situation can be read, accurately if not necessarily sufficiently, as something simpler and more familiar: Another new industry in the throes of massive expansion, its investors desperate for upside and its principal actors engaged in a ruthless land grab and fight for dominance that feels, to them, like a matter of life or death. That fight is all in pursuit of an outcome that they've explained is (1) probably inevitable and (2) might be pretty bad, and which therefore sounds awfully predatory.

It can be deflating to reimagine the Al boom as a more pedestrian business story with particularly colorful executives expressing contempt for their rivals and making things personal on the way to, say, packaged-beverage dominance. But the maximally dysfunctional dynamics of the pre-takeoff AI industry can also be read as an early, bad sign of how things might play out for everyone else: like they always do, but maybe worse. Here is a visible, prepared, and substantively aligned "small group of elites," including a few of the richest people in the entire world, suggesting that it's time to collectively
"rethink the social contract" and warning that we're about to be "tested as a species," as they're in the process of succumbing completely to a crude, winner-take-all market logic, utterly failing to coordinate among themselves, fighting regulation with lobbyists, getting pissed as hell in public, and opening up a bunch of fronts in a total industrial war for scarce resources — power, compute, water - with immediate and unmitigated externalities. (Granted, comprehensive high-level coordination among them might look like something else people don't particularly love: a cabal.) Individually, to receptive audiences, they can explain how all this happened and rationalize their own roles. To much of the rest of the world, though, they just look like a group of people who worried about building the thing and then couldn't figure out how not to, who cautioned against getting trapped in an arms race and then started one anyway. They see people warning about the speed of change as they step over one another to make it accelerate. They see people urging humility and accusing one another of having God complexes while engaging in a naked struggle for power.

John Herrman·Apr 21

What are we to supposed to make of the AI elite's open, seething hatred for one another? nymag.com/intelligence...

"Been thinking a lot about whether it's possible to stop humanity from developing AL," wrote Sam Altman to Elon Musk in 2015, shortly after Google had acquired DeepMind. Given that it appeared it would happen anyway, he wrote, "it seems like it would be good for someone other than Google to do it first." Musk, who had told Altman that DeepMind was causing him "extreme mental stress" and that, should Google "win," it would be
"really bad news with their one mind to rule the world philosophy," was receptive after recently failing to lure Hassabis to his constellation of companies instead. Soon, they became co-founders of OpenAI. By 2018, a bitter power struggle had led to Musk cutting ties with OpenAI, resulting in years of court battles, some still ongoing. Now, the men tweet openly about how much contempt they have for each other. (Altman on Musk: "I don't think he's, like, a happy person. I do feel for him." Musk on Altman:
"Scam Altman lies as easily as he breathes.") Anthropic's founding was the result of a core group of researchers and employees leaving OpenAI over concerns about its approach to safety, but also over concerns about Altman's character specifically. (Amodei on Altman in 2021: "The problem with OpenAI is Sam himself." In 2026, after OpenAI seized on Anthropic's conflict with the Pentagon, Amodei described Altman as telling "straight up lies" and
"gaslighting.") In 2023, Musk, now in possession of Twitter and a clearer public political identity, finally founded his own firm, xAI, to build a "maximum truth-seeking AI that tries to understand the nature of the universe," but also because Sam Altman was making ChatGPT "woke," which he said could be "deadly." (Elaborating on the theme, and making sure not to miss anyone, Musk posted at Amodei earlier this year: "Your AI hates Whites & Asians, especially Chinese, heterosexuals and men. This is misanthropic and evil.")

John Herrman·Apr 7

The AI industry is about to lose control of its story, and on some level it knows this nymag.com/intelligence...

Whatever the merits of their defenses — and, as Mike Masnick at TechDirt argues, there are more than some online-safety advocates are comfortable admitting — they have lately been, to be blunt, losing the public debate. The legal and rhetorical framework within which Meta and others long built their businesses — the lares protecting platforms from liability aren't perfect, but without them, the internet as we know it couldn't exist — suddenly tells us less about the possible futures of social media than a single unguarded quote from a Los Angeles juror, who told reporters after the verdict, "We wanted them to feel it. We wanted them to realize this was unacceptable." No need to be too specific about what "this" is
- everyone gets the idea.
Here, for people like Altman, is a glimpse of the future:
Nobody wants to hear from social-media companies, while everyone wants something to be done to them. This punishing dynamic will consume their next decade in the form of rolling public-relations crises, lawsuits, regulations, and law, which they will have to deal with in the manner of other entrenched and unpopular industries, with lobbyists and lawyers, rather than as privileged stewards of the economy's most exceptional story.
Today, AI leaders still demand immense amounts of attention, and their predictions - alongside leaks from their companies, cryptic posts from researchers, and viral X essays galore — are largely driving the story. But what about tomorrow? In the weeks before social media's "tobacco moment," there were already signs of angst surfacing in the AI industry about its loss of narrative control. Anthropic had publicly clashed with a right-wing Department of War that wanted control over its technology and for the company's founder to sit down and shut up. Meanwhile, the most popular left-wing politicians in the country were suddenly advocating for an extreme AI
"pause" that would halt development in its tracks, citing Al leaders' own warnings about its risk

OpenAI's "Industrial policy for the Intelligence Age," a set of proposals mingling plans to coordinate against AI security threats with outlines of redistributive government programs, is a sensible (if vague) document made incredibly strange by its source: Maybe we see evidence here of a different sort of AI bubble, one in which Sam Altman seems to believe that America is waiting on people like him to start a conversation about renegotiating the social contract, rather than a country spring-loaded for a tech backlash to end all tech backlashes
- the same bubble where, last year, as few will recall, Zuckerberg, trying out a new identity as an AI CEO, wrote a manifesto about how "superintelligence is now in sight."
But in OpenAl's substantial and much quieter lobbying efforts around, among many other things, "child safety" and data-center deregulation — and perhaps in its nascent new media strategy — there's evidence of a more paranoid and ruthless approach to managing the story. AI won't just have a "big tobacco" moment, a point at which years of misleading marketing, personal angst, and visible harms gradually build to a moment of reckoning. It's likely to have something more intense, and soon, arriving ahead of, not after, its diffusion through society and the economy. The last story the Al firms will have been in control of will be the one where they said they were about to change everything. It helped them raise a lot of money to build a lot of data centers and train better models. What happens next might not be up to them at all.

John Herrman·Mar 19

Two reasons the U.S. is losing the slopaganda war: 1) it makes the most formidable cultural exporter in the world look like a cheap spam operation 2) the most popular AI videos seem to tap into fantasies of American failure, and slop merchants follow the market nymag.com/intelligence...

On one side, you've got the superpowers that initiated the war, states with near-infinite resources to marshal for their military campaigns and accompanying messaging strategies. On the other, you've got, well, Iran. Besides the contrast in subject - upbeat, chest-thumping aggression versus a call to avenge the deaths of schoolchildren - the use of cheap, fast, and widely available AI video generation has a powerful flattening effect. If you've played with these AI tools, you'll easily be able to imagine fragments of the sorts of prompts that were fed into software like Kling, Veo, Runway, Grok, or Seedance: the aftermath of an attack on a school in Iran in the style of LEGO animation; LEGO man Donald Trump reading a notebook of the Epstein files, Satan is there, realistic, cinematic; militant bowling pins approach holding sign reading "We won't stop making nuclear weapons," Dreamworks style, smirking faces, angry eyes; bowling-alley style animation of a "strike" using military jets.
Nothing here took much effort, and the outputs end up inhabiting the same narrow aesthetic zone, easily and cheaply obtainable, associated with widely available AI tools that have rapidly commoditized such videos. For Iran, Al video propaganda might represent a slight increase in capability, a way to produce higher volumes of passable legible, and persuasive material for a wider
range of audiences. For the United States, this sort of automation provides speed and flexibility, sure, but also draws its communications strategy down into a far more symmetric fight than the war itself, where everyone has the same weapons (i.e., AI video-generation tools that any curious teenager could use at this level) and the battlefield - where the possible audiences for such videos are to be found - is the entire internet, where such videos overlap, stylistically, with spam

Into the void of information coming from the Middle East flowed an endless supply of videos suggesting unlikely, extreme, and enticing outcomes: swarms of missiles obliterating Dubai and Tel Aviv; societal breakdown in the United States; American warships exploding and jets falling out of the sky. (The AI slop war is producing blowback, too: Last week, Benjamin Netanyahu was compelled to make a proof-of-life video after social media users spread claims that, during a recent address, the Prime Minister had too many fingers
— a telltale sign, social media users said, that it had been generated by AI.) If you're a professional slop merchant
- the Trump Hormuz video above originated from a meme account allegedly based in Switzerland - you know that rendering stran
re and unbelievable scenarios
is a good way to get views. If you've started working around the war, you've probably also noticed that audiences seem more interested in stories representing U.S. and Israeli failure and overreach than in content that suggests its campaign is going well. That may be true not just abroad but also here in the U.S., where the young war is unpopular to an unprecedented degree. (One notable exception is the popularity of content posted in support Reza Pahlavi, the son of the deposed shah of Iran, whose followers in the diaspora seem receptive to fantastical AI content teasing his glorious return.) To paraphrase Field Marshal Montgomery, General MacArthur, and Vizzini, never get involved in a slop war on social media.

This dynamic ends up creating an uncomfortable alignment for the mostly American social-media companies, stuck between the incentives created by their platforms and the military aims of the country against which their home country has declared war. To take a purely sociopathic, market-centric view of what's happening — one that I describe here mainly because it's pretty close to how social-media CEOs talk about their platforms and users in the abstract — Al is helping to meet existing demand for war-related content online.
Perhaps it's even inducing more demand, as fake, misleading, or trivializing video makes the already dehumanizing and trivializing treatment of wars on social media as a fleeting content category that much more compelling and/or worse. (Congrats to AI!)

John Herrman·Mar 16

Arguments about AI aren't just about the future — they're trapped it in nymag.com/intelligence...

Communication between characters as constitutionally different as Amodei and Hegseth was always going to be strained, but the breakdown was about more than "vibes and personalities." It was about an AI company and the Pentagon trying to account for disparate and extreme visions of tomorrow in clauses and provisions written for today. It was a collision between two movements that believe, in different and inherently incompatible ways, that they might be on the cusp of achieving absolute power, either for themselves or for the systems they're helping to build, and that current concessions will extrapolate into total failure.

Of course, Hegseth isn't really thinking about the singularity or ways in which today's Claude is still in ways unreliable (if AI supremacy is a core part of your national security philosophy, you don't hobble your country's leading lab over a procurement dispute). He's thinking about consolidating power, and his conception of the potential of AI is subordinate to that, and him, rather than a future superintelligence. This is why the situation escalated the way it did, with the government not just walking away but attempting to punish the company for asserting itself at all. If not for this retaliation, the government's narrow defense — that if a private company wants to contract with the military, it shouldn't expect to be able to micromanage how its tools are deployed and should reasonably expect to be implicated directly or through reputation if, for example, the military were to then bomb a school - would make sense on its own terms. But the Trump administration's belief in its right to total power, expressed by and in the belligerent figure of Hegseth, is central to both its "vision," such as it is, and to understanding the way it conducts itself here and elsewhere.

View all posts on Bluesky

Menu

John Herrman

Top posts

Latest posts