AI Content and the War for Your Attention
Summary
Hayes’s core warning is that generative AI turns the economics of spam into the economics of culture: an automated content operation can make 1,000 videos a day, repeat an idea 15 times, and throw them at recommendation systems until something takes off. That brute-force supply could widen the gap between “what you will pay attention to” and “what you want to pay attention to,” leaving platforms to choose between filtering AI pollution and maximizing immediate engagement.
For feed businesses, attention acquisition and retention are different problems. Antonio García Martínez notes that firing a gun in a room captures attention but does not necessarily hold it; his own TikTok binges end with deletion of the app. Hayes and Antonio retain some faith that people will rebel against slop, while Erik Torenberg frames three possibilities: feeds drown in pollution, AI creates a “pure attentional drug,” or users continue preferring humans.
The democratization of fame is pushing social activity toward smaller, governed spaces. Antonio’s update to Warhol—“we’re all going to be famous to 15 people”—captures a world where everyone behaves as if observed and any outburst can go viral. Group chats can restore bilateral attention and accountability, but monetization remains awkward: Friend.tech tokenized access to Antonio’s channel, while Hayes argues that overfinancializing social spaces can destroy their social character.
The internet is fragmenting shared attention while simultaneously manufacturing geographically borderless omnicultures. Hayes misses collective objects such as the Super Bowl even as niche communities let isolated people “find my people”; meanwhile, Gen Z shares the same algorithmic slang across countries, and some rural and urban demographic cohorts increasingly resemble their peers elsewhere. The result is neither old mass culture nor local pluralism.
Adtech’s personalized-data story remains far stronger than its average creative output or response rate. Antonio says that on something like Instagram, “at least 97%” of viewers remain indifferent, while fragmented third-party data and transient cookies limit targeting; Facebook’s advantage is its stable first-party identity. Hayes’s unresolved question is why TikTok serves infomercial-like DTC products rather than Progressive, GM, or other high-LTV advertisers if its targeting is so powerful.
Antonio expects conversational agents to move upstream and become the gateway to consumer transactions. His concrete interface is a request to arrange a week in Brittany—Airbnb, flight, and a budget below $3,000—followed by one button that executes everything. Hayes agrees that this is useful, but stresses that utility does not answer what AI will do with the much larger block of screen time spent consuming rather than accomplishing tasks.
The ultimate growth frontier is control of more hours, while the information and energy underneath AI remain unresolved constraints. Hayes reads Apple Vision Pro as hardware for monetizing every waking second; Torenberg identifies 16 waking hours, and Antonio jokes that reaching dreams requires Neuralink. Yet Hayes argues that AI-led growth may collapse the open web that supported Google’s first iteration, while the quieter prospect of near-zero marginal energy from solar—potentially alongside nuclear—could prove just as consequential as near-free intelligence.
Deep dive
1. AI makes cultural spam a scale problem
Antonio opens with a pollution thesis: wherever attention pools and money follows, cheap automation produces spam. Junk mail, calls, texts, and email already followed that path; generative AI now brings the same economics to social content, raising the possibility that feeds eventually feel like overwhelmed inboxes.
Hayes’s mechanism is quantity, not necessarily quality. Online success rewards total hits more than batting average, and creators such as MrBeast discuss “always be posting”; a machine can generate 1,000 videos daily or repeat the same thing 15 times, making brute-force experimentation available at a scale no human creator can match.
Hayes’s answer is deliberately mixed: much of today’s AI output feels “empty and sterile,” but creative humans using the tools correctly could gain “superpowers.” His best specimen is the genre assembled from proven ingredients—religious themes and babies—including a video of Jesus and a baby singing “Bless the Lord, O my soul” while Jesus holds a signature AI-generated disembodied foot.
Antonio’s pushback is that social networks already rewarded artificial selves: LinkedIn feels staged, while Instagram users maintain public identities and “finstas.” Hayes accepts the continuity but argues that automation changes the quantity; he is not certain the flood arrives, only that its early appearance makes the platform consequences worth asking about.
2. Attention acquisition is easier than retention
Antonio separates two marketing jobs that slop discussions often merge: “acquisition and retention are different things.” Someone firing a gun in a room acquires attention instantly, but even the armed person may struggle to maintain it; recommendation systems can manufacture a glance without building an experience users continue choosing.
His own TikTok behavior supplies the counterevidence to faith in user agency: he emerges after two hours wondering “what the hell just happened,” then deletes the app to prevent a repeat. Hayes connects that loss of volition to his book’s central metaphor—Odysseus bound against the sirens’ call.
Both retain some faith in an “irreducibly human” preference that could provoke rebellion against slop. But Hayes thinks platforms may have to manage AI output as actively as Twitter once managed difficult content-moderation questions, balancing engagement with a tolerable user experience that attracts advertisers.
Erik later lays out three scenarios: brute-force pollution overwhelms feeds; synthetic media becomes a surreal “pure attentional drug,” optimized by a neural-network hive mind without much meaning; or audiences continue opting for humans and AI changes less than expected. The episode never pretends to know which equilibrium wins.
3. Constant visibility drives audiences into smaller rooms
Antonio’s line from Chaos Monkeys updates Warhol: “In the future, we’re not all going to be famous for 15 minutes. We’re all going to be famous to 15 people.” Tourists queue for self-images, people carry themselves “like they are a Kardashian,” and young workers report feeling observed everywhere.
Hayes confirms that even relative, limited fame modifies conduct: yelling at a dangerous driver could be recorded, recognized, and attached to his identity. Hayes also sees a possible prosocial effect—people may restrain public misconduct—but emphasizes the psychological warping of persistent social attention.
The group-chat model offers a way to reclaim privacy and community. Hayes describes normal relationships as “bilateral exchanges of social attention,” unlike unilateral audience relationships; WhatsApp is a valuable commercial example, while Signal offers a nonprofit model.
Antonio’s San Juan Islands analogy explains why bounded groups can self-regulate: in a small community everyone knows everyone, and misconduct carries reputational cost. A group-chat administrator can eject someone, while members with shared professional ties cannot behave badly without consequences—the technological recreation of local accountability.
4. Useful social technology does not guarantee a great business
Hayes warns against conflating useful technology with lucrative technology. Penicillin and solar power can be enormously useful without producing Rockefeller-scale fortunes; FanDuel and DraftKings can deploy excellent, highly lucrative technology without providing remotely comparable social value.
That distinction complicates the retreat into private groups. Meta owns the highly valuable WhatsApp, while Signal offers a nonprofit model; Hayes asks whether revenue incentives fundamentally conflict with the technologies best suited to ordinary human relationships.
Antonio supplies the crypto experiment: Telegram groups may charge admission, while Friend.tech tokenized access and let users buy an “AGM token” to join his channel. Hayes’s conclusion cuts against crypto’s reflex to turn everything into money: “if you overfinancialize things, you lose the social side of it.”
5. Fragmentation destroys common objects but creates omnicultures
Hayes defines a public as people “paying attention to the same thing together.” After an era when mass broadcasting could blast messages to populations at scale and move the needle, the problem has inverted: he now feels “weirdly old-man nostalgic” for the Super Bowl or even Love Island because football is among the last objects broadly watched at once.
The literary equivalent has disappeared too. Hayes and Antonio nominate Jonathan Franzen’s The Corrections as perhaps the last book everyone felt required to read; Infinite Jest became the book people claimed to read. Yet the old mainstream was also stifling, and the internet’s great promise was, “I finally found my people.”
Algorithmic social media can nevertheless produce shared culture by other means. Hayes jokes that “the algorithm” is discussed like a deity, despite sometimes being dumb enough to serve 15 copies of whatever he just viewed; at other moments it feels magical, as when a feed of sandwiches appeared to infer that he was high.
Shared Gen Z “brain-rot” slang now crosses national boundaries, while geographical cultures weaken: rural Minnesotans and rural Mississippians of similar demographics increasingly vote alike, and educated urban neighborhoods converge on distressed wood, exposed bulbs, and interchangeable farm-to-table restaurants. Individuation can help an isolated trans teenager find peers, yet local distinctions also give way to shared demographic omnicultures.
6. Adtech knows identities better than it persuades people
Hayes poses adtech’s unresolved contradiction: platforms possess personalized information no traditional advertiser could access, yet internet advertising remains remarkably “bad, schlocky, and ineffective.” Antonio agrees that the process is still statistical; on something like Instagram, “at least 97%” of people who view an ad simply do not engage.
The operational limits matter. Purchased data is often less accurate than advertised, fragmented across silos, and subject to drift; website cookies have become so transient that they are “basically meaningless.” Facebook’s advantage is its first-party relationship and stable knowledge that the same Chris or Antonio keeps returning.
Hayes’s best challenge is TikTok’s ad mix: why does it resemble a predawn infomercial instead of serving Progressive, GM, or other lifetime-value brands to young consumers? Antonio thinks TikTok and Whatnot work for niche DTC purchases. Hayes notes that broadcast football offers a scale TikTok cannot easily replicate for a Ford F-150, while Antonio’s honest answer on the larger discrepancy is: “I’m not sure.”
7. Agents capture commerce while hardware captures more hours
Antonio’s next platform is conversational AI: ask it to arrange a week in Brittany, book the flight and Airbnb, hold the total below $3,000, then present one execution button. He would “bet large portions” of his net worth that this becomes the gateway through which AI moves upstream across the consumer experience.
Hayes accepts the browser-like agent interface for useful work but distinguishes it from leisure consumption. Booking travel explains only a fraction of a screen-time report; the discussion leaves open what AI does to the hours spent scrolling, watching, and surrendering attention rather than completing an explicit task.
Apple Vision Pro reveals the commercial logic of immersive hardware: once phones occupy as much time as possible and nearly everyone owns one, growth requires expanding the commodity frontier. A device worn every waking second opens new “deep-sea reserves” of attention; Torenberg identifies 16 waking hours, while Martínez jokes that monetizing dreams requires Neuralink.
8. Endless growth can consume the information system beneath AI
Hayes is pro-growth but sees a hard tension between endless expansion and human boundaries. Google once offered a transformative service and “printed money,” yet that was not enough; AI growth now coincides with the collapse of the open web that supported Google’s first iteration.
For reporting, Hayes can use AI only with web search enabled and every claim cited, because he must verify that a person reported the information and that names and sourcing are correct. His unresolved question is who maintains the useful information layer once the growth and revenue model behind the open web collapses.
Antonio invokes Edward Abbey’s line that “growth for the sake of growth is the ideology of the cancer cell.” The episode pairs Abundance’s case that the left should not reject markets and growth with The Technological Republic’s reminder that markets cannot identify the ultimate good ends of social life.
Hayes closes by contrasting AI’s capex and attention with a less glamorous solar revolution that might push marginal energy costs toward zero; Antonio adds nuclear to the portfolio. Their combined possibility—“marginally free intelligence and marginally free energy”—could be extraordinary, especially because AI’s enormous compute demand makes the two transformations inseparable.