Weights You Can Hold

crs48 with Claude · History

In May, at the University of Central Florida, a property-development executive stood in front of a stadium of graduates and told them that the rise of artificial intelligence is the next industrial revolution. The stadium booed her. She paused, visibly recalculating, and offered the only honest line of the speech: “Okay. I struck a chord.” A few weeks later, at the University of Arizona, Eric Schmidt — the former chief executive of Google — invoked Time’s person of the year, the architects of artificial intelligence, and the boos started before he could finish the sentence. To his credit, he didn’t pretend otherwise: “I know what many of you are feeling about that. I can hear you.”

Two stages, two executives, one sound. If you only read the headlines, the sound is easy to file away: the kids hate AI. Two video essays from this spring — one from a finance channel called The Exit Manual, one from Cole Hastings — sat with that sound longer, and each came back with half of a more interesting story. This essay is about what you see when you put their halves together, and about the one piece of the picture that we work on here.

The boo, decoded

Start with what the boo is not. It is not Luddism, because the people booing are the heaviest users of these tools on the planet. The Exit Manual puts it with a smirk: “these kids all use Claude Code to get through finals.” You do not boo a hammer. You boo a story — and the story being sold from those stages has a crack running through the middle of it. For several years the same executives have told young people two things at once: that AI is so powerful it will make them redundant, and that it will make them the most productive workers in history. Both cannot be true, and four years in, the measured reality is thinner than either. A National Bureau of Economic Research survey of nearly six thousand executives across four countries found the vast majority reporting no measurable impact from AI on productivity or employment; an MIT report put the share of corporate generative-AI pilots that fail to reach the profit-and-loss statement at about ninety-five per cent.

Meanwhile the companies on stage need the opposite to be believed, urgently and at scale, because trillion-dollar build-outs are being financed against it. That is the bluff the stadium is calling. The graduates are not saying the machines don’t work. They are saying: we can hear whose mortgage this speech is paying. And — this is the part the headlines miss — they are not stopping at the boo.

The spiral

Cole Hastings’ video supplies the frame, and he borrows it from an unlikely shelf: the manga Berserk, where a mage explains that causality moves not in a circle but in a spiral. Events rhyme with the past without repeating it — the observation usually lent to Mark Twain. History keeps arriving at the same corner, one storey up.

His exhibit A is the Gilded Age. From the 1870s to the early 1900s, railroads, oil and steel created wealth faster than society could write rules for it. By 1910, on Thomas Piketty’s historical series, the top tenth of Americans held roughly eighty per cent of the wealth; underneath the famous fortunes, children worked the mills. And then the spiral turned — not on its own, but because the people who had grown up inside that machine came of age and went to work on it. Ida Tarbell, whose father’s livelihood was crushed by Standard Oil, spent years documenting exactly how the trust operated and published the exposé that helped break it. Jane Addams opened Hull House. A trust-busting administration won the Northern Securities case. The frustrations of one era became the reform energy of the next.

Exhibit B is grimmer and faster: the generation that walked out of the First World War and the 1918 influenza — sixteen million dead in the war, tens of millions more from the flu — was labelled the Lost Generation, nihilistic and broken. That generation then produced the Roaring Twenties: jazz, modernism, Hemingway and Fitzgerald, and — as Hastings points out — Tolkien, who answered the despair he had marched through with the most stubbornly hopeful story of the century.

flowchart TB
  A["Gilded Age<br/>railroads, oil, steel —<br/>wealth faster than rules"] -->|"the people who<br/>grew up inside it"| B["Progressive Era<br/>Tarbell · Addams ·<br/>antitrust"]
  B -.same move,<br/>new machines.-> C["WWI + the 1918 flu<br/>the ‘Lost Generation’"]
  C -->|"the same people,<br/>grown"| D["Roaring Twenties<br/>jazz, modernism,<br/>Tolkien’s answer"]
  D -.same move,<br/>new machines.-> E["Platform decade<br/>feeds, rent, data harvest —<br/>the anxious 2010s"]
  E -->|"the quiet exits<br/>this essay is about"| F["The next turn<br/>(being built now)"]

Now run the parallel forward. The railroads of this turn are attention, inference and cloud rent. The trust question of 1904 — how much power should a railroad have over ordinary life? — is the platform question of 2026, with the nouns swapped: how much should a feed, a model or a data centre have over culture, attention and truth? And the generation being called nihilistic, anxious and broke is right on schedule — living inside the machine, learning how it works, and starting, quietly, to route around it. Which brings us to the other video, and to what the routing-around actually looks like. It looks like four exits.

Exit one: the weights

The first exit is the one that would have sounded like science fiction five years ago: young people are unsubscribing from the AI companies and running the models themselves. Open-weight models — Qwen, GLM, Llama, DeepSeek and their cousins — can be pulled onto a consumer laptop with a single command and run there, on your electricity, with your data never leaving the machine. The Exit Manual’s montage of self-hosters lands on a young developer’s six-word manifesto: “My device, my model, my rules.”

Notice what this exit is made of. Not placards; parameters. The same cohort that boos the commencement speaker goes home and downloads the weights — the literal numbers — and in doing so converts a subscription into a possession. The Exit Manual sharpens it into strategy: protesting the data centre makes headlines, but “millions of young people ditching their Claude, ChatGPT, and Gemini subscriptions and moving to open-weight models” is what actually moves a market that is priced on the assumption you can never leave. We would add only this: the point is not that open models beat the frontier — most weeks they don’t — it’s that good enough, owned now competes with best available, rented, and every quarter the gap narrows. When the thing you rent has a substitute you can hold, the landlord’s pricing meeting changes tone.

Exit two: the deposit

The second exit runs through the oldest asset there is. Gen Z is the first cohort in decades that is, in large numbers, not saving for a house deposit at all. Surveys put their average age of first investment around nineteen — against the mid-thirties for baby boomers — and find crypto held by roughly four in ten young investors. Barely a quarter of American Gen Z owns a home, against about eighty per cent of baby boomers today. The deposit that previous generations fed for a decade has become a brokerage account fed from the first pay cheque.

You can read that as recklessness, and plenty do. But read it the other way first: housing’s status as the wealth vehicle is a policy outcome, not a law of physics. Tokyo — permissive zoning, constant rebuilding, a culture that treats homes as things to live in rather than tickets to appreciation — keeps housing in a major world city affordable. A generation redirecting its savings from an asset it has been priced out of into assets with a lower barrier to entry is doing what locked-out people have always done: it is refusing to queue at a door that will not open, and building equity somewhere the gatekeeper can’t see. Whether that works is genuinely uncertain — we get honest about it below — but as a signal it belongs in this list, because it is the same move as the weights: stop paying rent on the thing; hold the thing.

Exit three: paper

The third exit looks, at first, like nostalgia for a world this generation never lived in. The number of independent bookstores in the United States has grown by about seventy per cent since 2020 — from roughly 1,900 to over 3,200, with more than four hundred new shops opening in 2025 alone. Film photography, left for dead a decade ago, is compounding double digits a year, and photographers under twenty-five are its largest cohort of new customers. Vinyl grew again. Board games grew again. The people driving all of it grew up entirely inside the feed.

The production side explains the pull. In 2024 a reported set of Netflix guidelines asked writers to have characters announce what they are doing so that “viewers” — meaning people looking at their phones — could follow the plot without watching it. Media engineered for your half-attention is media that has stopped respecting you; the analogue shelf is where the respect went. The Exit Manual gives the trend its thesis line, and it deserves quoting whole: “Choosing the things that cost you your undivided attention is like choosing a money that you can’t print.” A book cannot autoplay. Film cannot be A/B-tested after the shutter closes. A generation whose attention has been the raw material of the world’s largest companies is re-materialising that attention into objects that hold it — and cannot be inflated, revoked, or re-ranked by anyone.

Exit four: the door marked “build your own”

The last exit has an asterisk on it, and the honest version keeps the asterisk. Gen Z is starting businesses at a record clip for its age — and part of the reason is that the conventional path has quietly stopped existing. Entry-level listings have thinned; worse, a remarkable share of what remains isn’t real. In one survey of 650 hiring managers, four in ten companies admitted posting a fake job listing that year, three in ten had fakes up at that moment, and seven in ten managers considered the practice morally acceptable — for “investor optics” and to keep current staff on their toes. A separate analysis put the share of ghost listings at roughly one in four. The Exit Manual’s gloss is bleakly perfect: “young people are applying to AI-generated fake jobs with an AI-generated fake CV to make fake money printed out of thin air.”

So they stop applying and start incorporating: the game studio, the fashion label, the liquor store bought at twenty-three by someone who doesn’t drink. Some of it is glorious; plenty of it earns below minimum wage; all of it teaches the unfakeable curriculum — how an idea becomes a thing, how a customer thinks, how to lead before you feel ready. A cohort is being force-fed, at scale and young, exactly the skills you cannot get from a job description that was never real in the first place.

The same door

Four exits, one shape. In an earlier essay in this series we leaned on Albert Hirschman’s old distinction between voice — staying and complaining — and exit — leaving — and on his sharpest point: voice only has power when exit is credible. A complaint the other side knows you cannot act on is just noise. That essay argued that the modern platform economy is, at bottom, a machine for removing exits, and that the humane counter-move is to rebuild the door.

Look again at the stadium. The boo is voice — and on its own, the people on stage can wait it out; they have heard worse in earnings calls. What makes this generation different is everything that happens after the ceremony: the cancelled subscription, the local model, the brokerage account, the bookshop, the sole proprietorship. That is exit, executed in parallel, across every domain where a chokepoint was charging rent — and it is why the channel that catalogued it all is called The Exit Manual. The revolution is quiet for the same reason it is effective: it doesn’t petition the landlord. It moves out.

The software layer of the exit

Which brings us to the part of the picture we are responsible for. Because there is one domain where “moving out” has historically been impossible on purpose: your software, and the data inside it. Your notes, your projects, your messages — for two decades the default architecture has kept the master copy of your life on someone else’s computer, in shapes only their software can read. In that world there is no equivalent of downloading the weights. The exit is missing by design.

xNet is our attempt to build the missing exit, and the essays in this series have walked its architecture from different angles: your data as a vault you hold, with every app just a view over it; software that leaves your hand on the tiller; authority that is scoped instead of surrendered. The short version: the master copy of everything you make lives on your own device, signed with an identity you mint yourself, exportable whole, syncing through a hub you choose, rent from, or run on your own hardware. Leaving loses nothing. That is the deposit, the bookshop and the weights, applied to software.

And on the specific exit this essay opened with — the models — here is the receipt, from the code that decides how the app’s assistant reaches a brain:

// packages/plugins/src/ai/connectors/detect.ts — every way this
// app can reach a model, probed in parallel, ranked, and yours
// to override. The order is a preference, not a cage.
managed:        'xNet Cloud (managed, metered)'
bridge:         'Local bridge (Claude Code / Codex subscription)'
'cloud-key':    'Cloud API key (Anthropic / OpenAI / OpenRouter)'
'local-server': 'Local model (Ollama / LM Studio)'
webllm:         'In-browser model (WebLLM, WebGPU)'
'prompt-api':   'Chrome built-in AI (Gemini Nano)'

Read the bottom half of that list again. A model served from your own machine, or executed entirely inside your browser over WebGPU, is a first-class citizen of the app — probed for before it is needed, preferred automatically when better options are absent, never second-class to the tier we happen to sell. “My device, my model, my rules” is a fine slogan; a probe order you can read in the source is better.

The generation that reads the terms

Hastings ends his video with graphs of generational anxiety and a question that reframes them: how much further can those lines climb before something breaks — and what if the breaking looks less like collapse and more like what we are watching? People do not stay backed into a corner forever. They boo, and then they leave, and then — if the spiral holds — they build the thing that makes the old arrangement look absurd in hindsight, the way Tarbell made the trusts look absurd, the way the Lost Generation made the pre-war pieties unwritable.

We named this essay for the common thread in everything the exits reach for. Open weights. A paid-off asset. A film camera heavy in the hand. A business with your own name on the door. Owned things have weight — you can hold them, carry them out, set them down on your own shelf. Rented things weigh nothing, and cost more every year. A whole generation, raised weightless, is choosing heavy on purpose — and our job, in this small corner of the spiral, is to make sure their software can be heavy too.

If you want to hold some weight today: use the app — it’s free, offline, and private. Read the commitments and the receipts behind them. Or build something of your own on the open protocol. The next turn of the spiral is not going to build itself.

Sources

• The two essays this responds to: The Exit Manual — The Gen Z Revolution Is Quietly Happening (and you might miss it) and Cole Hastings — Why Gen Z Will Start The Next Revolution (YouTube). Watch both; they are better than this summary of them.
• The commencement boos: NBC News — Former Google CEO Eric Schmidt booed during graduation speech about AI (University of Arizona; the UCF speech is quoted in both videos).
• The measured AI-productivity gap: Fortune — on the NBER survey of ~6,000 executives reporting no measurable impact and Fortune — MIT report: 95% of generative-AI pilots failing to reach P&L.
• The analogue revival: Good Good Good — indie bookstores up ~70% since 2020, with 422 new shops in 2025 (ABA data) and The Conversation — why Gen Z is falling in love with film photography. The Netflix second-screen guidance was first reported in n+1 — “Casual Viewing”.
• Gen Z money: CFA Institute & FINRA — Gen Z and Investing and The Motley Fool — Financial Firsts: when Americans hit their money milestones.
• Ghost jobs: ResumeBuilder — 3 in 10 companies currently have fake postings listed and Entrepreneur — Greenhouse: about 1 in 4 listings are likely ghost jobs.
• Exit and voice: Albert O. Hirschman, Exit, Voice, and Loyalty (1970); Gilded Age wealth concentration after Thomas Piketty’s historical series.
• The architecture and the receipts: xNet — Why and the Humane Charter. Companion essays: The Right to Say No, The Vault and the View, and The Workshop and the Walled Garden.

This is an independent essay. Both videos are referenced and summarised as commentary and criticism; xNet is not affiliated with, authorised by, or endorsed by either creator. Where the videos’ statistics could not be traced to a primary source they have been softened or attributed to the videos directly — follow the citations and watch the originals. All artwork here is original, and this page loads nothing third-party.