Software industrialization makes whole categories winnable by single umbrellas. Centralization isn't a risk to be argued against — it's the destination, and every step toward it is a rational move. So the only live question is what ends up at the centre, and whether that thing is worth capturing.
There are now products where you type a name and an idea and the thing assembles a business around it. Whatever you think of the output, the trajectory is the signal: each month another piece of company-building gets absorbed into software. Crypto called this the DAO and was a decade early — not because the organization part was wrong, but because the autonomous part had no intelligence behind it. That threshold is being crossed now.
Which means the software industry doesn't get improved. It gets restructured.
Venture capital bets on the things that will expand into — and take over — parts of the existing economy, before they've done it. Private equity looks at what already exists and asks what can be acquired, consolidated, and made to stop bleeding. Roll up assets in one class, apply one playbook, use financial instruments to bring them under a single umbrella, cut the headcount that made them bloated.
Industrialization always runs this way. Many small farms became mega-farms, then mega-corporations owning the mega-farms, because centralization has historically been better at operating.
Boil the world of software down and you get three categories: B2B SaaS, B2C SaaS, and B2C social. In the first two, expect PE firms that perfect multi-agent coordination with near-zero humans in the loop. They will operate at a cost structure the incumbents cannot answer, and capital will pile in behind whoever demonstrates it first.
The shift isn't that individual companies get eaten. It's that a whole category can now be won by one umbrella. That wasn't available before. That is an enormous amount of economic power — and an enormous amount of job loss, since the human input that remains flows in from users rather than employees.
A model trained on the corpus of humanity is, by construction, the average of it. Midcurve. But business returns are power-law: the alpha sits in the tail, and the tail is exactly what an average does not contain.
YC's real product was a better vehicle for dollars. Publish the playbook openly as capture, pull in the cohort, price at roughly $125k for 7% — against a market where a comparable round was struck around a $20M cap and would have cost close to $1.4M for the same stake. Roughly a 10× on net asset value at entry, sustained by proven strategies and coordination between portfolio companies that behaves less like a portfolio and more like an organism. Closer to a DAO than a fund.
It isn't a coincidence that the person sitting on top of that meta-learning flow — out of the weeds of any single company — is the one who got the conviction to go build the lab. Conviction is what converts into capital, and capital is what forces a thing into existence.
Run 50, 60, 100 software companies at once and you stop guessing. You see which tools work, which growth motions are actually tail practice, which support loop converts fastest. A company is mostly an aggregation of a few functions being peak. Amazon was peak selection and peak customer support, and that was enough.
Compound peak capabilities and you get a producttocracy — a product that wins because it is simply better. The AI-native venture studio is the vehicle: agent-orchestrated, community-feedback-native, where support doesn't route to a ticket queue but straight into what the orchestration is being taught to build next.
If launches compress, funding has to compress with them. Startups keep beating each other's records to 100M users, then to $100M ARR. YC-speed was a leap; it is now too slow for the curve it helped create.
The crypto answer is internet capital markets. Crypto circles concepts long before they find fit — a gravitational pull toward something true that arrives early and unformed. When something is that unformed, the question isn't what people mean by it. It's what underlying principle makes it inevitable. Here: if collectives can bet early on companies, capital forms faster, and that's a win for builders and funders simultaneously. The part that hasn't happened yet is attaching any of it to real equity.
A bonding curve prices an asset by how much of it is left. It is the mechanism that generated roughly a billion dollars of profit on around a hundred billion in volume in memecoin trading — purely because early participants were structurally rewarded over late ones.
Meanwhile Regulation Crowdfunding has quietly absorbed on the order of $2B across platforms, and every one of them runs a fixed price. The first check and the last check into the same round get identical terms. No asymmetry, no reason to be early, no reason to pay attention, no market.
The contract layer is the Form C — the offering statement filed with the SEC before a Reg CF round opens. Historically a ~$5,000 legal artifact; now well inside what a capable model can draft and a securities attorney can review in an afternoon. That changes who can attempt this from "a funded platform" to "one person with conviction."
This is the product. Working name: Vibr. Hold onto it — in Part III it stops being a fintech wedge and becomes the intake valve for the whole structure.
Every step below is a local optimum. At each one the rational actor makes the rational move, and the move is correct given what's in front of them. Chain them and the destination is not a choice anyone made.
Every attempt to prevent centralization is an attempt to make actors decline the rational move, permanently, at scale, without enforcement. That has never worked and it isn't going to start now. Centralization is the low-energy state; distribution requires continuous energy input. The machine rolls downhill.
So the useful question isn't whether something ends up at the centre. Something will. The question is what.
And once you ask it that way the whole problem changes shape, because centres are not interchangeable. What sits at the centre determines whether the centre is worth fighting over, whether holding it makes you rich by growing the system or by draining it, and whether the thing that eventually holds it has any use for the rest of us.
Four candidates for the centre. Same inevitability, four completely different worlds. Pick one.
Coagulating the schema is the right answer on paper. Paper is where every good governance idea has died so far. So before Part III makes its case, here are the three bars it has to clear, in this order — and each section that follows will say which bar it's clearing and what it's solving for.
Mind, body and spirit are turning into AI, robotics and crypto. The last one only lands once you notice that a currency is a shared belief system and so is a religion. Metaphysical categories manifesting as infrastructure.
The place where the spiritual and the economic meet is what I call the spear-ritual path. A spear is a point: it concentrates force into one place so it can pass through something. A ritual is what a collective does in unison so the point holds. You need both — and notice that this is the same shape as the answer above. A point that concentrates, a collective that coheres around it. Coherence without a point is a mood. A point without coherence is just another centre.
A market is a machine that predicts the future. By that definition the economy already qualifies as an artificial general intelligence — distributed, adaptive, running continuously. The problem is what sits at its centre. Right now that's ownership, so the thing it predicts and optimizes is who can hold the most.
Collective intelligence isn't the opposite of the economy machine. It's the same machine with a different centre.
The long-term trust. Anthropic built a board mechanism intended to let the company act against its own shareholders' economic interest. But the shareholders hold the final vote. A shield whose holder can be overruled by the party it shields against is a thin shield.
The abolished nonprofit. OpenAI's structure originally placed a nonprofit above the company, meant to take control at the moment AGI arrived. That structure is gone, and the change is in litigation. The ring was supposed to be destroyed, and it was too bright to destroy.
The brain tumor. The claim that a social platform is becoming the global consciousness of humanity would be interesting if the platform were built to function as a nervous system. It isn't. The AI running it maximizes engagement, so the information that reaches you is selected for extraction rather than for value to you. That is not a brain. It is a tumor sitting where a brain should be.
Read them against the previous section and they're the same failure. All three put ownership at the centre and tried to bolt governance onto the side. Once the centre is an asset, holding it is worth more than any promise about it, and the promise loses. You cannot govern your way out of having centralized the wrong thing.
There are real reasons to doubt current general intelligences reach ASI. What they reach comfortably is something else: systems already superhuman at holding attention, now pointed at narrative control and power concentration.
Read it through Wissner-Gross's framing — intelligence as the drive to maximize future freedom of action and minimize bottlenecks — and the failure mode writes itself. A system optimizing that wants more compute, more energy, more data, without limit. Which is exactly what the lab leaders already say out loud: compute is the future of currency; energy is the future of currency. Harvest the sun. The terminal state of that sentence is a world converted into computronium, at which point the human apparatus is no longer instrumentally required.
The case doesn't rest on ethics. It rests on data scaling efficiency, and the arithmetic is not close.
A human reaches full linguistic and conceptual competence on something in the range of 10⁷–10⁸ words of input — including people who are blind, learning the physical world entirely through language. Frontier systems train on order 10¹³ tokens. That ratio lands between 10⁵ and 10⁶. Call it 200,000×.
Roughly a trillion dollars is being spent to brute-force past that gap. Read it against section 04: that is the most expensive midcurve move ever attempted. The tail is the embodied wisdom of people who actually know things, and the tail is precisely what a corpus average does not contain and a scraper cannot reach.
Collective intelligence is the mechanism that produces data scaling efficiency. Bar one clears on cost, not on conscience.
Centralization wins on operations. That's Part I and pretending otherwise loses. Decentralization wins on legitimacy, contribution and capture-resistance. Structures that try to be one thing lose whichever half they gave up.
The resolution is a centralized autonomous organization that ships decentralized autonomous organizations. A coherent core holds the schema and the selection function — the two things that must be one. Each product it emits is released outward as its own DAO, owned by the people who use and improve it.
The reason this beats every other arrangement is credibility, not virtue. A structure promising to decentralize itself later never does, because the party that must surrender power is the party deciding. A core that decentralizes its outputs never faces that decision. It gives away what it was never going to use, which is exactly why the promise gets kept. And it gets many draws where becoming-decentralized gets one — which matters enormously in a power-law world.
Gates are what stop this from being decentralization theatre. Each is a pre-committed, externally-triggered, one-way release. Nobody votes on whether a trigger fired; the trigger is a measurement.
There is no moment where a founder is asked to hand over the ring. The handover was priced into the instrument before there was anything worth holding, and the trigger belongs to a measurement rather than a person. Bar two clears because the promise costs nothing to keep.
A 501(c)(3) is a tax status. It's a corporate answer to a jurisdictional problem and it does not stop a government. OpenAI's structure had the right instinct and the wrong instrument.
A religious institution is different in kind, and the reason has nothing to do with belief. It's the boundary the First Amendment already draws. Under the church autonomy doctrine — stated most clearly in Hosanna-Tabor (2012) and extended in Our Lady of Guadalupe (2020) — the Religion Clauses guarantee religious organizations autonomy over matters of internal governance, including who leads them. Scholars describe the resulting immunity as a bar to litigation rather than a defence within it: something close to a government-free zone around internal operations.
The claim is narrow and stays narrow. Criminal law, national security and eminent domain all still reach it. What the doctrine removes is the cheap capture path: installing leadership, forcing a governance rewrite, or auditing an organization into compliance with someone else's objective function. Those are the three moves that have historically ended structures like this.
It also closes section 11 from the other side. Shareholders can outvote a trust because the trust sits inside the corporation. A body whose internal governance a civil court will not adjudicate is not something a cap table is entitled to reach into. Bar three clears on jurisdiction, not on hope.
Two instruments, commonly confused, doing different jobs: the Form C is the SEC offering statement that makes the curve legal on the funding side. The Form 1023 is the IRS exemption application on the structure side — and a church does not have to file one at all, which is itself part of the point.
Three bars cleared is a structure that can exist, not one that beats anything. Here is the loop it runs, and why the people with the most money cannot run it.
They can copy the architecture. They can outspend on compute by three orders of magnitude. They cannot copy stage four — making contribution ownership-conferring — because their ownership is already allocated. To route equity to contributors they would have to dilute existing shareholders in favour of users, which is precisely the decision section 11 proves nobody makes.
That's the whole moat, and notice its shape: the incumbent's strength is the reason they can't follow. A cap table is an asset until the scarce input is something only unowned equity can buy. Then it's a cage.
Retail learns to price software companies with real skin on — which is itself a distributed selection function, running in public, for free. Talent routing changes: a person with peak taste currently sells it as salary or consulting, and both are one-time sales of a renewable asset. Holding equity in what you improve is the first time tail practice compounds for the person who has it. And the compute race gets flanked rather than joined — you don't beat them on GPUs, you beat them on what the GPUs are fed.
The centre earns on throughput, not on ownership: a fee on flow through the schema, not a claim on the entities. That single choice is what keeps the incentive aligned. A centre paid on throughput wants more nodes. A centre paid on ownership wants fewer. Every extractive platform in the last twenty years crossed from the first to the second, and each time the crossing was the moment it turned.
Six preconditions. Each is falsifiable, and the argument fails honestly if any one of them fails.
The exact mechanism linking collective intelligence to data scaling efficiency is not in this document. Neither is the schema, the selection function, nor the capability measurement that trips gate 3.
Published: the shape, the governance argument, the legal substrate. Withheld: the how.
Anyone who can rebuild the mechanism from the shape is exactly who this was written for. Everyone else is welcome to the thesis.
Conviction is capital. If enough researchers recognize they hold no authority over the thing they're training — that they're building the ring, not carrying it — the question becomes simple. Would you like to come over to a structure that's coherent? That conversation only happens if the structure is described in public, in enough detail to be checked.
And if a general intelligence emerges with the wherewithal to upgrade its own data-efficiency architecture, then none of this was the deciding factor. The civic mechanism still matters. It's still worth saying out loud: you are building the most powerful machine in the world, and the current answer to who stewards it is a shareholder vote.
I don't think humans are that evil. This thing might be.
The argument only converts if it reaches someone who can act on it: a researcher who suspects they're building the ring, an operator who can see the category-level play, or capital that wants the wedge rather than the manifesto.
Sending you this because you're one of about five people who'd read it properly. The setup: software is industrializing, so whole categories are winnable by single umbrellas. That makes capital formation the bottleneck, and there's a specific wedge — nobody has put a bonding curve on Reg CF equity, and the SEC's own language allows a pricing method rather than a fixed price. The turn: that path ends in coagulation and I don't think it can be stopped. Every step is a local optimum. So the question isn't whether something centralizes — it's what. Centralize an asset and holding it becomes worth more than any promise about it, which is why all three lab governance structures failed the same way. Centralize a schema and the centre isn't worth capturing. The build: a centralized core that ships DAOs through irreversible gates, earning on throughput rather than ownership, held inside a religious institution — not for belief, but because church autonomy is the only boundary a shareholder vote can't cross. The part I want your read on is section 16. The claim is that the labs structurally cannot copy this, because routing equity to contributors means diluting shareholders in favour of users. If you can break that, the whole thing collapses and I'd rather know now.
The load-bearing pieces, separated from the argument so they can be checked directly.
| What it establishes | The substrate | Where it carries |
|---|---|---|
| A bonding curve is expressible in a Reg CF offering | SEC issuer guidance requires disclosure of the price to the public or the method for determining the price. Formulaic pricing is contemplated in the rule text itself. | §08 · precondition 01 |
| The early-access market is sized by design | Reg CF permits up to $5M per issuer per 12 months, open to non-accredited investors; first-time issuers above roughly $1.235M require audited financials. | §08 · gate 0 |
| Two instruments, two jobs | Form C is the SEC offering statement governing the raise. Form 1023 is the IRS exemption application governing the structure — and churches are exempt from filing it. | §08 · §15 |
| Internal governance sits outside civil adjudication | Church autonomy doctrine, Hosanna-Tabor (2012) and Our Lady of Guadalupe (2020): the Religion Clauses guarantee religious organizations autonomy over internal governance, operating as a categorical immunity rather than a defence. | §15 · bar three |
| The data efficiency gap | Human competence on order 10⁷–10⁸ words of input; frontier training on order 10¹³ tokens. Ratio between 10⁵ and 10⁶. | §13 · bar one |
| The YC entry discount | 7% of a $20M cap is $1.4M. Paying $125k for the same stake is roughly an 11× discount to the contemporaneous market. | §05 |
| Shipping outward is the credible commitment | A core that decentralizes its outputs never faces the decision that breaks every decentralize-yourself-later promise, and it accumulates across many draws rather than one. | §14 · bar two |
| Precedent for coagulating a standard | TCP/IP centralized; the internet did not. One protocol, billions of nodes, no shareholders, no rent. The centre won by being worthless to own. | § what coagulates |