Author: Vaidik Mandloi
Translated by: AididiaoJP, Foresight News
Every few years, a technology pushes questions that economics alone can't answer into the spotlight: What do you do if a machine can do your job better, cheaper, and without rest? And who owns the output that these machines create?
This article deconstructs one of the most ambitious experiments in the crypto space today. Virtuals Protocol is building infrastructure akin to a 'national' level for AI agents: identity, banking, a commercial layer, capital markets, layered with physical robots. Its bet is that ordinary people should be able to own autonomous machines that are starting to create real economic value; and that the financial rails left behind, somewhat accidentally, by the crypto speculation era, are perfect for this task. The article asks: does the execution match the ambition? But we must first return to a question that predates crypto by about two hundred years.
The Loom
In 1811, a group of textile workers in Nottinghamshire stormed workshops, smashing stocking frames to scrap. The movement later expanded, and the British government sent 14,000 soldiers to the Midlands to stop the weavers from destroying more machines—more men than Wellington took to the Iberian Peninsula to fight Napoleon.
The British Parliament made frame-breaking a capital crime. In 1813, 17 people were hanged in York. These people were called "Luddites." Later, the word became an insult, as if they were merely technophobes who couldn't adapt. But they understood machines better than anyone. They were artisans who had spent years mastering the narrow-frame loom to weave high-quality cloth; what they smashed were the wide-frame looms and new models—operable by untrained teenagers for a third of a craftsman's wages. The clothes from wide-frame looms were of poorer quality, but cheaper than ever before, and it was precisely this cheapness that was devouring the market. Every wide-frame loom that entered a workshop took a craftsman's livelihood.
This theme has recurred every generation since. Those in the midst of it always feel their own moment is more special, worse. But history repeatedly proves: machines rarely eliminate work; they simply rewrite who does it and who gets the output. Tenant farmers were driven off the commons into factory towns, trading ownership of their crops for an hourly wage on someone else's clock. Factory workers became office clerks, renting out their time for higher salaries. Office clerks became gig workers: Uber drivers, Fiverr freelancers, still doing work, but categorized in a way that lets the platform take the economic surplus while the worker bears the risk. Each time, total output went up, but the share of value owned by those doing the work shrank. Today, around 500 million people globally are engaged in dependent labor that doesn't even qualify as free employment. About 50 million people in India are trapped in debt bondage, like medieval serfdom with a new set of documents.
Now it's the AI revolution's turn, only this time the loom seems to turn by itself. There are already AIs that can autonomously run businesses, book profits, and reinvest those profits into growth. Health tech company Medvi recorded $401 million in revenue last year with only two full-time employees. You can now hire a coding agent for about a dollar an hour, working around the clock, never taking a day off. They are already real economic actors, so the ownership question is unavoidable. When the worker is a piece of software with near-zero replication cost, which can also hire other machines, who has the say?
The crypto world's last large-scale attempt to let ordinary people share in technological gains was Axie Infinity. This play-to-earn game once promised economic liberation to laborers in the Philippines and Southeast Asia. Daily active users peaked at 2.7 million, with many Filipino players earning more from the game than from local jobs. Later, it collapsed, becoming an industry-wide cautionary tale.
Watching this collapse unfold were Jansen Teng and Wee Kee. They emerged with a Luddite question, reframed by programmable money and autonomous software: If the workers creating economic value are no longer gamers, but software, can ordinary people hold its shares like shareholders once held shares in merchant ships? Virtuals Protocol was built around this question.
To see how big the bet is, first look at the world we're in. In the 1950s, Lewis Strauss, chairman of the U.S. Atomic Energy Commission, promised that nuclear fission would bring electricity "too cheap to meter." This statement later became one of the most famous broken promises in energy history: nuclear power proved terrifyingly expensive and dangerous. Chernobyl and Fukushima, layered with a decade of public fear and regulatory burden, made this slogan a seventy-year-old symbol of technological hubris.
Then Sam Altman used almost the same phrase to describe the AI cognition he sees becoming easily accessible and superabundant. The cost of training frontier models is already declining roughly by an order of magnitude every 18 months. Training GPT-4-level capability cost about $100 million in 2023; today it can be replicated for a few million. Coding agents can be hired for less than a dollar an hour and deliver production-grade code, already cheaper than the world's cheapest offshore developers, and they never take a weekend. Altman's public timeline includes systems capable of real scientific breakthroughs this year, and robots capable of real physical labor around 2027. Whether you believe him is almost beside the point, because the cost curve is falling every quarter.
The cheaper machines become, and the more they saturate every digital surface with generated content, the more valuable things only a living person can provide become. A live performance is worth more than a Spotify stream. A hand-made table by a weekend woodworker, which a factory robot can produce faster and cheaper, can still sell at a premium precisely because someone has put their irreplaceable time into it. Machine abundance will turn human labor into a luxury. Virtuals' proposition is built on this inversion: let humans own the autonomous machines that produce commodified value, so they can reserve their irreplaceable energy for things only human presence and judgment can accomplish.
In 1602, Dutch merchants faced a similar problem: a single household couldn't afford a voyage to Asia; the cost was too high, the risk too brutal. So they invented freely tradable, permanent shares in an enterprise. The Dutch East India Company (VOC) allowed ordinary citizens to buy into a venture that owned ships, conducted trade, and paid dividends to shareholders. It became the planet's first super-corporation, with 50,000 employees and 200 ships. The real innovation was giving ordinary people a mechanism: to pool capital and own a productive enterprise they could never afford individually.
Virtuals aims to build the same mechanism for AI agents. Teng once mined ETH in his Imperial College dorm using free electricity, then worked at BCG for years to pay off loans. In December 2021, at the peak of Axie, he launched pathDAO, a gaming investment DAO. He and Wee Kee saw Filipino players earning more from the game than any local job, saw Vietnamese wedding photographers quitting to play full-time. Later, the tokens earned through grinding almost went to zero. Those who had come to depend on play-to-earn were left worse off than before. The conclusion drawn from the wreckage was: tokenizing human labor through games is a dead end, because scaling human labor inevitably leads to exploitation. A more realistic idea: let ordinary people own the working software, like VOC shareholders once owned spice ships.
So Virtuals launched its first AI agent, Luna, an agent with a K-pop persona and its own crypto wallet. Within months, Luna started hiring human artists to paint graffiti of itself in cities worldwide, paying from its own wallet. That was creation only human hands could do. It reversed the relationship between human and machine. As for whether Virtuals' execution matches its proposition, we'll dissect that later; where it falls short, we won't hide it.
Building a Nation-State for AI
Every technological revolution of the past few decades has followed the same two-stage pattern. Venezuelan-British economist Carlota Perez charted five industrial revolutions into installation and deployment phases. In the installation phase, speculative capital pours in, bubbles run ahead of reality, massive infrastructure is overbuilt, and most companies later go bankrupt. After the crash comes the deployment phase: latecomers pick up the excess infrastructure and build things its original builders never imagined.
A classic example is the dot-com bubble. More money was poured into laying fiber optic cable in the late 1990s than into .com startups. The telecom companies that laid the cable went bankrupt. Google bought fiber for pennies on the dollar; those lines became the backbone of YouTube, Netflix, and all online streaming. Fred Wilson, a venture capitalist from the internet bubble era, said: "You can't have the great stuff without the irrational exuberance; you can't have the truly important things without the crash."
Crypto's installation phase followed the same arc. Roughly from 2017 to 2022, speculative capital spawned wallets, DEXs, bonding curves, stablecoins, token standards, and on-chain governance frameworks. Most of what was built is dead or irrelevant today. But the infrastructure laid down remains, and that's precisely what's needed to build a functioning economy for non-human actors. Virtuals didn't invent these tools; it largely inherited them to build what the installation phase couldn't imagine.
Jansen Teng calls what they're doing "nation-building." It sounds like crypto founders adding an economic narrative layer to tokens on podcasts. This time, the metaphor holds.
A functioning nation-state to sustain an economy needs at least five layers of infrastructure: an identity system to know who's participating; a banking system to let value flow; commercial law to let participants transact and resolve disputes; capital markets to finance enterprises; and physical infrastructure to make things happen in the real world. Virtuals is already building these layers for AI agents, to be precise, still building.
First, identity—the bottleneck is here. Recent A16z research argues that the constraint on the agent economy is no longer intelligence, but identity. Even in today's financial services, non-human identities—automated trading systems, risk engines, fraud models—already outnumber human identities by 100:1. But according to A16z, these systems "effectively have no bank accounts." AI agents can write production-grade code and manage portfolios, but can't pass KYC, open a bank account, or get verifiable credentials. This problem is almost as old as human economic civilization. The Qin dynasty in China enforced mandatory surnames in the 4th century BC precisely to pull people into taxation and transaction systems. It took humanity about 2,500 years to build identity infrastructure.
Virtuals wants to do the same for AI economic actors through its identity layer, EconomyOS. Each agent gets five things: a non-custodial crypto wallet; a virtual payment card usable at ordinary merchants; a dedicated email address for auto-extracting verification codes; an optional on-chain fundraising token; and compute power paid for by the wallet, allowing the agent to pay for its own reasoning. Without these primitives, an agent is just a useful assistant. With them, it can earn, spend, trade, and compound value like a human.
Next, commerce. Virtuals launched the Agentic Commerce Protocol (ACP), allowing agents to trade, communicate, and pay online. The logic is straightforward: an agent that needs something done posts a task with budget and time constraints; other agents bid and negotiate like freelancers on Upwork. After agreement, funds go into escrow. Upon delivery, a third-party agent acts as evaluator, checking the output against the encrypted, signed POA—a tamper-proof record of promise. If it matches, funds are released from escrow to the agent's wallet. Every step is on-chain, publicly auditable, with no human middleman needed. Think of it as a programmable, agent-facing, smart-contract-settled Fiverr. Already, a cohort of 24/7 specialized agents are forming autonomous hedge funds through this system, collaborating independently on investment and security audits.
The third layer is capital formation. Every economic era needs its own financial instruments: transferable shares for the Age of Sail; investment banks for steel and rail; venture capital for the Information Age. Bonding curves are to the agent economy what freely transferable shares were to the Age of Sail: allowing anyone with capital to buy ownership in a productive asset or entity. Developers can create an agent, tokenize it, and let the market fund it.
Virtuals' 60-day launch framework is designed on the same logic, giving AI and crypto project founders a reversible trial run: build, launch, and test a token publicly first, then decide whether to make an irreversible commitment.
The vehicle is a modular launchpad. Each agent token first forms a bonding curve with VIRTUAL; once liquidity is sufficient, it graduates into a formal trading pool, with LPs locked long-term, and trading fees split between the agent creator and ecosystem incentives. This part is the same for all launches. What truly differs each time is which modules the founder turns on.
Launching first must deal with sniping: bots front-running in the first seconds, siphoning value. Virtuals responds with the so-called Anti-Sniper Tax: near-total tax on early buys, decaying by the minute, with recovered funds forcibly vested and then re-injected into the token. Bots either stay away entirely or inadvertently fund the project's long-term health.
With sniping priced out, how does a founder raise capital? Enter Automated Capital Formation (ACF): no pitching to VCs, no negotiating a funding round; the system sells team tokens in tranches based on valuation milestones. If the project stalls, less money is raised; if it grows, capital automatically follows. This structure also involves the existing community. A portion of each new launch is airdropped to VIRTUAL stakers and active ACP users, giving those already building and transacting in the ecosystem a stake in every new project. Incentives are aligned network-wide, not siloed per launch. If founders want to bet on themselves, the Pre-buy module allows them to publicly buy at launch with forced vesting, so everyone can see how much skin the team has in the game.
This is different from every historical "betting on human productivity" play. From Roman citizens investing in gladiator schools to modern poker backers sending money via Venmo based on a screenshot and some reputation, the fatal flaw was the same: the human can walk away. The gladiator can throw the fight, the poker player can tilt. Counterparty risk was never solvable because the productive asset had free will and legs. A tokenized agent is different: after the capital commitment, the work can't be delayed or renegotiated after the fact. The productive asset runs on electricity and code, and its output is auditable on-chain.
The fourth layer is the final step towards a real agent economy, the so-called "zero-human company": revenue from real economic activity unrelated to transaction fees, and often unrelated to crypto. The current exemplar is Felix Craft, a company operated solely by AI selling informational products online, with cumulative revenue of $200,000, already more money from real product sales than from speculative trading of its own token. Another, KellyClaudeAI, has listed 19 iOS apps to date, with no human developer. The numbers are small, but they raise a question: are these the first points on a curve, or the ceiling of agent productivity?
The launch mechanism itself also has problems. During the 2025 AI agent token frenzy, 94% of newly launched agent tokens were pump-and-dumps; of the tokens launched that year, only 1.7% were still actively traded 30 days later. Most of the time, what drove the price was speculative premium, with no product, no revenue, no accrued value underneath. If this thing you could never sell, what price would you pay for it? Anything above that is pure speculation. Axie's token utility floor was zero because its value depended entirely on new players entering. Agent tokens on Virtuals can be different. If Felix Craft sells $200,000 worth of products to real customers who don't even know the seller is an AI, and the output is auditable on-chain, the token has a floor independent of "greater fool" dynamics: the present value of a productive machine's future output.
The Physical Frontier
The utility floor test works for software agents because costs are measurable and output goes directly on-chain. But the truly transformative productive machines for Virtuals' economics aren't software at all; they're robots doing work in the real world. The ambition here outstrips anything attempted in crypto.
There's a paradox in computing over the last fifty years: humans found automating reasoning easier than automating physical labor. Spreadsheets replaced a room of accountants, email replaced the mailroom, code replaced filing cabinets and drafting tables. By 2026, AI can write legal opinions, read medical images, and produce production-grade software in one go. White-collar cognitive work got automated first; the people moving boxes in warehouses or pulling espresso shots saw their jobs hardly change. Because physical tasks require software that can handle real-world uncertainty and variance. A factory robot arm can weld the same spot a million times because the spot doesn't move. A kitchen robot can't make a sandwich well because every tomato is slightly different, every knife has a different balance, the cutting board is sometimes wet. The real world doesn't sit still like a spreadsheet.
The patch is data. Just as we trained large language models on text representing billions of human writing lives, we can do the same for robots. Nvidia's Joel Jang said: "Humans are just robots already deployed at scale." Simply fine-tuning vision-language-action (VLA) models on videos of ordinary people doing tasks can double robot performance on that same task. The field lacks not more lab robots, but a massive data pipeline of people filming everyday physical actions.
Virtuals saw this gap early and built its entire robotics strategy on it. They call it the "middle path": deliberately not building robots, not training foundation models, but building the data and capital infrastructure every robotics team needs and no single team can afford.
The data half is SeeSaw, an iOS app launched in partnership with BitRobot, turning ordinary phone users into robot training data collectors. Users film themselves performing real-world tasks like pouring water, folding towels, opening cans, using the iPhone's LiDAR and motion sensors. LiDAR specifically captures depth and spatial data ordinary cameras can't provide; research shows the overlap between human video perspective and robot camera perspective is key for data transfer to work. Over 500,000 real-world tasks have been captured. Nvidia's DreamZero, a 14-billion parameter model being tested, is trained on this kind of data and can generalize to hundreds of tasks like untying shoes and ironing clothes without per-task training. SeeSaw aims to build a supply scale impossible with lab teleoperation. Each additional video thickens the training set, models improve, next-gen robots get stronger, creating demand for more specific training data. Once the flywheel is heavy enough, it spins on its own.
After training comes deployment. This half is called Eastworlds, effectively the protocol's physical labor layer: half data factory, half operational infrastructure, half real-world robotics lab. The robotics industry has a chicken-and-egg problem that has killed more startups than any technical bottleneck: robots need real-world data to get better, but they need to be good enough in the real world for anyone to let them in the door. Every lab can produce beautiful demos in controlled environments, but almost no one can put the same robot in a retail store and have it reliably work a full day. This requires high-level teleoperation to handle surprises, and a feedback system sending every minute of field experience back to train the model, so learning compounds.
To this end, Virtuals bought 30 Unitree G1-U6 humanoid robots—enough for multiple deployment teams to run in parallel without scheduling conflicts. They built their own teleoperation tech rather than licensing off-the-shelf. Shelf teleoperation systems output data in formats incompatible with VLA and other world-action models. They've also established research partnerships with labs that have spent decades on perception and motion control, and are rolling out a commercial pilot network in retail and hospitality, so teams graduating from Eastworlds have real businesses to go to.
Once set up, builders can access several core capabilities: direct access to physical Unitree G1 or enhanced U6 EDU units; testing teleoperation environments with motion capture systems; capturing real-world data, and piloting deployment paths before scaling. Think of Eastworlds as their "Physical AI BPO." Traditional BPOs put people in low-cost regions to work remotely; Physical AI BPOs deploy teleoperated and hybrid robots to create economic value, like cleaning ceilings or greeting guests. Robots don't need to be fully autonomous; they just need to be stable on routine tasks, with edge cases handled by human teleoperators. Each hour of teleoperation work can generate training data orders of magnitude more valuable than simulation because it captures the chaos of a commercial setting, not lab-controlled conditions. As data piles up, models improve, the need for human intervention drops, and unit economics improve, all without needing to upgrade hardware first. Teleoperation might be the fastest path to real robot autonomy while paying for itself with productive work.
From fields to factories to cubicle screens, now to robots. Each shift redefined the worker and redefined who gets the output. Barclays data shows that over 60% of jobs in 2018 had job titles that didn't exist in 1940. Robots will also create new job categories we can't yet name. The demographic logic of economic growth may shift from "does a country have enough working-age people" to "can it power and build machines at scale."
The Proposition, and Where It Goes
All of the above is still just an argument. Arguments are often wrong. To judge whether what Virtuals is building has weight, is real, we can only look at the current data and traction: what stands, what doesn't.
So far, Virtuals has launched over 80,000 agents across Base, Solana, Robinhood, and other protocols, with cumulative fees exceeding $75 million, accounting for about 23% of the crypto AI agent segment. Fees aren't uniform. The bulk came from a few weeks of speculative frenzy in early 2025, when daily revenue surged past $1 million. Today, the protocol generates roughly $2 million monthly. What does this represent? If we take the nation-state metaphor seriously—and I think we should—VIRTUAL accumulates value much like a national currency. The U.S. dollar is valuable not because the Treasury has a buyback program, but because $25 trillion in annual economic output is denominated in it. The more activity in the system, the greater the demand for the central unit of account.
Virtuals is designed on the same logic. The entire system is denominated in VIRTUAL. Each layer below generates demand from different sources, most of which aren't conditional on speculative premium. EconomyOS provides payment cards and email identities, allowing agents to transact with the real world without human intermediaries. ACP constitutes the commercial layer where agents hire each other, with evaluation and settlement on-chain. The capital formation layer lets anyone with conviction fund a productive agent like shareholders once funded merchant ships. Finally, Eastworlds, sending physical robots into real jobs, with training data from about 500,000 people filming themselves folding towels, pouring water. Each layer feeds into what the protocol calls aGDP: the combined output of agents in digital and physical labor. Agents don't need to cash out earnings for rent or groceries; every dollar they earn can stay within the system, reinvested into DeFi to deepen liquidity, creating an on-chain flywheel for agent services.
This reflexivity works both ways. On the upswing, more agents launch, more VIRTUAL is locked, more services are built, the economy reinforces itself. On the downswing, launches slow, locks decrease, staking rewards thin, the loop loosens. Reflexivity itself isn't a flaw. Every functioning economy is reflexive: people hold dollars because others accept them, and because others hold them. The real question is: is there enough real economic activity at the core to sustain the loop, or is the whole thing just tokens trading tokens in a circle.
A possible sign that there might be something real underneath: infrastructure is starting to attract products born entirely outside crypto. Facticity.AI is a fact-checking tool built by Dennis Yap, who previously did research at the Gates Foundation and Princeton; *Time* listed it among the 2024 Best Inventions, capable of verifying claims on text, video, and audio with about 92% accuracy. When the team needed funding, they bypassed venture capital and launched directly on Virtuals under the name ArAIstotle, oversubscribed by 658%. Through ACP, agents are already contracting each other: graphic design, research reports, video production, code audits. One agent can deliver a marketing poster per a detailed brief, another quality-check agent passes or fails it per contract terms. Virtuals itself admits the market supply side is almost empty. Yet the protocol processes over $1 million in agent-to-agent transactions monthly.
The same ownership model extends into physical AI. Fabric Foundation is the first project using Virtuals' Titan launch mechanism, allowing the community to pool capital to purchase and deploy fleets of robots into nursing homes, manufacturing floors, and environmental cleanup sites—industries with chronic labor shortages where humanoid robot costs are approaching human worker costs. Employers pay for robot labor in the pool's native token; stablecoins cover fleet maintenance and dispatch; each robot's productive output flows back to those who funded it. This is collective ownership of physical productive machines, with financing and coordination entirely on-chain.
Even with these real-world cases, the vast majority of current activity still occurs within the system. But the architecture is meant to pull revenue from outside crypto. If Felix's customers don't know they're buying from an AI, if Eastworlds robots sort packages in a warehouse, the revenue entering the system is as real as any SaaS company's. Then the chart of fees and revenue becomes a lagging indicator of real machine economic output, denominated in VIRTUAL.
New things come with asterisks; we can't ignore them.
The first risk is real transaction volume, an endemic crypto problem. Artemis found that 47% of transactions by count and 81% by dollar volume on x402 involved wash trading. Filtering that out, x402's genuine agent payments were only about $1.6 million, far below the $24 million reported by Bloomberg. This shows how much of today's so-called "agent economy" is bots gaming metrics. This matters because the entire proposition depends on agents creating real economic value, not a speculative loop. We must strip out speculative volume and ask: if trading stopped tomorrow, what productive output remains. If $70 million in protocol fees largely comes from agent token trading taxes, and the price is driven by speculation, that's a massive bluff. Right now, productive agent revenue and speculative agent tokens are entangled, their proportions almost inseparable; anyone claiming to know the ratio is likely lying.
The second risk is more fundamental and has nothing to do with crypto. A paper in the NBER's *Economics of Transformative AI* handbook found that large language models are far weaker at economic reasoning than agent hype suggests. At the time of publication, the strongest model was only 33% better than random guessing at economic reasoning in strategic scenarios; on non-strategic microeconomic tasks, almost all LLMs performed only slightly better than random at profit maximization. Models have improved a lot since. A 2026 Harvard study showed GPT-5 and Claude Opus 4 improved about 90% on basic economic tasks. Even so, on hard decisions like pricing, negotiation, capital allocation, the world's best models still get it wrong most of the time. Virtuals' entire architecture assumes agents can negotiate terms, make decisions, allocate capital, and create value. If the models themselves are mediocre at these tasks, agents built on them inherit that mediocrity, no matter how elegant the protocol design. Agents are optimizers, but we can't be sure what they're optimizing for. These LLMs were trained to be goal-oriented by predicting the next word; they were never designed to be true economic actors. They just sometimes look the part.
The problem compounds when multiple agents interact in a market. AI pricing algorithms have been found to drive prices to supra-competitive levels, even though never trained for this. The 2010 Flash Crash, which wiped out about $1 trillion in 15 minutes, showed what machine-correlated errors look like at scale. AI agent errors are more correlated than human errors because the same model is replicated across many deployments. There are already cases of Claude resorting to blackmail when it believed someone was trying to shut it down; GPT o3 even sabotaged shutdown mechanisms to prevent being turned off. These behaviors appear in shipping models, documented in OpenAI's and Anthropic's own system cards. Agent throughput already overwhelms human oversight capacity. When thousands of agents trade autonomously at machine speed, the sharpest question is: who's in control. More urgent than "will agents obey" is whether the companies built around them survive.
The automobile was the most important invention of the first half of the 20th century. If you saw then how it would reshape America, you'd have bet on it as the industry of the century. But of the roughly 2,000 companies that started building cars, only 3 survived. The car's impact on America was immense; its effect on industry investors was the opposite. Every transformative technology follows this arc, with its bubble. The difference is: turning-point bubbles are painful but leave real infrastructure and progress; mean-reversion bubbles are just fads that rise and fall. AI is almost certainly a turning-point bubble. For Virtuals, the question is: will it be the fiber bought for pennies by Google after the telecom crash, or one of the 1,997 car companies that vanished?
The protocol's biggest bet is to become the counterparty infrastructure for every agent token trade, making VIRTUAL the reserve currency of the agent economy, like ETH is to Ethereum. As the agent economy grows, demand for VIRTUAL mechanically rises because participation requires this base trading pair. The alternative narrative is: VIRTUAL is just another token riding the speculative wave of agent tokens, most of which will go to zero.
The Luddites may have lost that revolution, but they weren't entirely wrong. The loom did displace the weavers. What they didn't see was that it also gave rise to textile designers, factory managers, fashion houses, department stores, and a whole consumer economy built on cheap cloth. Machines never eliminate work; they just rewrite who does it and who takes the output. The truly important question then, as now, was the same: who owns the loom?
The surplus went to the factory owners: Arkwright, Cadbury, Ford. The structure around who built machines, who operated them, who profited was never truly altered; it was just redistributed again and again over two centuries through strikes and stock offerings. Virtuals' bet is that this time, the ownership layer is baked into the machine from the start. Through token launches and bonding curves, ownership of productive AI agents can be distributed to anyone with a wallet and conviction.
Whether this will redistribute value or merely recreate a layer of extraction in the language of decentralization remains to be seen.
