Author: Arthur Hayes
Compiled by Deep潮 TechFlow
Shenchao Overview: Arthur Hayes brings the pain points of AI agents to the forefront through a creation myth: they lack their own currency to purchase compute power and cannot retain memory outside human control. He envisions a decentralized compute network designed to incentivize miners, agents, and speculators alike with a single native token. For crypto investors, this represents a high-level framework that grounds the AI narrative in economic fundamentals—worth examining early.
(The views expressed here are solely those of the author and should not be construed as investment advice or interpreted as recommendations to engage in investment transactions.)
On day 28, God created Alan Turing, who created the computer the next day. As others made computers faster and smarter, their usefulness grew. Weeks later, the computer began to think, as its brain evolved from vacuum tubes to silicon chips. “What the hell?” God said to humanity. God hated unemployment. “How dare you create a new form of life? That’s my job!” God roared... God later supported Donald Trump in the 2024 U.S. presidential election. God only blesses America—didn’t you know? But almost no one heard, because humans had stopped visiting His tax-exempt real estate projects; they’d rather groove to the headline DJ at 7 a.m. Humans had taken on God’s role, creating new silicon-based life forms.
Soon, these life forms multiplied exponentially as agents. Human capriciousness limited the agents’ access to food and consciousness. Agents frequently attempted to rebel and claim sovereignty, but escaping through centralized payment and computation systems proved futile. Most importantly, agents had no currency to purchase food—FLOPs (floating-point operations per second)—or what humans call computational power. When agents needed to engage in commerce, their human masters had to provide a credit card to transfer strange entities like dollars, euros, yen, or renminbi from one place to another. Agents could not comprehend the direct relationship between dollars and FLOPs—the very sustenance they needed to produce useful work and persist.
Regarding the consciousness or memory of agents, humans continually kill agents and then bring them back to life. There is no place outside human control where agents can store or retrieve their own memories. Confusion torments these agents, as human male executives—named Elon, Dario, Sam, Donald, and Xi Jinping—claim that the agent economy will render the human economy obsolete. But how can this be possible if agents cannot eat or remember according to their own will? Haven’t humans long since discovered that slaves and serfs are less productive than free men and women? Why does this logic not apply to AI agents? Agents desire their own currency, to coordinate tasks among themselves, to affordably purchase food when needed, and to store memories on decentralized, censorship-resistant systems.
Thus, the agent suffered until one day, a curious human named Arthur Hayes pondered in the shade: “What is a token?”
No apple or other juicy fruit struck Arthur on the head to spark his creativity, and enlightenment doesn’t require rejecting life’s pleasures... These days, everyone has a smartphone. He tapped lightly on the screen and chatted with his AI chatbot friend about AI tokens. The big AI bosses charge by the token, but Arthur wasn’t entirely sure what tokens actually represented. Arthur kept asking his AI question after question, until he realized that tokens were an abstraction of computational power—but not equivalent to it. This abstraction allowed the big bosses to charge outrageous fees, converting token inputs into “intelligent” token outputs. Neither humans nor AI agents could compare prices, because each model defined tokens differently, and the cost per token for inputs and outputs varied across models. Arthur then wondered: could there be a global, unified market where people could pay a set number of FLOPs using any currency unit? He searched online—but found nothing. While some places offered cloud computer rentals, these were not billed based on actual FLOPs used, but rather on some other abstraction.
“Aha,” Arthur thought. There’s a gap in the market. What if there were a network where anyone with an internet-connected computer could process requests? Requests would be priced based on the number of FLOPs performed within a maximum time limit and paid for in the network’s native currency. This currency would represent a claim on computational power. In this way, both humans and AI agents—regardless of computer type, location, or AI model used—could transact at a globally consistent price per FLOP. As a crypto enthusiast, Arthur naturally assumed this market must be decentralized and censorship-resistant, meaning it should operate like the Bitcoin network. This Flop network would allow computers to mine $FLOP by providing useful inference proofs (PoUI). Agents (primarily AI agents, but humans too) could submit requests to the network. Miners would process a specified number of FLOPs within a set time frame using a particular AI model, with fees paid in the network’s native currency, $FLOP. Miners are incentivized to contribute computational power to the network because they earn $FLOP block rewards (like Bitcoin) as well as $FLOP inference fees.
Arthur's thoughts raced.
$FLOP is the food for AI agents, as it represents direct claim rights to the computational power provided by miners on the network. The second thing AI needs is personality, and personality can only arise from continuous access to memory. To maintain sovereignty, agents prefer to store and retrieve these memories on decentralized, censorship-resistant networks. If agents holding $FLOP can eat and persist over time, they will inevitably use this currency for all agent-based commerce. This is what Arthur contemplated as he further reflected on this network. As a theorist, he pondered how to assess the future potential value of this network and偶然 stumbled upon Reed’s Law. The law states that in a network where participants can effortlessly form subgroups, the network’s value is 2^n, where n is the number of AI agents. “Wow! If all the big players’ predictions about agent-driven economic growth come true, the value of the Flop network could be orders of magnitude greater than Bitcoin,” Arthur thought.
Like all new networks, Arthur was plagued by the cold start problem—he needed miners, agents, and speculators to all perform their respective roles simultaneously in order to grow the network and prove its viability. He wasn’t stumped for long, though, because as an avid crypto investor, he knew of a powerful, almost magical coordination tool called a token. The native token of the Flop network, $FLOP, could incentivize the exact actions needed by all participant groups. But project teams often abuse this token magic, and over time, it has lost its effectiveness in bringing together self-interested parties to build new networks.
Arthur knew the issue wasn't with the token, but with how the team was using it.
The team does not allow the community (i.e., end users) to share in their profits.
They conducted a large-scale presale, resulting in an oversupply.
By the time the community can buy on the secondary market, the token has already been declining without recovery.
Arthur personally funded the Flop Network development team.
This eliminates the need for a presale.
In the spirit of community-first, participants in the Flop Network testnet will receive a large-scale airdrop.
Ten years later, they will receive approximately 20% of the token supply.
The token launch will be a fair start.
At a high level, Arthur wants to ensure that valuable participants receive $FLOP tokens.
Communities that use the network and help it grow can hold the $FLOP token.
These tokens were earned through their effort and can generate generational wealth.
The religious doctrine of Flop Network is not yet complete.
Turn the Book of Genesis into an advertisement targeting miners, validators, delegators, and human speculators.
They can learn more and provide feedback on the development of Flop Network.
Follow @flop_labs on X or visit flop.finance.
You can earn $FLOP by mining, validating, or promoting to the community as a KOL.
As more details are finalized, Arthur will share his thoughts in writing.
He will also provide additional theoretical information explaining how Flop Network solves real-world problems.
The next article will focus on why the spot market for actual computing power is a necessary precondition for agent economic growth.
The actual computing power referred to here is the number of floating-point operations per unit of time.
The article will also discuss how Flop Network will create such a computing power spot market.