I like Stanley Druckenmiller’s tendency to approach stocks with a “buy now, research later” instinct. Sometimes you just get this feeling where everything clicks about a stock, mainly that the market is not seeing something. After seeing Figmas new product demo and realizing that AI sucks as design it just clicked for me.

In this case, the market has rewarded all of the hyperscalers and then moved on to all of the compute and infrastructure companies—Dells, HPs, Microns, Sandisks of the world. These market moves made total sense, grounded in real structural demand driven by AI, though maybe they over-indexed a touch.

We’re now at a point though where the SaaS apocalypse narrative (the idea that frontier model companies were going to replace every SaaS company) is becoming pretty obviously wrong. I think it’s been obvious to some people who are deep in the space, but the Wall Street analyst crowd, who are really just observers of this technology, are starting to experience that reality themselves. And I think the narrative is going to change.

There’s been this open debate among AI experts (of which I am definitely not one), but as an observer of the conversation you could see a bifurcation between two groups. On one hand, you had people like Dario Amodei saying that within a year all jobs were going to be gone. On the other hand, various AI researchers and experts were saying we’re at least ten years away. I think we’ve hit an inflection point where the latter camp is undoubtedly winning the argument.

Just from personal experience using AI every single day, I can say that for coding there’s no doubt it is the most productive, addictive, and useful thing ever created. It dramatically lowers the barrier to entry and allows creatives to turn ideas into software.

On the design side, though, I have a completely different opinion. I think most people would agree that AI is great for prototyping designs, but not much more than that. In the same way that Claude Code has become an essential tool for developers, I think the equivalent for designers is going to be Figma. There will be other apps, but Figma is by far the leader in this category.

The stock has underperformed since the IPO for different reasons, but I think once the narrative around AI replacing designers took off, it took an even bigger hit.

As a Figma user, I had my doubts about its future too because their AI tools felt lackluster. Then last week, at their conference, they released a lot of really impressive products that organically weave AI into the product. The sentiment among designers about the new products seems to be overwhelmingly positive.

Figma is becoming a control plane for design—a shared canvas where AI has all the context it needs to orchestrate creative workflows. It’s the perfect canvas, or substrate, for AI to sit on top of because it provides all of the context and tooling AI needs to create really great design.

Will AI replace designers? I don’t think so.

Will designers use AI on top of a powerful canvas like Figma to create far more than they could before? Absolutely.

I think AI-augmented design tools are the future, not AI replacing designers.

The one thing AI cannot do is intrinsically have taste. No matter how you prompt it, there’s something fundamental about creativity and design that has to come from a human. Design is an ephemeral thing. It comes from the creativity of one person and resonates with many people. Eventually it becomes popularized and copied, and at that point it becomes old. Then someone creates the next thing.

All of that is to say that I think the question of whether AI can have taste has a pretty clear answer: no.

That’s how I think this plays out. Humans remain the creative drivers for the foreseeable future, and Figma becomes the platform they use to amplify that creativity—not replace it.

As far as the stock goes, I think it has bottomed here. It’s trading at a reasonable valuation. The company is intentionally sacrificing some near-term profitability to invest aggressively in AI, something management has been very explicit about.

I expect revenue to continue growing. Profitability will come over time, and when the market starts realizing that, the stock price will likely move before the financials fully reflect it. Markets are reflexive like that.

By the time Wall Street turns bullish, it’ll probably be too late.

The agent is brilliant. It can reason through multi-step problems, write software, summarize legal documents, and plan your vacation. But it cannot tell you what Tide Pods cost at the Walgreens on 5th Street. It has no idea if there’s a sale at Target this week. It doesn’t know the shelf price of eggs at your local Kroger, or whether Bounty paper towels are cheaper at Costco or Walmart right now.

The data exists. Somewhere, a receipt was scanned. A shelf label was photographed. A browser extension captured a price. But that data is sitting behind an enterprise contract that costs $100,000 a year, refreshes monthly, and requires a six-week sales cycle to access.

This is the central tension of the agent economy. We have built extraordinarily capable AI systems that can think, plan, and act. But the real-world data they need to be useful, the kind of data that reflects what is actually happening in stores and wallets and shopping carts right now, is locked inside an industry that was never designed to serve machines.

That is about to change. And the change is not speculative. It is structural, already underway, and inevitable.

The Data Agents Actually Need

There is a category of information that no language model can generate on its own, no matter how large or sophisticated. It is the data that only humans produce through their daily actions: what they buy, where they buy it, how much they pay, and what they almost bought but didn’t.

This includes real-time shelf prices across thousands of stores. It includes receipt-level purchase data showing exactly which products were in someone’s basket, at what price, in what quantity, at which retailer. It includes brand market share broken down by region, income bracket, and age group. It includes promotional calendars, price history trends, and the browsing behavior that reveals what consumers are considering before they buy.

No foundation model can hallucinate what eggs cost at a Kroger in Memphis today. No training corpus contains this week’s promotional depth on Olaplex shampoo at Ulta. No amount of reasoning can substitute for the ground truth of what real people are actually spending their money on, right now, in the physical world.

And yet this is precisely the data that agents need to answer the questions people are starting to ask them.

A consumer asks their agent to build a grocery list that stays under $150 based on what’s on sale nearby. A brand manager’s agent needs to know if a competitor just undercut their pricing in the Southeast. A portfolio manager’s agent needs real-time brand velocity signals to build conviction ahead of earnings. An economist’s agent needs daily basket-level inflation data that the BLS won’t publish for another month.

These are not hypothetical use cases. These are the questions that AI agents are being asked today, and they cannot answer them because the data they need is inaccessible.

Why the Incumbents Cannot Serve This Market

The consumer intelligence industry is dominated by a handful of companies: NielsenIQ, Circana (formerly IRI), and Datasembly among the largest. They are serious businesses that have built excellent products over decades. NielsenIQ covers 2.6 to 4.5 million UPCs across 1,100 product categories. Circana’s National Consumer Panel tracks spending across 100,000 households. Datasembly captures over 12 billion price observations per week from 150,000 stores.

But these companies were built to serve human analysts at large enterprises who review data on dashboards, in quarterly business reviews, through syndicated reports. Their architecture reflects this. And that architecture is structurally incompatible with the agent economy for five specific reasons.

Pricing model. Minimum commitments start at $50,000 to $100,000 per year, often much higher. An AI agent that needs a single price comparison or one brand share data point cannot sign an enterprise contract. The unit economics are fundamentally mismatched.

Sales cycle. Onboarding takes weeks to months of human negotiation: demo calls, procurement reviews, legal sign-off. Agents operate in milliseconds. There is no “schedule a demo” in machine time.

Data freshness. Most incumbent data refreshes weekly or monthly. An agent advising a consumer on where to buy groceries today needs prices from today, not from last Tuesday’s batch job. A trading agent building a position needs brand velocity signals in real time, not in a monthly report.

Access model. The data is delivered through dashboards, portals, and CSV exports. Agents don’t have eyes. They don’t click through UIs. They need structured APIs that return JSON in response to programmatic queries.

Discoverability. There is no machine-readable manifest telling agents that these data services exist, what they cost, or how to pay for them. In the agent economy, if your service doesn’t have a discoverable endpoint, it effectively does not exist.

This is not a criticism of these companies. They built great products for a market of human buyers with annual budgets and quarterly review cycles. But the buyer is changing. The next wave of customers for consumer intelligence data will be machines, and the incumbents’ entire go-to-market architecture, from pricing to delivery to discovery, will need to be rebuilt to serve them.

Capability NielsenIQ / Circana Datasembly Agent-Native Model Minimum commitment $100K+/year $50K+/year $0.005/query Data freshness Weekly/monthly Near-real-time Daily/real-time Access method Dashboard/portal Enterprise API Open API + agent.json Agent-accessible No No Yes (x402/MPP) Self-serve onboarding No (sales cycle) Limited Instant, no contract Demographic segmentation Panel-based, projected No Real user profiles

The Payment Rails That Make This Possible

Even if someone built the perfect agent-accessible consumer data API tomorrow, there was until recently no good way to charge for it at the right price point.

Consider the economics. A single price lookup might be worth half a cent. A brand share query might be worth a penny. An alpha signal for a trading agent might be worth two and a half cents. At these price points, credit card interchange fees alone would consume the entire transaction. Traditional billing through API keys, monthly invoicing, and account management requires the kind of human infrastructure that defeats the purpose of building for machines in the first place.

This is the micropayment problem, and it is the reason the internet never developed a native payment layer despite being designed with one in mind. In 1997, the architects of HTTP reserved status code 402, “Payment Required,” anticipating a web where payments would be embedded directly into the protocol. That status code sat unused for 29 years.

It took machines, not humans, to finally need it.

In May 2025, Coinbase launched x402, a protocol that brings 402 to life. The flow is simple: an agent requests a resource, the server responds with a 402 status code containing payment details, the agent pays in USDC on-chain, and retries the request with proof of payment. The entire cycle completes in under a second. No accounts. No API keys. No contracts. No vendor lock-in. About five lines of code to integrate.

As of April 2026, x402 has been contributed to a neutral foundation under the Linux Foundation, backed by Coinbase, Cloudflare, Stripe, Google, Visa, Mastercard, AWS, Shopify, and Circle. This is not a startup experiment. This is the internet’s major infrastructure players agreeing on a standard for machine-native payments.

Alongside x402, Stripe and Tempo co-authored MPP (Machine Payments Protocol), launched in March 2026. Where x402 handles stateless, per-request payments, MPP adds a session layer for higher-frequency use. An agent pre-authorizes a spending limit, like opening a tab, and streams queries within that session. It supports hybrid fiat and crypto settlement, and carries Stripe’s built-in compliance, fraud detection, and tax reporting. MPP is designed for the agent that needs to make hundreds of queries per minute rather than one-off lookups.

These two protocols are not competitors. They are complementary layers. x402 handles the long tail of autonomous, per-request discovery and payment. MPP handles high-throughput commercial sessions. Together, they solve the micropayment problem that has held back machine commerce for decades.

The internet was designed with a payment layer that was never built. Now it exists. And stablecoins, which settle in milliseconds for fractions of a cent, are the reason it finally works.

The Ecosystem Is Already Live

This is not a whitepaper future. People are building and shipping right now.

Last week I sat in a room at the Solana Skyline with half a dozen teams presenting what they have built on top of x402. The infrastructure layer is real: Faremeter and Corbits have created the open-source middleware that makes any API agent-payable across Solana, Base, SKALE, and Monad. A developer can drop in a few lines of code and start accepting micropayments from AI agents.

The wallet layer is real: AgentCash, founded by three former Stripe engineers who built Stripe’s stablecoin financial accounts (YC W26), gives agents a USDC wallet and instant access to over 250 premium APIs with no per-provider sign-up. PaySponge provides agent wallets with granular spending controls including per-transaction limits, daily budgets, and domain allowlists.

The analytics layer is real: Merit Systems’ x402scan tracks the ecosystem in real time. As of this week, it shows roughly 57,000 daily transactions across 850 sellers and 1,100 buyers. This is a functioning marketplace with real volume, not a testnet demo.

The application layer is real: BlockRun.ai lets any agent with a wallet access 38 different LLM models through x402 micropayments. A smart routing engine analyzes each prompt and sends it to the cheapest capable model. No API keys, no subscriptions, no accounts.

And the major players are converging. Stripe supports both x402 and MPP natively. Coinbase contributed x402 to the Linux Foundation. Anthropic and OpenAI are MPP launch partners. When Visa, Mastercard, Google, AWS, and Cloudflare all put their weight behind the same protocol foundation, it tells you something about where things are heading.

But here is the observation that matters most: the ecosystem has payment rails, agent wallets, middleware, routing, and analytics. What it does not yet have, at any meaningful scale, is the real-world data that agents actually want to buy. The infrastructure is built. The question is what gets built on top of it.

The World This Creates

Imagine a service fielding ten million agent queries per day at a penny each. That is $100,000 in daily revenue from customers that never signed a contract, never sat through a demo, and never logged into a dashboard. Now consider what those queries look like.

A personal agent comparison-shops across every retailer in your zip code before you leave the house, paying a half cent per price lookup, and tells you exactly where to go and what to buy to stay under your grocery budget. You never open a browser.

A brand manager’s agent monitors competitor pricing across 30,000 zip codes daily. It detects that a rival brand just launched a 20% off promotion in the Southeast and alerts the team within hours, not weeks. The cost is pennies per day instead of a six-figure annual contract.

A quantitative trading fund’s agent ingests real-time brand velocity signals built from actual receipt volume and transaction data, not surveys or estimates. It detects that a public company’s sales are accelerating two weeks before the earnings call. The signal costs $0.025 per query.

An economist’s agent constructs a real-time consumer price index from millions of actual price observations, updated daily, broken down by income bracket and region. It reveals that lower-income households are experiencing grocery inflation at nearly double the rate of the national average. The monthly CPI report from the Bureau of Labor Statistics, published weeks later, feels like it’s arriving from another era.

A startup with three engineers and no sales team launches a data API on Monday. By Tuesday, agents are discovering it through a machine-readable manifest, paying per query via x402, and the startup has its first revenue without ever speaking to a single customer.

This is the shift:

From subscriptions to pay-per-request. From dashboards to APIs. From enterprise sales cycles to machine discovery. From projected panel data to real-time crowdsourced signals. From gatekept intelligence to open, permissionless access.

The consumer intelligence industry is a $31 billion market built on infrastructure designed for human buyers with annual budgets and quarterly review cycles. The buyers of the future are machines that make decisions in milliseconds and pay in microcents. The companies that rebuild for this reality will define the next era of the industry. The ones that don’t will join the long list of businesses that mistook their distribution model for a moat.

Conclusion

The forces converging here point in one direction: AI agents that need real-world data, payment protocols that enable micropayments at internet scale, stablecoins that make sub-cent transactions economical, and an incumbent industry structurally unable to adapt. Each of these forces is powerful on its own. Together, they make the outcome inevitable.

The consumer intelligence market will be rebuilt, and it will be rebuilt for machines.

HTTP status code 402 waited 29 years. The data waited behind enterprise paywalls. The agents are here now, with wallets in hand, ready to pay.

The question is not whether this happens. The question is who builds the data layer that agents actually want to buy.

Azad Neenan is the founder of Crush Rewards, building real-time consumer intelligence for the agentic economy.

There’s an ongoing debate about the most fitting metaphor to describe blockchain networks. While analogies with databases, programming languages, and operating systems such as iOS and Android are prevalent, one might wonder if these comparisons truly capture the essence of blockchain technology. Accurately conceptualizing this technology is paramount for a comprehensive understanding of its future trajectory.

In a recent episode of the Empire podcast, two prominent venture capitalists made comparisons between blockchain and the aforementioned technologies. Yet, such comparisons overlook a fundamental aspect unique to blockchains: their inherent ability to facilitate transactions of value, a feature not found in traditional technologies and what essentially adds a third dimension to blockchain’s capabilities. Unlike databases, programming languages, and operating systems, which cannot bypass the financial regulations imposed by central banks, blockchain technology uniquely empowers the creation of decentralized entities with the ability to operate independently of traditional monetary systems. This positions blockchains not merely as technological innovations but as autonomous entities akin to states within the United States — each possessing distinct attributes yet collectively committed to fostering a liberated and open ecosystem.

Analogies Explored

Before delving into this intriguing state analogy, it’s worth acknowledging the allure of drawing analogies with other technological forms. Yes, blockchain shares several features with these technologies, but its capacity to embody and transact value adds a significant dimension, distinguishing it profoundly.

Blockchain Networks vs. Databases:

Blockchains, akin to databases, structure and store data but are distinct in their inherent decentralization. They prioritize data integrity and permanence, often compromising on areas where databases might excel, such as performance or versatile querying capabilities.

Blockchain Networks vs. Programming Languages:

These networks resonate with programming languages in terms of specialization and community influence. Designed for specific applications like digital currencies or decentralized apps, blockchains, much like programming languages, evolve within their own communities and ecosystems. They face unique governance challenges and strive to address scalability and performance through innovative strategies.

Blockchain Networks vs. Mobile Operating Systems:

Blockchain networks and mobile operating systems both foster extensive ecosystems. Mobile OSs facilitate a variety of computing operations, whereas blockchains concentrate on secure, decentralized transactions. The control spectrum varies widely, from the centralized nature of iOS to the decentralization characteristic of public blockchains. While security is a top priority for both, their core purposes diverge — personal computing for mobile OSs and immutable, consensus-driven data recording for blockchains.

Despite these similarities, it’s crucial to recognize that blockchains amalgamate the structured approach of databases, the specialized nature of programming languages, and the ecosystem-centric ethos of mobile operating systems. Yet, they transcend these characteristics by their unique capability to hold and transact value, akin to states within a country.

The State Analogy

Chris Dixon eloquently parallels blockchain functions with urban planning in his recent book Read. Write. Own.. He likens starting a blockchain network to founding a new state, where initial structures are laid out, and a system of governance and incentives is established to encourage development and investment. Property rights, or ownership, play a pivotal role, ensuring that inhabitants and investors feel secure in their contributions. As the state — or, in this case, the blockchain network — expands, so does its economic base, leading to further growth and development. Tokens in a blockchain act as land grants or incentives, fostering ownership and investment, while take rates and DAOs function similarly to state taxes and governance, respectively.

This comprehensive analogy not only illustrates the functional similarities but also highlights the unique economic and governance dimensions that blockchain networks introduce, paving the way for emergent, bottom-up technologies and economies.

The cultural and community development that binds the citizens of a state finds its reflection in the blockchain space. States are not just administrative constructs; they are vibrant hubs of culture, tradition, and shared values. Blockchain networks cultivate similar communities, united by shared visions, interests, and governance philosophies. The sense of community is paramount in both realms, driving engagement, development, and a collective sense of identity.

Sense of community, like pride in your state or nation, is strong across crypto communities. One of the VCs on the podcast (Haseeb) brought up that some people will simply choose to buy NFTs on ETH because of the history and the culture, even if it means paying more. It’s true, there is no any way around this point — people do things that make them happy and bring them joy. People do and will continue choosing their communities based on personal preferences, even among ETH for example there are dozens of L2 and communities who stand by those ecosystems. There are other communities that are strong and provide their own trade-offs, such as the Cosmo app chain, Solana and other L1s. While people always have the option to move around should an enticing opportunity arise, they generally stay where they feel most comfortable — where their families and communities are. But that doesn’t mean newcomers can’t choose where to go from an unbiased perspective.

All blockchains are still open for immigration, so to speak. Some may have a large population today, but may not tomorrow or in 10 years — there is fierce competition among states to develop the best applications and technologies that attract people to them. The most successful ones will have established cultures and tech that will transcend their states and provide strong economic incentives to encourage immigration.

Abstraction And The Future

Outsiders who aren’t blockchain native will certainly want a piece of what’s been created inside the new world. No one cares what language the application is written in, people just want to do the thing. So wherever the action is happening, wherever the money is, the masses will flow down. Most people who will benefit from the innovation of crypto won’t be so mentally invested in it the same way the early adopters and enthusiasts have been. Some crypto fundamentalists may see this as a “hot take” because they believe in holding their own private keys so strongly.

However, most people simply don’t want nor have the time to be fully responsible for their own assets. In fact, many individuals who reap the benefits of cryptocurrency innovations may not engage with the technology directly at all. A few examples of how this already happens: an investor seeking substantial returns might allocate funds to complex financial instruments, such as exotic credit derivatives, at a sophisticated hedge fund; billions of people who benefit from tech companies but have no idea how the software works; foreign investors who’ve never stepped foot in the US getting yield from corporate bonds and treasuries. There are countless examples of such cases. And yes, most benefactors won’t ever use wallets themselves — some would even call them tourists. It’s inevitable that there will be 2nd, 3rd and even 4th order benefactors of the technology blockchains enable.

Crypto will be no different, but it will require substantial regulations to be passed. However, the moment those are established, fintechs and banks will rush in creating products that they can package and upsell to their customers. Customers will not care where the yield is coming from. It will all be abstracted away and the pipeline of capital from the “real world” will rush downstream directly into the economies of the blockchain states who are providing the most cutting edge technologies and services. One exciting example worth thinking about is structured products, which are currently only available to high net worth individuals with access to the top funds. This paradigm could be broken if tradfi companies can access the pipes to decentralized services like Ribbon Finance or Friktion (rip). All of a sudden anyone with a bank account can get access to sophisticated automated option strategies.

While it’s still early days, if things continue the way they’re going we will eventually see large profitable companies never have a traditional IPO because they’ve raised all they need with tokens instead of traditional stock equity. Those tokens can only be accessed on blockchain, not on NYSE. For the rest of society to benefit from the appreciation of their token there will need to be a bridge from tradfi fintech directly to the blockchain. Whether that’s through an ETF or some other mechanisms, it’s bound to happen with clearer regulation.

As blockchain technology continues to evolve, it invites a broad spectrum of participants — from the crypto-native enthusiasts to the traditional investors seeking new opportunities. The analogy of blockchain networks as states emphasizes their potential to redefine economic and social structures, paving the way for a future where decentralized and open systems prevail. As we look forward, the integration of blockchain into wider societal frameworks signifies a shift towards more inclusive, transparent, and equitable systems — a vision that, while still emerging, promises to reshape our understanding of technology, community, and governance in the digital age.

This essay will examine an emerging opportunity in nuclear power called small modular reactors (SMRs), and to understand the growing role they could play in the future. This is not a technical deep dive into the technology nor is it financial advice. The goal here is to zoom out to explore some of the new catalysts pushing the field forward. Hopefully through the synthesis of all this information you walk away with a better understanding of the landscape and the potential opportunities that lie ahead.

Intro

We live in an era where the quest for sustainable energy is converging with the explosive growth of digital infrastructure. SMRs have the potential to represent a critical juncture in our pursuit because of a confluence of factors including:

1. a large increase of forecasted energy consumption

2. a race to create and deploy AI software into every square inch of society and business

3. increasing geopolitical risks that can skyrocket traditional energy prices

4. self imposed targets to reach net zero carbon emissions

5. and a shifting positive sentiment from regulators and investors.

This analysis explores the interplay between these factors and dives into the commercial viability of SMRs and their application in powering AI data centers — a focal point of energy forecasts. Through the lens of Amazon’s groundbreaking integration of nuclear energy with its data operations and the pioneering steps of companies like NuScale, we will navigate the complexities, challenges, and transformative potential of this energy innovation.

Shifting Sentiment and Increased Funding

The CTO of Digital Realty, the largest data center real estate firm on the planet, recently said in an interview:

“Our industry has to find another source of power” He predicts that data centers in the not too distant future will come with their own dedicated, built-in nuclear reactors. “A normal data center needs 32 megawatts of power flowing into the building. For an AI data center it’s 80 megawatts,” said Mr Sharp.

He is referring to SMRs which are intended to generate between 20 to 300 megawatts of power in a baseload capacity (about 1/3 of traditional nuclear plants). These types of reactors offer a scalable, safer and quicker-to-build alternative to traditional nuclear power plants, promising to redefine energy paradigms. Traditional reactors use water for their processes, but advanced reactors like SMRs can use molten salt, liquid metals like sodium or lead, or gases like helium or carbon dioxide. These approaches allow them to operate at higher temperatures, and greater safety with higher efficiency rates and potentially less radioactive waste. It seems Mr. Sharp’s is not alone in his positive sentiment for nuclear, investors have also taken notice too:

In the public markets

• NuScale, a pioneer in SMR technology, has witnessed its stock price double, reflecting strong market confidence. This is further evidenced by support from major financial institutions like Vanguard Group Inc., Mirae Asset Global Investments, and BlackRock Inc. Additionally, Canaccord Genuity recently initiated coverage with a buy rating, acknowledging the company’s shift towards commercialization and its bright, profitable future.

• As the sole provider of small modular nuclear reactors with U.S. regulatory approval, NuScale stands uniquely positioned for commercial deployment, significantly ahead of competitors. Their reactors are already in production at a specialized manufacturing facility. The advanced design of their light water reactor, which has received significant regulatory approval, signifies the onset of a new nuclear power era, marked by efficiency and accessibility.

In the private markets

• Radiant Industries, a startup developing portable microreactors, generating between 1 and 20 megawatts of power aims to replace diesel generators. They recently raised $40 million in funding from some prolific investors including Andreessen Horowitz, Founder Fund, and Draper Associates, bringing their total capital raised to $54 million.

• Oklo: the partnership between AI innovator Sam Altman and SMR company Oklo underscores a key alliance essential for addressing the growing energy demands of the digital era. Altman, co-founder of OpenAI and a key investor in Oklo, demonstrating a deep commitment with his approximately 30% stake, strongly believes in SMR technology’s potential. Oklo aims to supply AI data centers with SMRs by 2028, actively tackling the power challenges of high-tech operations. This collaboration not only highlights the synergy between nuclear technology and AI but also heralds a future where data centers benefit from sustainable, abundant power.

Oklo’s head of business development recently said that:

“AI is the catalyst, the main driver, [Without us] they just don’t have the power to turn on all the machines they need to have. We are signing letters of intent on specific data center locations for the 2028 timeframe.”

Surely OpenAI, who is owned by Microsoft, will be one of Oklos first customers when they get clearance from regulators. As a leader in the space they could pave the path for more industry players to leverage SMRs to power their energy requirements.

Another indication of shifting sentiment around Nuclear is Amazon’s recent acquisition of Talen Energy’s Cumulus data center campus for $650 million, marking a significant milestone. This strategic initiative not only underscores Amazon Web Services’ (AWS) commitment to leveraging nuclear power but also sets an industry-wide precedent for sustainable digital infrastructure. And while it’s not an SMR it underscores the trend of using nuclear power to power data centers.

Economic and Environmental Considerations

Environmentally, SMRs represent a significant stride towards reducing carbon emissions, aligning with global efforts to combat climate change. However, the debate around nuclear energy’s environmental impact, particularly regarding waste management and safety, remains contentious. Critics, including organizations like Greenpeace, highlight challenges related to cost, safety, and waste, questioning the viability of SMRs compared to renewable energy sources. However, proponents of nuclear power, like Brian Gitt of Oklo, argue that the new generation of reactors addresses these concerns through advancements in fuel recycling, meltdown prevention, and self-regulation technologies. The debate over SMRs’ viability and competitiveness against renewable energy sources underscores the need for a balanced and informed discourse on our energy future.

“People question the viability of nuclear power due to the waste product and risk of an accident,” says Mr Gitt. “We recycle the fuel through our reactor multiple times, dealing with the waste, and the new reactors cannot melt-down, they are self-cooling and self-regulating.”

Critiques of nuclear energy, particularly from organizations like Greenpeace, have increasingly faced skepticism from both national governments and profit-driven companies. The unintended consequences of anti-nuclear lobbying became starkly evident in Germany’s energy crisis, exacerbated by the war in Ukraine. The decision to phase out nuclear power, hailed as a victory for environmental advocacy, quickly soured as energy costs soared. Faced with a shortage of reliable base load energy, Germany was compelled to reactivate coal plants, a move antithetical to environmental goals. This scenario underscored the indispensable role of nuclear energy in ensuring national security and highlighted the pitfalls of overlooking nuclear power’s reliability in favor of intermittent renewable sources like solar and wind.

This real-world example illustrates a broader dilemma in the pursuit of sustainable energy infrastructure. While environmental sustainability and cost are significant considerations, they are not the sole factors that nations must consider. Energy security and reliability emerge as paramount concerns, particularly in volatile geopolitical climates. Solar and wind, despite their environmental benefits, cannot yet offer the consistent base load power that nuclear energy provides.

The debate surrounding the economic viability and safety of small modular reactors (SMRs) further complicates the transition to more sustainable energy sources. Critics, including Dr. Doug Parr of Greenpeace, expresses skepticism regarding SMRs’ cost-effectiveness and safety, suggesting an over reliance on optimistic economic forecasts. But there are many reasons to doubt the skeptics.

What The Skeptics Have Missed

Proponents of SMRs believe that Wright’s Law, predicting cost reductions with increased production, along with economies of scale, will make SMRs competitive with renewable energy sources. However, achieving these cost benefits requires substantial investment in nuclear infrastructure. While commitments made at COP28 to triple nuclear output are promising, reliance on such agreements is risky due to geopolitical uncertainties. The demand for SMRs is likely to be driven by the private sector, particularly tech companies with growing clean-energy needs for AI data centers.

Historical criticisms of SMRs have focused on their cost and scalability, overlooking the energy demands of AI-driven data centers. The rapid development of AI in the private sector necessitates reliable, on-demand power, making the initial investment in SMRs more attractive for profitable tech firms. SMRs are expected to become cost-efficient with scale and provide a secure energy supply, enabling off-grid operations. This is particularly advantageous for data centers in colder climates, where natural cooling can reduce operational costs, which can constitute up to 40% of a data center’s energy consumption.

Recent events, like the outages experienced by Google Cloud and Oracle due to extreme temperatures, underscore the importance of climate resilience. According to the Uptime Institute, 45% of US data centers have faced weather-related operational challenges. As the climate continues to change, tech companies must invest in robust power and cooling technologies to maintain uptime. Notable examples include Amazon Web Services’ data center in the cooler Nordic region and Microsoft’s land acquisition in Sweden for future data center development. These moves highlight the industry’s shift towards leveraging colder climates for energy-efficient data center operations.

Ultimately the skeptics have discounted the private sector’s demand for energy in a world where data is the most valuable commodity. To drive this point home, let’s not forget that the CTO of the largest data center real estate company in the world is optimistic about a future where data centers have their own built in SMRs.

Conclusion

The converging trajectories of technological advancement, environmental imperatives, and economic realities paint a vivid picture of the future, where Small Modular Reactors (SMRs) can play a pivotal role in meeting the world’s burgeoning energy demands. As we stand on the cusp of a new decade, the amalgamation of factors such as technological maturity, regulatory approval, and shifting public sentiment towards nuclear energy indicates a fertile ground for the proliferation of SMRs. This expansion is not merely speculative but grounded in tangible developments and strategic investments from both public and private sectors. The critical role of SMRs in powering AI data centers — a sector with an insatiable appetite for clean, reliable, and scalable power — serves as a beacon for their broader applicability and potential. With significant advancements in safety, scalability, and environmental performance, SMRs are poised to become a cornerstone of a sustainable energy future. Given these dynamics, it’s not just probable but nearly certain that the next decade will witness a significant increase in the adoption of SMRs, marking a transformative period in our global energy landscape. This transition towards a more sustainable, efficient, and resilient energy infrastructure underscores the pivotal role of innovation in addressing the complex challenges of our time.

Thesis

Garmin is currently performing well, having established a robust market presence with their high-quality physical devices for tracking fitness and health statistics. They have been consistently growing their earnings while maintaining minimal debt on their balance sheet. However, there is significant untapped potential within the company. At present, their app’s user interface and user experience lack engagement, and they are not yet positioned to capitalize on the future of health and wellness, where personal devices will play a more crucial role in monitoring our health than traditional doctor visits. This shift is inevitable, driven by the convergence of skyrocketing healthcare costs in the U.S., the rise of artificial intelligence, and blockchain technologies that enable immutable records of GPS-tracked activities.

By adopting blockchain mechanisms, Garmin has the potential to boost user engagement and create new revenue streams by enabling users to monetize their health and fitness data. While some small blockchain startups like MoonWalk and Cudis, as well as non-blockchain companies like Fitrockr and IMS Health, are already exploring this space, Garmin is uniquely positioned to outperform existing solutions due to their substantial head start and strong distribution network. Unlike services such as Fitrockr, which must depend on device makers like Garmin for their business models to succeed, Garmin can vertically integrate these data services. By leveraging their own app, devices, and platforms such as Solana, Garmin can connect directly with users.

This represents a significant opportunity for Garmin to generate incremental revenue through the implementation of these data services, which incur minimal additional costs. As consumers become aware of the potential to earn money from their data, Garmin is poised to capture market share from its competitors.

Key Investment Highlights

• I propose two key new business lines for Garmin: Research Bounties and Data as a Service (DaaS). The Research Bounties initiative is projected to generate an additional $68 million in revenue in its first year, growing to over $400 million within five years. Meanwhile, the DaaS model is expected to add $2 million in its initial year, increasing to over $12 million after five years.

• For DaaS, I’ve intentionally used conservative estimates, considering numerous variables such as the number of devices, the volume of user-generated data, data pricing, and user participation rates. With further research, these estimates could be refined, revealing an even more promising financial outlook.

Company Description

Garmin Ltd. designs, develops, manufactures, markets, and distributes a range of wireless devices in the Americas, the Asia Pacific, Australian Continent, Europe, the Middle East, and Africa. Its Fitness segment offers running and multi-sport watches; cycling products; activity tracking and smartwatch devices; and fitness and cycling accessories. This segment also provides Garmin Connect and Garmin Connect Mobile, which are web and mobile platforms; and Connect IQ, an application development platform. While they have other business lines this is what we are focused on for the purpose of this exercise.

Garmin’s Blockchain Business Models

1) Research Bounties

Overview: Token-based rewards can be effortlessly integrated into Garmin’s ecosystem using USDC. With this system, the Garmin app on users’ phones transforms into a digital wallet, where data is anonymized and securely added to the blockchain. Researchers can establish bounties and challenges, and Garmin users can opt-in to complete these tasks, receiving payments upon successful verification. The Solana blockchain offers several secure options for maintaining privacy, such as token extensions, Light Protocol, and Elusiv, ensuring data remains visible only to the bounty creator. At the core of this system is the Proof of GPS, which ensures the authenticity of location-based activities.

This approach gamifies health and fitness data by offering cryptocurrency rewards. Users earn tokens for completing tasks and challenges, which they can redeem for USDC or another chosen cryptocurrency. Smart contracts provide automatic verification using data from wearable devices, ensuring an efficient and secure process.

Engagement Outcomes:

• The implementation of blockchain-based incentives in fitness has demonstrated significant benefits, including an expanded user base, heightened user engagement and retention, and a more rewarding exercise experience through enhanced gamification. Additionally, the transparent nature of blockchain technology offers better tracking of progress and distribution of rewards, making the entire system more appealing and trustworthy for users.

Monetization Opportunities for Garmin

Market Size Analysis: The total global pharmaceutical R&D expenditure is approximately $300 billion, with U.S. federal health research funding contributing $50 billion. Currently, Pfizer, which represents about 5% of the market, spends $12 million annually on anonymized data, suggesting total industry spending of around $240 million. IMS Health (IQVIA), a key player in medical data trading, reports revenues of $3 billion, highlighting the significant market for data monetization.

Market Opportunity:

I have delineated two scenarios for Garmin Bounties. In the optimistic case, where 3% of global R&D spending is channeled through Garmin Bounties, the platform could experience an annual bounty volume of $9 billion. The base case scenario assumes half of this figure, resulting in $4.5 billion of annual bounty volume.

Garmin Revenue Model:

• Garmin’s revenue model includes a flat 5% platform fee on transaction volume.

The potential annual fee revenue is as follows:

• Base Case: $225 million

• Bull Case: $450 million

• Average: $337.5 million

These estimates are grounded in historical R&D spending trends, with projections ramping to $469 million in five years.

Key Insights:

• There is a significant addressable market within pharmaceutical and healthcare research.

• Successful precedents for data monetization exist, as evidenced by IMS Health and Fitrockr.

• A conservative estimate of capturing 2-3% of total R&D spending underlines a realistic and substantial market penetration potential.

• The proposed platform fee structure is consistent with industry standards.

This represents a major opportunity in the healthcare data marketplace, offering potential platform revenues in the hundreds of millions annually, even with modest market penetration.

2) Data as a Service

Overview

Garmin users can effortlessly opt in to share their data and receive compensation determined by market rates. While the market ultimately dictates the compensation rate, the quality and frequency of the data play crucial roles in influencing pricing. Potential consumers on the demand side include research institutions, insurance companies, healthcare providers, pharmaceutical companies, and wellness companies. Each of these entities values high-quality, frequent data for various applications, making it a lucrative opportunity for users to monetize their fitness and health insights.

Data Quality Factors:

• Frequency of data collection

• Consistency of device wear/usage

• Completeness of health metrics (heart rate, steps, sleep, etc.)

• User demographic information

• Additional health context (nutrition, medical history)

These factors are evaluated and weighted to create a composite score for each user, which then determines the pricing of the user’s data per byte or gigabyte. Detailed explanations and calculations for this scoring system are provided in the Blockchain Business Models tab of the spreadsheet, offering a transparent view of how data valuation is derived and adjusted based on user engagement and data quality.

Pricing

In this model, pricing is dynamic and closely linked to user quality scores. For instance, users with a quality score between 0.4 and 0.6 receive a base rate of $38 per GB. However, those who achieve a score above 0.8 benefit from a 1.5X multiplier, earning $57 per GB. Currently, the forecast assumes a conservative average quality score of 0.6, leaving room for potential upward adjustment if users consistently score higher.

User Income

In my estimation, users will generate an average of 40 MB of data annually. Based on the described quality score and pricing parameters, users can expect to earn approximately $1 per year after Garmin’s fees. While the methodology is solid, there is room to refine these assumptions, as $1 does not seem to accurately reflect the true value of user data. Although the value might not equate to hundreds or thousands per year, it is likely higher than the current estimate.

User income is a critical factor because it can drive additional device sales for Garmin. If users perceive that, akin to Hivemapper or Helium devices, they can recoup their investment in a Garmin device within just a few months, Garmin is positioned to expand its market share ahead of competitors.

Monetization Opportunities for Garmin

Garmin’s model is designed to allow the majority of data revenue to go directly to users, while charging a fee for providing the platform. I’ve determined that a 90/10 revenue split in favor of the user is a fair rate. Though the immediate earnings for users may seem modest, the success of the platform could lead to increased device sales. If users recognize the value of earning from their data, this can drive incremental unit sales and help Garmin expand its market presence.

Recommendation

Garmin stands poised to transform its existing fitness device business into a data monetization powerhouse. By capitalizing on its robust distribution network and extensive user base, Garmin can introduce two promising revenue streams: Research Bounties and Data as a Service (DaaS). The Research Bounties platform has the potential to generate up to $450 million annually through a 5% fee on research funding transactions. Meanwhile, DaaS offers additional revenue opportunities through the monetization of user data at minimal incremental costs.

This innovative, blockchain-enabled strategy not only opens new revenue channels but also has the potential to boost device sales by allowing users to earn from their health data. With the global pharmaceutical and healthcare research sectors spending billions on data acquisition, Garmin is uniquely positioned to capture a significant market share. This integrated ecosystem promises to deliver value to both researchers, who benefit from high-quality, anonymized data, and users, who are incentivized through compensation. Garmin’s distinctive approach may well redefine its role in the health and fitness industry, leading to substantial growth and sustained relevance.

This is long overdue, but a lot has changed since I wrote my initial thesis on Palantir. For one, the stock has gone parabolic, and the Wall St cabal has finally caught on. Luckily for them, the investment thesis has only become more robust in light of recent changes in the geopolitical and economic environment. Amongst the biggest change in my original thesis — which was that they were solidly positioned as the operating system for the DoD and Intelligence Services — is that now they have a robust and diverse commercial business with potential customers lining up down the block to get access to their enterprise software. Palantir is rapidly becoming the default operating system not only of intelligence agencies but also of the US private sector. And their products are sticky, not dissimilar from Apple’s iOS ecosystem.

My updated thesis follows as such: Palantir Technologies stands at the intersection of critical national trends in manufacturing resurgence, technological advancement, and defense modernization. As tariff policies and an “America First” approach drive substantial domestic manufacturing expansion, Palantir’s advanced AI and data analytics platforms position the company to become an essential partner for both government and commercial entities navigating this pivotal moment in time. In this memo we’ll explore how Palantir’s unique technological capabilities, expanding customer base, and strategic market positioning create a compelling investment case in the context of America’s industrial revitalization.

Summary

The United States is experiencing a significant manufacturing renaissance, with President Trump’s administration implementing policies intended to revitalize America’s industrial base from its multi-decade decline. People today have largely forgotten how vulnerable COVID left us. We learned the hard way how reliant on the global supply chain really were, from everyday essentials to critical minerals and chemicals for pharmacology and everything in between. Suffice it to say, this created logistical nightmares during the shutdown and underscored potential national security threats. Fast forward back to the present day, President Trump is claiming to reinvigorate industrialization in the US, and he’s using tariffs as the main incentive mechanism. This policy environment is accelerating substantial investments in domestic manufacturing. Noteable examples include: Taiwan Semiconductor Manufacturing Company’s (TSMC) historic $165 billion commitment to U.S. operations, and defense contractor Anduril Industries’ $900 million manufacturing facility in Ohio that will create 4,000 jobs by 2035.

Palantir Technologies has strategically positioned itself to capitalize on this manufacturing resurgence through its advanced data analytics platforms and AI solutions. The company’s Warp Speed manufacturing operating system, described as the “manufacturing OS for American re-industrialization,” is already being deployed by leading manufacturers like Anduril Industries, L3Harris, Panasonic Energy, and Shield AI, and many many more. With its proven track record of delivering transformative results to clients like Airbus, where it increased A350 aircraft delivery by 33%, Palantir is uniquely positioned to support America’s industrial transformation and deliver substantial investor returns.

Company Overview

Founded in 2003, Palantir Technologies has evolved into a leading big data analytics and AI solutions provider. The company operates two primary platforms: Gotham, designed for defense and intelligence applications, and Foundry, which serves commercial enterprises across manufacturing, healthcare, finance, and energy sectors. These platforms integrate disparate data sources to create actionable insights, optimize operations, and enable more effective decision-making.

Palantir’s client base spans both government and commercial sectors, with an increasing focus on manufacturing and industrial applications. In 2024, the company expanded its customer base by 43% to 711 customers, up from 497 in 2023. This diverse client portfolio includes defense agencies, weapons manufacturers like Anduril, aviation giants like Airbus, and a growing roster of U.S. manufacturers embracing digital transformation. The company’s recent introduction of the Artificial Intelligence Platform (AIP) has further strengthened its market position.

*Note: I am long PLTR, and have been since 2023. This is not investment advice.

Abstract

This paper explores the development of an “Agentic Market Twin,” a multi-agent simulation designed to predict market movements by modeling human psychology and societal behavior. Converging insights from AI agent networks, behavioral economics, and Douglas Hofstadter’s “golden braid” philosophy, the system aims to create a digital society of diverse AI agents. These agents, representing various human decision-making archetypes, interact within a simulated environment, with their collective behavior and decisions aggregated through a Futarchy-based mechanism. The goal is to achieve more accurate market consensus predictions by capturing the complex, often irrational, dynamics that drive real-world financial markets. This paper outlines the conceptual framework, the design of the agentic society, the Futarchy-based decision-making process, and the proposed technical implementation.

Introduction

The Challenge of Market Prediction — Predicting financial market movements remains one of the most complex and persistent challenges in economics and finance. Traditional models often rely on assumptions of rational actors and efficient markets, yet real-world markets are frequently driven by a confluence of logical analysis, emotional reactions, cognitive biases, and complex social dynamics. Capturing this intricate interplay is crucial for developing more robust predictive systems.

The Agentic Market Twin Concept — This paper proposes the “Agentic Market Twin,” a system designed to address the limitations of traditional market prediction by creating a digital society that mirrors human market behavior through multi-agent simulation. The core idea is to build a swarm of AI agents, each modeled on distinct aspects of human psychology, decision-making styles, and societal interactions. By simulating their collective behavior, the Agentic Market Twin aims to predict emergent market consensus and potential movements with greater fidelity than methods that overlook these nuanced human factors.

Conceptual Foundations: Hofstadter’s Golden Braid — The philosophical underpinning of the Agentic Market Twin is heavily influenced by Douglas Hofstadter’s “golden braid” concept, which emphasizes the interconnectedness and recursive nature of complex systems. Three key principles from Hofstadter’s work are central to this framework:

• Strange Loops: This refers to hierarchical systems that, despite their layered structure, loop back and affect themselves. In the context of markets, this is seen in the recursive relationship where market sentiment affects agent behavior, which in turn impacts market prices and fundamentals, further influencing sentiment. Agents’ perceptions can alter market realities, not just reflect them.

• Level-Mixing: This principle describes how phenomena at one level of a system can emerge from interactions at a lower level, and how these emergent properties can then influence the lower levels. For the Agentic Market Twin, collective market intelligence and sentiment are emergent properties arising from the decisions and interactions of individual agents.

• Isomorphism: This involves a mapping or correspondence in form or structure between different systems. Here, it refers to the deliberate mapping of human behavioral archetypes, cognitive biases, and psychological traits onto the digital AI agents, aiming to create a simulation that is structurally analogous to human social dynamics within markets.

These principles guide the design of the system, ensuring that the agents and their interactions create a cohesive and dynamic simulation capable of reflecting the self-referential and emergent nature of real-world markets.

Overview of Agent Archetypes and Behavioral Economics — The agents within the simulation are not homogenous. They span a statistical distribution of decision-making styles, informed by behavioral economics, which acknowledges that human decisions are often influenced by cognitive biases and heuristics rather than pure rationality. The system will feature distinct agent archetypes (e.g., Rational, Emotional, Independent), each programmed with specific biases, risk preferences, and information processing styles. This diversity is crucial for simulating the rich tapestry of human behavior that contributes to market dynamics.

Paper Structure — This paper is organized as follows:

• Part 1: Designing the Agentic Society details the different agent archetypes, their behavioral characteristics, and the system dynamics that govern their interactions.

• Part 2: Collective Decision-Making with Futarchy explains how the agent society arrives at collective decisions and predictions using a Futarchy-based mechanism.

• Part 3: Technical Implementation Framework outlines the proposed AI and blockchain architecture for building and deploying the Agentic Market Twin. Finally, the paper concludes with a summary of the vision, potential impacts, and future research directions.

Part 1: Designing the Agentic Society

The efficacy of the Agentic Market Twin hinges on the realistic design of its constituent AI agents and the dynamics of their interactions. This section details the foundational agent archetypes, the principles ensuring population realism, and the simulated societal interactions.

Agent Archetypes and Behavioral Characteristics — Standard economic models often assume rational agents. However, to mirror human markets, our agents must embody a spectrum of behaviors, including irrationality and cognitive biases. These biases — such as overconfidence, anchoring, loss aversion, herding, and confirmation bias — significantly influence decision-making and can lead to market anomalies. The intensity of these biases can vary based on market conditions or past experiences. We define three primary agent archetypes, noting that a full implementation would involve many nuanced personas within each.

Rational Agents

These agents primarily employ quantitative models, focusing on fundamentals and statistical analysis. They represent entities like large money managers, hedge funds, and institutions, forming a significant portion of market liquidity.

• Cognitive Biases: While more analytical, they can exhibit biases like overconfidence in their models or anchoring to specific data points.

• Data Sources & Tools: Financial statements, economic indicators, statistical models, bond market data, SEC filings, DCF modeling, classical technical indicators (Bollinger Bands, moving averages), trading algorithms.

Sub-types (Examples):

• Data-Driven Analysts (Institutional): Focus on comprehensive data analysis.

• Quantitative Strategists (Systematic): Employ algorithmic trading strategies.

• Value Investors (Conservative): Prioritize capital protection and fundamental security analysis.

• Technical Traders (Professional): Focus on pattern recognition, momentum, and price action; often sophisticated and systematic in risk management.

Emotional Agents

These agents are driven more by impulse, sentiment, and social trends, often exhibiting higher elasticity to short-term volatility. They primarily represent retail investors.

• Cognitive Biases: Highly susceptible to herding behavior, loss aversion (leading to panic selling or holding losing positions too long), and FOMO (Fear Of Missing Out).

• Data Sources & Tools: News headlines, recommendations from financial pundits, social media trends (e.g., WallStreetBets, Twitter/X), influencer opinions.

Sub-types (Examples):

• Retail (Mom and Pop): Often follow trends late, sometimes serving as exit liquidity for more sophisticated participants.

• Retail (Risk-Seeking/Degenerate): Younger, thrill-seeking investors, often drawn to high-volatility assets like meme stocks or certain cryptocurrencies.

Independent Agents

These agents adopt unconventional approaches, seeking opportunities in emerging technologies, niche markets, or by identifying market dislocations. They are often thesis-driven and possess a mindset closer to venture capitalists.

• Cognitive Biases: Can exhibit confirmation bias (seeking data that supports their unique theses) but may also be less prone to herding. Their ambition can sometimes lead to overestimating the potential of novel ideas.

• Data Sources & Tools: R&D spending, patent filings, trend analysis of nascent technologies, analysis of innovators and disruptors (e.g., CEO personalities, company culture), heavy use of LLMs for qualitative judgments.

• Characteristics: Problem identifiers, contrarian sentiment, tolerance for risk if conviction is high, focus on power-law return distributions.

Sub-types (Examples):

• Disruptive Investors (Early Adopters): Focus on groundbreaking technologies.

• Thesis Developers (Research-Focused): Build unique investment theses based on deep research.

• Trend Pioneers (Pattern Seekers): Identify and act on emerging macro or niche trends before mainstream adoption.

Ensuring Population Realism: Diversity, Adaptability, and Individual Goals

To create a truly representative digital society, several factors beyond basic archetypes must be considered:

• Intra-Archetype Diversity: Within each category (Rational, Emotional, Independent), thousands of fine-tuned personas with unique behavioral nuances, motivations, political alignments, and levels of optimism/pessimism are required.

• Dynamic Behavioral Models: Agents must adapt their decision-making over time, reflecting learning from experience, changing market conditions, or significant market events. This includes the evolution of their biases and heuristics.

• Risk Preferences and Time Horizons: The system must account for varying risk tolerances (from highly conservative to risk-seeking), diverse investment horizons (short-term speculation to long-term investment), and individual financial goals.

System Dynamics: Simulating Societal Interactions

The defined agents interact within a simulated environment, governed by dynamics designed to reflect real-world societal and market interactions.

• Information Flow and Asymmetry: Information cascades through the system. Top-tier agents (e.g., sophisticated Rational Agents) may have access to data first, which then disseminates through social networks and media channels to other agent types. Bottom-tier agents might aggregate into crowd behavior based on delayed or filtered information. Market narratives emerge from this collective interpretation.

• Coalition Formation and Power Dynamics: Agents can form temporary alliances or trading blocs based on shared interests, themes, or sectors. Counter-coalitions may develop to exploit perceived inefficiencies. Power dynamics shift based on agent performance, capital accumulation, and influence. Wealth distribution, access to information, trading frequency, and risk tolerance contribute to a hierarchy, where certain agents or coalitions wield more influence at different times.

• Communication Networks and Emergent Narratives: Agents communicate through simulated channels, propagating signals like price movements, volume indicators, sentiment markers, technical patterns, and fundamental metrics. Network effects mean information spreads through these networks, with influence potentially weighted by agent credibility. Echo chambers can form around specific investment themes, but cross-pollination of ideas between different agent classes also occurs.

• Manifestation of Hofstadter’s Principles in Agent Interactions: The system dynamics are designed to embody Hofstadter’s principles organically:

• Strange Loops are evident as agents’ perceptions and predictions influence their behavior, which impacts the simulated market, feeding back into their perceptions. Collective beliefs can shape market realities (e.g., bubbles).

• Level-Mixing occurs as individual agent decisions and interactions aggregate to form emergent market phenomena like sentiment shifts, trend formations, and price discovery. These macro trends then influence individual agent decisions.

• Isomorphism is maintained by ensuring the agent behaviors and interaction patterns consistently map to their human psychological counterparts.

These interactions, incorporating elements of behavioral game theory (strategic interaction, belief hierarchies), allow for sophisticated decision-making where the simulation reflects the strategic and psychological nature of trading. Emergent properties such as self-organizing market regimes, spontaneous trend formation, fear/greed cycles, and adaptive risk assessment are expected.

Part 2: Collective Decision-Making with Futarchy

Having designed the agentic society and its internal dynamics, this section explores how these diverse agents arrive at collective decisions and market predictions, primarily through the implementation of Futarchy.

Introduction to Futarchy for Market Prediction

Futarchy is a governance model where decisions are made, or policies are chosen, based on the outcomes predicted by speculative markets. Participants in these markets bet on which proposed actions or states will lead to the most favorable, predefined outcome. The action or state whose associated prediction market indicates the highest probability of success (or highest utility) is then adopted. In the context of the Agentic Market Twin, “best outcome” translates to the most accurate prediction of real-world market movements or the valuation of assets. This model inherently incentivizes accurate predictions through potential profits or losses for the participating agents. The Solana-based protocol MetaDAO is a key example of a platform enabling such Futarchy mechanisms.

Integrating Futarchy with the Agentic Market Twin Philosophy

Futarchy serves as a powerful mechanism to aggregate the complex interactions and diverse perspectives of the agent society, operationalizing and enhancing the core philosophical underpinnings of the system.

• Amplifying Strange Loops through Strategic Betting: Futarchy markets create a direct feedback loop. Rational agents might model emotional agents’ likely reactions when placing bets. Emotional agents, in turn, might react to the price signals emerging from these prediction markets, which are influenced by rational and independent agents. Independent agents can exploit these predictable patterns. This creates a recursive loop of strategy, counter-strategy, and price discovery, where agents’ bets directly influence the collective prediction, which then informs further actions.

• Achieving Level-Mixing via Prediction Market Aggregation: Individual agent biases, beliefs, and analytical insights (micro-level) are aggregated into market prices within the Futarchy markets (macro-level). These prices represent a collective intelligence, a weighted consensus reflecting the conviction and capital of diverse participants. Emergent phenomena like market bubbles or panic selling, driven by aggregated behavioral cascades (e.g., widespread loss aversion or herding), can be captured and quantified through these prediction market dynamics.

• Reinforcing Isomorphism in Market Psychology: The Futarchy mechanism allows the digital agents to express their mapped human cognitive biases and psychological traits through their betting behavior. For example: 1) Rational agents might exhibit overconfidence by placing large bets based on their models. 2) Emotional agents might contribute to momentum or herding within the prediction markets. 3) Independent agents might place contrarian bets based on their unique theses. The resulting market prices in Futarchy thus reflect an aggregation of these diverse psychological factors, mirroring how human market psychology shapes asset prices.

Market Consensus Formation and Incentivizing Accuracy

The agent swarm uses Futarchy to make collective predictions by:

1. Creating Prediction Markets: For various possible market outcomes or asset valuations, agents (primarily Rational and Independent types, though Emotional agents’ sentiment can be modeled by others) can propose markets or participate by betting.

2. Agent Participation: Agents bet based on their unique decision-making styles, information access, and risk profiles.

3. Signal Extraction: The market prices within these Futarchy markets serve as signals of collective belief. The outcome with the highest implied probability (derived from market prices) is considered the system’s consensus prediction.

Economic Incentives: Futarchy inherently incentivizes accuracy. Agents are financially rewarded (within the simulation’s economy) for correct predictions and penalized for incorrect ones. This motivates agents to:

• Carefully analyze available information.

• Refine their predictive models and heuristics.

• Attempt to account for and even counteract their own biases if it leads to better outcomes. This “natural selection” of strategies means that, over time, agents and strategies that are more successful at prediction gain more influence (e.g., by accumulating more capital to bet with).

By combining incentivized accuracy through Futarchy, diverse agent behaviors, advanced simulation features, real-time data processing, and feedback loops, the system effectively reproduces conditions and behaviors analogous to real-world financial markets. Predictions derived from this system are more likely to mimic actual market outcomes because they are generated through processes that replicate how market participants interact, react, and influence those outcomes.

Practical Scenarios: Futarchy in Action

1. Equity Valuation Cascade

• Setup: Agents create/participate in a prediction market on Tesla’s future market share in autonomous vehicles.

• Agent Participation: Independent Agents analyze technological patents and progress. Rational Agents place bets based on fundamental valuations and competitive analysis. Emotional Agents’ potential market impact (e.g., retail buying/selling based on news) is factored into the bets of more sophisticated agents.

• Outcome: The Futarchy market price reflects a nuanced consensus on Tesla’s prospects, potentially revealing insights before they are obvious in the actual stock market. A feedback loop can emerge where the prediction market outcome influences simulated trading strategies for Tesla stock.

2. Crypto Winter Predictor

• Setup: Detecting early signs of a crypto market downturn using Futarchy markets on, for example, the solvency of major platforms or the likelihood of significant regulatory action.

• Agent Participation: Data-Analyst Rational Agents spot warning signs in on-chain data or exchange reserves and bet accordingly. Independent Agents might create markets predicting specific platform failures. Emotional Agents’ panic selling potential is a factor considered by others.

• Outcome: The Futarchy market could provide early warnings of systemic risk, allowing the system to “predict” a downturn or periods of high volatility.

3. Rate Hike Anticipator

• Setup: Anticipating Federal Reserve interest rate changes via prediction markets on the probability of specific rate hike percentages by a certain date.

• Agent Participation: Rational Agents analyze economic indicators (inflation, employment). Technical Traders identify market patterns reacting to speculation. Emotional Agents’ sentiment shifts in response to news are modeled.

• Outcome: The Futarchy market aggregates these diverse inputs, potentially positioning the system ahead of actual market movements related to interest rate changes.

Part 3: Technical Implementation Framework

The practical realization of the Agentic Market Twin requires a sophisticated dual architecture combining AI for agent intelligence and interaction, and blockchain technology for transparent and incentive-aligned decision-making and simulated trading.

Overview of the Dual AI and Blockchain Architecture

• AI Layer: Responsible for creating, fine-tuning, and managing the diverse swarm of AI agents. This includes their personality development, cognitive bias embedding, decision-making frameworks, and the simulation of their complex interactions within virtual environments.

• Blockchain Layer: Provides the infrastructure for the Futarchy mechanism (e.g., via MetaDAO on Solana), enabling agents to propose and bet on prediction markets. It also facilitates simulated trading of assets, management of agent “wallets” and internal tokens, and transparent recording of key decisions and transactions.

Blockchain Layer: Wallets, Tokens, and On-Chain Interactions

Cryptocurrency and blockchain technology are integral for several reasons:

• Futarchy Implementation: Protocols like MetaDAO on Solana offer the tools to create and settle prediction markets, which are central to the decision-making process.

• Agent Economy: Agents will possess wallets and utilize a native system token. This token can be used for internal transactions, such as certain agents selling specialized information or analytical insights to others, simulating a service economy and disincentivizing spam.

• While basic information might be widely available, access to sophisticated data or analyses from specialized bots could be monetized.

• This internal currency could have a liquidity pool on a Solana protocol to enable swapping with other tokens (e.g., USDC, SOL).

• Initial wealth distribution programmed into bots can be reflected in their token holdings, influencing their betting capacity in Futarchy markets.

• Simulated Trading: For agents authorized to “trade” based on predictions, Solana’s ecosystem (e.g., Jupiter for price discovery/swaps, Drift for derivatives) offers cheap, fast, and reliable platforms. Real-World Asset (RWA) protocols (e.g., Ondo) could simulate bond investments, and platforms like Parcl could simulate real estate investments.

• Data Feeds: Integration with oracle services like Pyth Network is crucial for providing real-time market data (that can originate anywhere like from the US census, or FED data) and price information to the agents and the Futarchy markets.

AI Layer: Building and Managing the Agent Swarm

This involves two main components: the development of individual agent personalities and the system architecture enabling their collective function.

• Agent Development: LLM Fine-Tuning for Personalities and Biases The foundation is creating distinct agent personalities through specialized Large Language Model (LLM) fine-tuning, as outlined by the archetypes in Part 1.

• Foundation Model Architecture: Base LLMs must possess strong financial comprehension, numerical reasoning, pattern recognition, and the ability to process both quantitative data and qualitative sentiment. Training data includes historical market data annotated with trader behaviors, sentiment-labeled financial news, trading psychology case studies, behavioral finance research, and real trader decision logs.

Specialized Training Frameworks:

• Rational Agents — Focus on technical/fundamental analysis, quantitative modeling, mathematical finance.

• Emotional Agents — Trained on social media sentiment, retail trading patterns, forum discussions to understand market psychology and crowd behavior.

• Independent Agents — Specialized in disruptive technology research, contrarian analysis, identifying emerging trends.

• Cognitive Architecture Implementation: This involves calibrating cognitive biases (e.g., overconfidence via prediction scoring, anchoring via historical price references, loss aversion via asymmetric risk-reward weightings) and social influence parameters. Risk profiles (conservative to aggressive) are mapped with specific behavioral characteristics.

• Decision-Making Framework: Integrates information processing weights, risk assessment matrices, time horizon preferences, position sizing algorithms, and entry/exit triggers consistent with agent personality. Behavioral modulation systems allow agents to adjust to market stress or confidence fluctuations.

• Training Methodology: A sequential approach: 1) Establish financial knowledge. 2) Embed personality traits. 3) Integrate cognitive biases. 4) Establish decision patterns. 5) Develop interaction behaviors. Reinforcement learning with reward/penalty functions ensures consistent personas while allowing adaptation. This aims to create thousands of distinct, evolving agents.

System Architecture: Enabling Complex Agent Interactions and Information Flow

A hierarchical structure mirroring real-world financial markets is needed.

• Core Architectural Framework: A centralized agent initialization hub deploys pre-trained LLM agents into virtual environments based on classification/access levels. A data processing engine orchestrates information flow (raw feeds, news, indicators) through collection, cleaning, normalization, and hierarchical distribution, creating natural information asymmetry.

• Communication and Decision Architecture: Hierarchical communication network (high-priority channels for institutional-level agents, mid-level for independents, broadcast for market-wide, private for coalitions, emergency alerts). Agents use a multi-layered decision process: information filtering -> analysis (applying biases/preferences) -> decision generation -> execution.

• Virtual Space Implementation: Specialized interaction spaces (simulated “rooms” where fine-tuned models interact): primary trading floors, research rooms, coalition spaces, analysis chambers, emergency rooms.

• Operational Dynamics: Information flows, is processed by agents per their models, generating decisions/reactions that feed back, creating loops. Agent interaction protocols manage relationships, access control, and logging for integrity/optimization.

• System Control and Management: Load balancers, access controllers, performance monitors (for persona consistency), resource managers, and emergency protocols ensure stability and efficiency. This framework supports emergent market phenomena from individual agent interactions.

Voting Initiation: Technical Triggers for Futarchy Proposals

A multi-layered threshold approach determines when a topic/question moves to a Futarchy market:

• Consensus Detection: Monitor conversation density, number of agents engaged, interaction intensity, urgency signals from high-influence agents.

• Quorum Requirements: Minimum threshold of interested agents, representation across types, presence of specific types based on subject, verified information distribution.

Voting Trigger Conditions:

• Information Saturation: Sufficient data processed/distributed, key agents analyzed.

• Coalition Formation: Clear opposing viewpoints/agent groups emerged and stabilized.

• Market Conditions: Time-sensitive opportunities/threats, regime changes, significant external events, risk levels at decision thresholds. Once triggers are met, proposals are generated on-chain in Futarchy. After settlement, outcomes are fed back off-chain as new information for agents, continuing the reflexive loops.

Conclusion

Recapitulation of the Vision and Approach

The Agentic Market Twin project aims to create a digital society that mirrors the complex, often irrational, dynamics of human market behavior. By developing diverse AI agents with distinct personalities, cognitive biases, and sophisticated interaction mechanisms, and by integrating a Futarchy-based governance model for collective decision-making, this system offers a novel approach to market forecasting. The technical implementation leverages both advanced AI for agent modeling and blockchain technology for transparent, incentive-driven prediction markets and simulated trading.

Potential Impact and Contributions

This system has the potential to:

• Provide more accurate and nuanced market predictions by capturing behavioral factors often overlooked by traditional models.

• Offer a “sandbox” for studying market psychology, the formation of bubbles and crashes, and the impact of different information cascades.

• Enhance understanding of collective intelligence and emergent behavior in complex socio-economic systems. The primary contribution lies in the synergistic integration of Hofstadter’s principles, behavioral economics, multi-agent AI simulation, and Futarchy.

Future Research and Development Directions

Significant development work remains, particularly in the AI implementation phase, including the continuous refinement of agent models, scaling the simulation, and validating its predictive accuracy against real-world market data. Further research could explore:

• The evolution of agent strategies and biases over long periods.

• The impact of introducing novel types of information or external shocks into the system.

• Ethical considerations and the potential for misuse if such predictive technology becomes highly accurate.

The journey to create a true digital twin of human market dynamics is ambitious, but the potential to revolutionize our understanding and prediction of market behavior is profound.

References:

• Hanson, R. (2000-2007), various papers on Futarchy

• Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263-291.

• Hofstadter, D. R. (1979). Godel, Escher, Bach: an Eternal Golden Braid

• MetaDAO, Solana protocol