Transcription
Since the days of Ian Nasir and his famous complaints about copper, the world has been built on arbitrage. It's been built on the cost of good X at port Y and how much you can sell it for down the camel caravan route for thousands of years. That's what we've been doing.
AI is changing all that. AI is reinventing the art of arbitrage and it is one of the underlying drivers of the economy that we need to talk about because it is changing everything. It's underneath labor. It's underneath tech. The fact that arbitrage is shifting is one of the foundational shifts in our entire economy that AI is unlocking that as far as I know nobody is talking about. I've looked who's talking about arbitrage.
Now, let's be really blunt here. Arbitrage is the art of getting rid of inefficiency. Most of the world runs on inefficiency, right? Not brokenness, not stupidity, inefficiency. Gaps between what something costs to produce and what the market is willing to pay for it. Those gaps are not bugs. They're the structure of the market. Right? Entire industries, career paths, entire business models exist because some inefficiency was just too expensive, too invisible to close.
So the law firm that bills 10 hours for work that takes two hours of thinking and eight hours of doing, that's not a scam. It's that the business model is built on the historical cost of legal research or the consulting engagement where the client is going to spend hundreds and hundreds of thousands of dollars on what amounts to a series of board decks. Really, what they're getting there is access to information that previously they couldn't get and the fact that it takes a while to produce it that's always been tolerated that's part of the arbitrage. Uh, similarly offshore development teams, right? They exist because a San Francisco engineer costs X and a Bangalore engineer costs Y. These gaps have been the water that the economy swims in, right? Most people don't see them, we don't think about them, they're everywhere.
AI is changing how all of that works. AI is closing those gaps, not slowly the way previous technologies did, not incrementally over decades because it took a long time to build railroads out and reduce the cost of time to get goods to market, which is a great example from the past. AI is closing these arbitrage gaps on the time scale of model releases in months, sometimes in weeks, and every time one closes, three new ones open up somewhere else. That's the dynamic that we're not talking about clearly. And I think it's the single most important thing if you want to understand the relationship between labor and capital in the age of AI.
I know this sounds abstract, so let me make it concrete for you. This is a true story. In late 2025, a bot on the prediction market PolyMarket turned $313 into $414,000 in a single month. It had a 98% win rate across 6,600 some trades. And it did not predict anything. All the bot did was exploit a simple fact. PolyMarket's short-duration crypto contracts updated their prices much slower than the spot exchanges where the underlying assets traded. So when Bitcoin moved very sharply on Binance, sharply enough to make the outcome of a 15-minute contract very nearly certain, PolyMarket was still showing roughly 50/50 odds. So the bot bought the mispriced side of the market over and over and over again while humans slept.
A developer reverse-engineered the strategy and claimed to have rebuilt a working version in Rust using Claude in just 40 minutes. So the full stack, right-time price monitoring is a big piece of that. Uh, the probability calculation, position sizing, automated risk controls, and all of that. So, the claim, the developer was generated from a single prompt session. What previously required a quantitative research team, software engineers, and risk managers now requires one person with a laptop and an API key.
And that bot is not alone, right? A separate Claude-powered system generated $2.2 million in 2 months using probability models trained on news and social data. A swarm model trained on three years of NBA data reportedly generated $1.49 million trading sports contracts. Comparative data showed bots using identical strategies to human traders. It's not that the strategy was better, it's that the execution was almost completely flawless. No fatigue at 3:00 AM. No oversized positions on confident bets. No missed trades during lunch. Humans knew what to do, but we haven't been able to do it consistently because, well, we're humans. The bots can do it for us.
I'm not telling you this to shill a get-rich-quick scheme, right? This is not all about crypto and how amazing it is. I'm telling you this because PolyMarket is one of the places where you see the mechanism that I'm saying is underneath the whole economy with perfect clarity. The data is on the chain. The trades are public. The compression of arbitrage is measurable. Literally, average arbitrage windows shrank from 12.3 seconds in 2024 to 2.7 seconds on PolyMarket in early 2026. You can literally watch the inefficiency in the market closing in real time.
That same mechanism, AI identifying a gap, building the system to exploit it, compressing the window until only the most sophisticated players survive. It's happening in every industry. You just can't see it as clearly because most industries don't happen to publish their pricing lags in public.
So, you might wonder, Nate, what's actionable here? How can I build off of this? I want to give you the taxonomy of where these arbitrage windows are closing all over the economy. So, you can pay attention. Whether you're a builder, whether you're aspiring to be a builder, whether you're a business leader, it doesn't matter. You got to pay attention here because it's going to shape how all of our careers work. Once you have this lens, once you can look for inefficiency that AI makes newly exploitable, you start seeing it absolutely everywhere.
So number one, look for speed gaps. One system updates slower than reality. This is the PolyMarket case, right? It's the easiest to understand. The bot reacted faster than the market could repric. But the same structure exists in any business where information propagates through intermediaries. Your competitor's pricing model. It updates in real time while yours updates weekly. Their customer support bot resolves issues in seconds when your team takes 24 hours. Their hiring pipeline screens candidates in minutes and yours takes weeks. Every one of these is a speed gap. And each one is now closable by whoever builds the faster system first.
Here's another one. Reasoning gaps, not speed interpretation. A Fed governor makes a statement. A regulatory filing drops. An earnings call reveals a change in strategy for a major company. The information is public. It's available to everyone simultaneously. The gap is how quickly and accurately someone can reason about what it means. Can you update your model of the world and act on the new probability? LLMs do this incredibly fast and more consistently than humans. Not because they're smarter, but because they don't get tired, they don't get distracted. They don't go for lunch. So, they can get the full context of a doc in a couple of minutes or a couple of seconds. One PolyMarket bot generated $2.2 million in two months using ensemble probability models trained on news and social data. Not by having information others didn't have access to. This was not insider trading, but just by interpreting public information faster and acting on the interpretation before the crowd caught up. In your industry, the analog is every decision that waits for someone to sit on it and read it and synthesize it and recommend. That wait time is a reasoning gap and it's closing.
Here's another one. Fragmentation gaps. The same thing is priced differently in different places because nobody's looking at all of those places at once. Sports arbitrage bots will scan PolyMarket bets and they're going to lock in arbitrage against traditional bookmakers and then lock in margins by buying both sides when the combined price implies a mathematical edge. That's the financial version, right? The business version is the consultant who charges so much for an analysis that synthesizes five publicly available data sources. The value is not in the data, it's in the aggregation. And AI now does that aggregation for free. So the intermediary whose value proposition is "I can see the silos you can't" is sitting on a fragmentation gap. And that gap is compressing fast because one thing LLMs are really good at is pulling information from disparate sources and putting them together. If the deep research report you can pull is better than your Big Four consultant, I got news for you, right? You're not alone. There's a lot of people who are discovering that everything was information silos before and now information silos are disappearing.
Here's another one. Discipline gaps. The inefficiency here is not in the market nor is it in the information. It's actually in the human executing on the information. So comparative data from PolyMarket shows that bots using identical strategies to human traders captured roughly twice the profit. Right? It's not because the strategy was different. It's because they were perfect on their positioning sizes. They had no emotional overrides, no fatigue, no missed trades, and they just made correct decisions over and over and over again, which is a lot of how you capture alpha in a market. That discipline gap is everywhere in business. The sales team that knows the playbook but does not follow it consistently. The content pipeline that produces erratic quality depending on who's working that day. The operations team that drifts from the protocol when the pressure is on. Anywhere human performance degrades under fatigue, AI doesn't just close the gap by replacing the human. It closes the gap by enforcing a consistency that the human cannot maintain alone.
Knowledge asymmetry gap. That's the last big one I want to call out. And this is a huge one. This is the macro layer. For 30 years, the dominant gap in the global economy was a labor pricing gap. Same work, different costs depending on geography. So offshore teams existed because San Francisco was expensive. AI now replaces labor arbitrage with intelligence arbitrage. The unit of value shifts from the person-hour to the outcome. And this is a really tricky one because it's, it's dependent on the ability of people to use this tool. So one prompt from the right person can generate a working system that scales extremely efficiently. But if it's in the hands of the wrong person, one prompt can generate a completely broken system. And that means that the company that produces a deliverable in three hours and its competitor is taking three weeks to produce that deliverable. The real value they have is the people who can use the AI tools that enable them to take advantage of that intelligence leverage, that intelligence arbitrage. In other words, the intelligence arbitrage is a function of your best people's ability to use cutting-edge models and consistently grow with those models. That is one of the most valuable gaps in the world right now.
A recent piece drew the parallel to CNC lathes in the 1980s. When computer-controlled machining arrived, a shop owner could buy one CNC lathe, hire an operator at 40% of a master machinist's wage, and produce precision parts in 45 minutes that used to take 10 hours of hand-milling. And the smart shops hid their machines in the back room and kept the machinist out front for clients. They charged the old rate for work done at the new cost. The margin for a while was staggering. Then everyone got CNC machines. Prices collapsed 60 to 80%. The bespoke premium evaporated because everyone realized it wasn't bespoke anymore.
This is the exact same arc playing out right now in every knowledge work industry. The agencies, consulting, and service firms that are currently using AI to produce deliverables at a fraction of the old cost and trying to claim it's bespoke. That's not going to last. You've got to actually produce real value to a company. You can't just produce thoughtfulness for a company, right? Like that, that whole idea that the future is going to be built by thought leaders. The future is built by builders in the age of AI because the future is not dependent on labor and labor arbitrage anymore. The future is dependent on intelligence arbitrage. And any company can get their best people inside the house now and have an intelligence edge over anybody else. And that is why there is such a fierce competition for the top 1% of AI talent. The top 1% of AI talent can write their ticket because they are the ones that can best leverage this intelligence arbitrage which is the new gold currency of the future economy.
But there's a darker side to the story. I've talked a lot about the edge. I've talked a lot about PolyMarket. One of the things that is true that we don't like to talk about is that just because AI tools are now available to everyone, just because they are democratized does not mean that the edge that they give is spread widely. In fact, you would think when I talk about this, when there's a million little linkbait posts about this, that everyone would be on PolyMarket making money. But it turns out the arbitrage is being whittled away and 94, 95% of PolyMarket wallets lose money. They're feeding the successful traders. In that world, one of the things that is hard to swallow and hard to recognize is that the availability of AI tooling does not equal success or meaningful outcome.
Look, in business, everyone has access to Claude now if they want. Not everyone has reorganized their workflow or their decision-making or their feedback loops or their quality systems around what Claude makes possible for you. So, the executive who deploys a chatbot and calls it transformation is like the PolyMarket guy copy-pasting the same prompt. The consultant that uses AI to write faster but doesn't change what the consultant does. All they're doing is an unsophisticated implementation that the market is going to eat for lunch because of intelligence arbitrage. So the gap here isn't "has AI" versus "does not have AI." That gap has closed. The gap that matters is whether you bolted AI onto your existing process wrong or whether you rebuilt the process around what AI makes possible. Correct. That is the new inefficiency and it's the one that separates the few percentage points of people and companies that are succeeding right now from the ones who are all like up to their necks and trying to figure out how to swim in the world of AI.
Now, this is the part of the story that almost nobody talks about, but it might be the most important. The conventional framing treats AI disruption as a one-time event. It is like a meteor that hits the dinosaurs. Technology arrives, disruption happens, new equilibrium settles in. That is incorrect. What is happening is a continuous rotation of exploitable arbitrage gaps, each opened by a new capability step and each one compressing on a shorter timeline than the last.
We got a really vivid illustration of this in just the last week. On March 27th, a configuration error in Anthropic's content management system accidentally exposed draft materials about a model called Claude Mythos. Anthropic confirmed it exists, described it as a step change in performance, and said it's the most capable we've built to date. Leaked drafts indicated the model dramatically outperforms current models on reasoning, on coding, on cybersecurity, with one draft describing Mythos as "currently far ahead of any other AI model in cyber capabilities" and warning, "it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." Right? Scary. Markets did not wait for the model to ship to react. The software sector ETF fell 3% on the rumor of a leak. Bitcoin tumbled back from $70,000 because of the risk on cybersecurity. What if Bitcoin gets hacked? Cybersecurity stocks dropped. Prices moved on the knowledge that the model exists before anyone outside a tiny, tiny early access group even touched it.
Now think about what happens when this model becomes available. Every existing AI system gets repriced. The PolyMarket bots running on current Claude become the slow horse all of a sudden overnight. Someone running the same strategy on a model with meaningfully better reasoning has a temporary edge. The edge lasts until everybody upgrades and then the window compresses again. Right? Maybe the two-year compression goes down to, uh, from 12.3 seconds to 2.7 seconds to to a half a second. Right? With a step-change model like we're talking about with LLMs, my point is that the arbitrage window shifts overnight and it shifts in ways that are hard to predict in advance.
So the compression by itself is not the important part here. The rotation is the important part. So Mythos's reported strength in cybersecurity creates an entirely new gap that didn't exist before. Right? The distance between organizations that have hardened their defenses against Mythos-class capabilities and those that don't yet know the threat model has changed. Defensive security firms that got early access will have a window, a real monetizable edge in protection calibrated to what this particular new model can do. And that window will only last until that defensive tooling is widely deployed and then it will close and a new one will open somewhere else.
Similarly, Mythos's improvements in reasoning will create new capability thresholds. Agentic workflows that could not reliably handle complex multi-step tasks on current models may well work on Mythos and that unlocks new opportunities for automation. It creates new gaps between early adopters and everybody else. And all of that compresses as the capability rapidly becomes table stakes. And Anthropic is not alone on launching models like this. OpenAI reportedly finished pre-training its own next-generation model the same week as the Mythos leak from Claude. Sam Altman told employees that "things are moving faster than many of us expected." Both companies are racing toward IPOs, potentially late this year, which means the cadence of capability releases is about to accelerate. You thought it was fast already? It's going to get faster. Meanwhile, Google, Meta, and a half a dozen other labs are on similar timelines. Every single release in this market is a perturbation, a change. Every change opens new gaps across multiple domains at the same time. And every set of gaps is compressing faster than the last one because the adoption infrastructure is improving every cycle.
So think about the velocity. In 2024, a major model release happened every few months. In 2025, releases were roughly quarterly and absorption compressed into just a couple of months. And in 2026, we're watching markets repric within hours of a leaked draft of a model that isn't available. And the major labs are shipping every single day. So the cycle time between "new capability exists" and "the market has priced it in" is absolutely collapsing.
Which means the old mental model that has sustained our economy for thousands of years where you have a disruption, you have a transition, and you have an equilibrium. That model of slow, stable arbitrage change is fundamentally broken. There is no equilibrium. There is only the next rotation of the model. The world doesn't settle into a post-AI steady state. It enters a permanent condition of rolling disruption where the specific inefficiencies that define your industry, your role, your competitive position are reshuffled with every significant model release.
If this sounds tiring, I get it. I'm also tired sometimes. So the question then becomes, how do you keep your sanity? How do you see what changes next? If you accept the frame that the world is built on slowly exploited inefficiencies on arbitrage and AI will collapse them on the time scale of model releases and every collapse opens up new opportunities, then the question becomes, how do you have a practical lens to read the future? This is the biggest question in business right now.
If you ask yourself what's underneath that question, you get three root cause issues you need to understand. Number one, for any industry, any role, any business model, ask yourself this: What inefficiency is this built on? Every business model rests on a gap. Information asymmetry, execution difficulty, aggregation complexity, whatever it is, name the gap. If you can't name it, you're not going to see it closing. And you won't see it closing until someone else has built a system over the top of it. This is true, by the way, for careers as much as it's true for business models. I'll give you a specific example. Do you know the arbitrage gap that product management was built on? It was built on the fact that engineers did not want to take meetings and were considered too valuable to be in meetings. That was the arbitrage gap that founded the entire career of product management. Now, it's changed. It's evolved over the years, but that's where it started. That's the inefficiency it started on. Is that getting reshaped in an age when we don't need meetings as much and we have smaller, leaner teams? Absolutely.
Number two, the second question you should ask yourself in the age of AI arbitrage that's dawning here. How fast can AI close that gap? Some gaps are really structural, even in the age of AI, and they're going to stick around for a long time. I'll give you an example. Regulatory modes, they're not going away anytime soon. Relationship-dependent trust, physical world logistics, moving atoms around, being in the atoms business, genuine creative taste, hard-won domain judgment. A lot of the others beyond those are very much informational or cognitive, and those are closing very, very fast, like on a time scale of quarters when they used to take decades to close. Let's be concrete. Let's say you're a law firm versus you're a surgeon. A surgeon's judgment, that's not going to close as fast as the law firm's ability to do research and bill for that research. The law firm is going to be in more trouble. Another example, right? An agency's ability to control its production costs and bill for production costs at full freight is going to collapse faster than a therapist's ability to charge for the empathy that they provide. The insurance company's actuarial analysis gap is going to close fast because you just don't need as many people to do actuarial analysis as you used to. And the negotiator's relationship equity is going to stick around because a negotiator, whether they're a real estate negotiator or whether they're doing deals other places, that's a relationship job and it's hard to arbitrage that out. So be honest about which kind of gap you're sitting on and whether it's a structurally stable gap or not.
And then question three, ask yourself, what new gap does the closure create? This is where the opportunity lives, right? And it's the question almost nobody's asking. Every time AI closes one inefficiency, it's going to create some adjacent ones. When AI collapses the cost of producing content, the gap shifts to distribution and taste. Anyone can produce content, but not everybody can reach an audience. Not everybody can curate quality. When AI collapses the cost of code generation, the gap shifts to system design. It shifts to integration. Anyone can generate functions, but not everyone can architect systems that are going to work reliably at scale in the age of AI. When AI collapses the cost of legal research, the gap shifts toward judgment, toward client trust. The research gets commoditized down, but the legal counsel does not.
So here's the pattern. The new gap is always upstream of the old one. Closer to judgment, closer to taste, closer to relationships, closer to systems-level thinking. It's further from production. It's further from execution. It's further from information retrieval. This is the migration path that's stable. In a world of change, this is what is steady and it's really predictable. Once you see the pattern, you can trace where value is headed in your own industry even before it gets there.
Let me give you a concrete example. Right now, a junior financial analyst job is roughly 70% data gathering and formatting and 20% analysis and like 10% judgment calls. AI is collapsing that 70% of the job towards zero. The naive conclusion is that you need fewer analysts. The better conclusion is that the analyst role in its entirety is migrating upstream. The same person, freed from gathering and formatting, can now spend 60% of their time on analysis and 40% on judgment. And the gap shifts from "who can compile my data?" which is the lower-level question, to "who can interpret the data in context and make a defensible recommendation?" That is a harder gap to close because it requires domain knowledge. It requires institutional context and it requires the kind of integrative reasoning that develops those upstream skills and it requires the kind of integrative reasoning that current models don't do very well and they're not necessarily gaining in that capability at a super high rate. The analyst who recognizes this migration and develops those upstream skills like judgment, like communication, like contextual reasoning, really, really, really fast and deliberately, they're positioning themselves for the new gap. The one who's just using AI to compile the data faster is in trouble.
Now, you can argue that some junior analysts are just not going to do that. And I would agree with you. This is very much a case where the revolution is made of individual people and many individual people are not going to make that jump. One of the things that I want to call out is that that window to make the jump on your own voluntarily is not going to be there forever. At some point, companies are going to decide to cut bait. And you may not agree with their decision, but at some point, they're going to say, "The junior analyst that has not made progress or career growth in the last three years, we're not going to stick around with them. We would rather just stick with our ones that are growing." Please, please, please pay attention to how fast your peers are growing in your role because if you are not near the top of the pack, you are at risk.
Okay. What are your takeaways here? If you are running an organization, I want to be really blunt with you. You are living in a world of collapsing arbitrage. Your company has existed maybe for a long time on arbitrage inefficiencies that took decades to close. All of that is going away. And it is worth it for you to name the arbitrage you're building your business model on. Name whether it's structural and sticking around or not. And then name how you are going to take advantage of the new arbitrage opportunities that are opening up in the age of AI. There is no other path forward. And if you aren't thinking in terms of arbitrage, you're going to get blindsided. Think about whether you're building toward specific edges that are going to compress with the next model release or think about whether you're building toward edges that are structural. The difference is a really big deal. If you chase the last window that was open, you're always going to be behind for the next window. If you think about the next window and think about whether it's structural, you're going to be in a better place.
If you're an individual contributor, the intelligence gap that I talked about is the most relevant one to your career. Right now, the difference between what an AI-augmented professional can produce and what a non-augmented professional can produce is absolutely enormous relative to what the market pays for either of them. Most salaries and most freelance rates still reflect pre-AI productivity assumptions. If you can do in three hours what used to take you 30, you are capturing surplus. The market is paying you for 30 hours of value whether it knows it or not. The question for you at this point, are you going to be like the machinist that took advantage of that short-term win in the front of the shop and just smiled and greeted the customers and tried to like make it look like you were hand-delivering all of this stuff when in fact you were automating it with AI and all of that advantage is going to go away in a year or two? Or are you going to take advantage of this moment and move from being the machinist in front of the shop to being the person that knows how to make the machines? Because that's really where the market is starting to go, right? If you want long-term durable value for your career, you can't just pretend like you're delivering hand-rolled data in a world where it's all AI. You have to assume that the market is going to start to price those skills as commodities and move up the value chain to something that's more durable.
Please, please, please take this moment seriously. You have a chance to take advantage of intelligence arbitrage and show that you are someone who can architect intelligence systems that deliver outcomes. That is where the value is going to lie for labor.
Ultimately, the world has been built on slowly exploited inefficiencies for thousands of years. That's why I started this video talking about Ian Nasir and camel caravans. That's how old this is. The "slowly" part is over. What replaces it isn't efficiency. It's a faster cycle of inefficiency creation and destruction. We're not talking about a perfectly efficient, flat market. It's actually the opposite. It's a very turbulent market, but it's micro-turbulent. All these inefficiencies are created and they're closed very, very quickly, and you have to look underneath that structure to see the larger trend lines, which is what I tried to do with this video. I want you to see where the inefficiencies are moving and migrating you toward. That is why we talk about taste. That is why we talk about judgment. We are talking about things where the market is essentially slowly chasing people because it's closing up these inefficiencies all around them. The PolyMarket bot is just the clearest example of this. The Mythos leak is a preview of what is coming next. The only losing move in this market is to assume that where you are standing is steady state. It's not steady state and you need to plan accordingly. Plan for a world where we have rapid opening of arbitrage opportunities, rapid closing of old arbitrage opportunities. And we need to assume that we are looking for stable structural gaps in the market to build businesses around and to build careers around. And if you're not thinking that way, you are frankly going to get "arbitraged out." And that is the world we're all living in. Best of luck. Find those durable gaps and make sure that you are using AI to build into the intelligence disruption, not to be built out by it. Because effectively, in a lot of cases, I see people who are being built out by intelligence disruption. All of the intelligence that's getting built is crowding them away, and they're not seeing this strategy, right? They're not seeing that fundamentally there are a few gaps that are structural that AI is not closing, and they're not jumping onto those islands. They're just trying to sort of run faster and pretend they do AI and price like they always did. And that's like being the machinist with the CNC shop. That's just not going to last. And everything's going to collapse on that side because it's going to get "arbitraged out." Take arbitrage seriously if you, if you haven't gotten excited about it. It's a fundamental force in the economy. It's how a lot of value gets created. Understand how it works. Understand how it works in the age of AI. And best of luck out there. Cheers.