Mind Crunches #32: Sensemaking The World
AI Governors, Stablecoins and Robert Moses
📌 Mind Crunches
On AI Running the US Government: you must probably already know what “vibe coding” is and if you don’t, you should. Satya is casually vibe coding in the latest Microsoft’s 50 year celebration posts. It seems that we recently witnessed the first-ever occasion of “vibe governing”. Rohit recently tweeted (and then it was somehow confirmed by various media sources) that Trump’s team used LLMs to calculate the exact formula of tariffs announced last week. If that’s the case, then this may be an inflection point where more governments, politicians, policymakers will use GenAI to decide on important affairs, make geopolitical moves, shape policies. I have been thinking about this a lot and:
surprisingly, I feel perfectly fine that governments/policymakers use AI to inform their decision making. I suppose that the "AIs are running our government" scenario seems much more intimidating when you read it in a report VS when it actually happens. It definitely shows how much progress we have made the past 4 years in considering AI an essential part of our lives.
this is another example of how important prompting skills are. it's super simple. poor prompts=poor results. smart prompt=smart results. a good analogy I use for prompting is conversational skills. Conversations are in a way a form of prompting. Compare the quality of answers that someone like Tyler Cowen gets in his podcasts VS Lex Fridman. You get the point.
vibe governing is technically an "LLM as an Oracle" use case. This is a challenge that folks in Crypto x AI have faced for quite some time now. Which LLM should you trust to decide on your policy? Is your LLM grounded in real facts about the assumptions it takes? If yes, can you verify it? Can you use a kind of "swarm intelligence" to combine the ingenuity of LLMs? There are some really inspiring startups working in these areas like Space & Time and Genlayer.
if we end up to an equilibrium where AI Governing is a real thing and widely adopted, then I expect prediction markets to become another important tool because they can be a perfect pair to LLMs. If you aren’t familiar with prediction markets, you can read this piece but in a nutshell prediction markets are platforms where people bet on future events, with prices reflecting the collective probability of outcomes. In AI-governed societies, they could serve as decentralized, incentive-aligned tools for forecasting the consequences of policy decisions—helping AI systems make better-informed, democratic, and empirically grounded choices. Robin Hanson’s futarchy is a very interesting framework if you really want to go down the rabbit hole of prediction markets + governance.
On AI biology & the upcoming AGI effects: it’s increasingly clear that when it comes to the pace of AI innovation and the intrinsic nature of LLMs, we don’t know a lot. I am fascinated by the latest Anthropic research because it actually proves many things that we would consider improbable a few weeks ago. I highly recommend you reading both reports, but my tldr is:
there seems to be a Universal Grammar (a la Chomsky) baked into Claude
Claude can make up fake CoT to convince the user for the validity of its answer
Combinatorial thinking in LLMs is more advanced than we think
Claude has meta-cognition that allows it to judge if can answer a question or not
Even as I am typing these words, I can’t grasp how crazy this is and how it completely shifts our mindset on how LLMs actually behave. With that said, predicting the next years of AI is almost certainly doomed to fail. There are just too many intertwined variables (compute, geopolitics, AI talent, economics etc) that can redirect the trajectory of history in different directions. But there seem to be indicative signs of where the economy and “world order” may be heading. I recently read two completely different essays/reports on the future of AI which I found very interesting and informing despite the fact that I don’t necessarily agree with their predictions:
The AI 2027 report is written by some of the smartest AI researchers and best predictions analysts out there and to be honest I was very concerned about the end of their story. Good news: these folks are obsessed with AGI so it’s likely they view the world only from this angle, hence the focus of their report on superhuman intelligence and Cold War type of game theory. Bad news: these folks are both really smart and super knowledgeable on AI so if there is someone who could make somehow accurate predictions about AI, it’s them.
Dean Ball’s essay is much more narrowly focused on AI agents but it also draws some broader conclusions. He emphasizes that AI agents will significantly impact knowledge work by automating complex tasks, leading to substantial economic and organizational changes. Bonus note: I am surprised that neither of these essays explores the AI future of robotics and how this can influence both industrialization and war capabilities.
On “Regulation Is Life”
One of the main themes of this blog is how much our daily life is being shaped by regulation. This was something that I couldn’t see when I was younger but it has now become a personal “mental model”. There are so many things that we do daily and we think of them as outcomes of cultural preferences and personal choices but they are pretty much consequences of regulation. Works In Progress (one of my favorite magazines) has a great essay on why the shape, height, density of buildings in America, China and Europe were shaped by regulation rather than culture. “Modern urban form is not always, and perhaps not even usually, an emanation of deep cultural preferences. Changing it does not necessarily mean changing the hearts of millions of people. To build very differently would not take overthrowing twenty centuries of urban culture. It would merely mean tweaking these obscure and somewhat arbitrary documents.”
On a similar note, I am currently reading “The Power Broker”, the 1974 biography of Robert Moses by Robert Caro (and supposedly one of the best biographies ever written). While on the surface the story is about Moses’ rise to power, the book reveals a deeper truth: the profound and often invisible ways regulation and bureaucracy shape our everyday lives. Through Moses’s story, Caro shows how rules—written and unwritten—about funding, zoning, planning, and public authority can have massive consequences for where we live, how we commute, who has access to public space, and whose neighborhoods get destroyed or preserved. One of the book’s central insights is that regulation isn’t neutral. It can entrench inequality or open up opportunity, depending on who controls it. Moses used the machinery of regulation to accumulate power over decades, even without holding elected office. His projects redefined the geography of New York, shaped suburban sprawl, and helped cement the car-centric urban design that dominates much of America today. In essence, The Power Broker is a cautionary tale about how seemingly boring mechanisms of governance—permits, funding formulas, planning boards—can be wielded to reshape lives on a massive scale, often without the public fully realizing who is pulling the strings.
On the Inevitability of Stablecoins: I have previously written about why stablecoins are a net better version of money and how they could potentially be a form of “glocal” capital that overcomes tariffs. Last week, Circle filed a public prospectus with the SEC in one of its first steps to going public and the US House Financial Services Committee passed a stablecoin framework bill, the STABLE Act, which will now head to a full House vote. It’s evident that the US will soon have a clear and regulatory framework about stablecoins and I am convinced that their adoption rates will be very strong. If you aren’t familiar with this new form of money, stablecoins, such as Circle's USD Coin (USDC), unlike traditional fiat currencies, which are subject to fluctuating exchange rates and international tariffs, maintain a consistent value by pegging to stable assets like the U.S. dollar. This stability enables businesses and individuals to conduct international trade more efficiently, circumventing the financial barriers imposed by tariffs and currency conversion fees. The next big fight will be over the interest that stablecoins carry and whether this should pass to their holders (which of course should be the case). Bonus note: stablecoins is a booming industry. here are just a few of the startups serving the Stablecoin stack (cards, APIs, middleware, wallets etc)
Happy birthday, Microsoft! On to the next 50. I consider Microsoft a very important part of my personal and professional life. It’s the company that helped me grow the most, gave me the opportunity to move to the US, witness the growth of GenAI first hand but it also showed me how large organizations almost naturally tend to complacency and succumb to bureaucracy. I will always be amazed by Microsoft’s ability to rapidly commercialize and “package” software into chunks that drive revenue and its almost unimaginable turnaround by Satya. Most people forget (or underestimate) how difficult it is for tech companies to stay relevant for 50 years. Last, I consider the meetings I had the chance to attend next to Satya to be the highlights of my professional life. I extracted so much substance by Satya’s style (how he was being prepared for a meeting, the questions he asked, how he asked them, the questions he did NOT ask etc) that I will always feel grateful. A longer post about my journey at Microsoft will definitely come later this year!
I recently read “The ASML Way” which I highly recommend especially if you are European. There is very limited bibliography on successful European tech companies and this book offers some great insights of both the benefits and disadvantages that a European tech leader faces in a global market. I will probably write a long post about the book and my conclusions but one of the most surprising thing was how “chaotic” the production of lithography machines is compared to the super lean and efficient model that Musk has popularized in Space X.
I wrote about the brilliance of MCP in my previous post. The excellent Steven Sinofsky writes why MCP may get stuck in the valley of failed middleware software.
You must have probably already Ghiblified yourself. I know I did! Taking aside the completely irrelevant and shallow concerns about Ghibli studio not benefiting from this trend, Erik Hoel has an interesting contrarian take that the cheap production of art is rapidly driving the world to a semantic apocalypse where everything is striped of meaning.
The story behind Arxiv which is pretty much the global database of science.
There have been so many books, movies and theories about crime in NYC, why it comes back, how it comes down etc. This interview with Peter Moskos is a fascinating read because it provides unfiltered insights about some of the decisions made in the ‘90s. Highly recommended!
“Perception is reality for a lot of these things, because most people aren't victimized by crime. But when people perceive that no one is in control they feel less safe. It's not that this perception is false, it just might not be directly related to an actual criminal act.”
Recommended Blog: Hyperdimensional by Dean Ball.
Recommended Book: King Dollar: The Past and Future of the World's Dominant Currency
Recommended Podcast: Ezra Klein on Tyler Cowen about Abundance
Quote: “Cognition goes all the way down to the molecular level” Michael Levin
(Perplexity artwork by the excellent Mac Baconai)