I want to lay out what we see, how we think about it, and what it means for where we deploy capital.

Waking up from a 60-year stupor

For the past 60 years, we have delegated virtually everything to a single, all-powerful entity: the nation state. Technological development, education, healthcare, security, money, and oftentimes thinking itself. We outsourced all of it with the internalized belief that we could stop thinking about it, and we could now take it for granted as it would keep working. Well, it’s not working anymore. Nation states are now drowning in liabilities they will never be able to pay off, running on infrastructure they can no longer maintain, and regulating technologies they fundamentally do not understand. This letter is about people: people noticing this even if in a subconscious way, and waking up (in very different ways).

On one end of the spectrum, normal people are realizing the ground is shifting and they are getting both scared and angry. We fully expect the general population to become massively adversarial to fundamental technologies like AI, robotics and genetic engineering. As an industry, I’m afraid we are 1) not taking this seriously enough and 2) not developing good enough answers to the very reasonable concerns that people have. Most people care first about their own welfare in the short term, with the global advancement of technology being very low on their list of priorities (contrary to yours truly, as here we believe the two things are very much correlated).

We posit that the political consequences of this reaction are not priced in. Governments will try to arouse and leverage this fear to engineer even more state control, precisely as their ability to deliver on promises evaporates.

They will instigate the Butlerian Jihad of the 2000s, and the world will be very much different based on its outcome.

But on the very opposite end, and our end if we must say, something much more exciting is happening: a real, tangible awakening. People are starting to realize the limits of the nation state’s ability to be the best and cheapest provider of all these services, and they are concluding that they might have to take them on themselves if we want to keep the quality of life we’ve grown accustomed to. We see young founders everywhere (US, EU, India, China, Japan, Korea) embracing a completely new way of building companies, one defined by a set of words that has so far completely eluded the tech sector: greatness, audacity, urgency and consequence.

Gone is the time of the quick app to make a quick buck. It is the time of founders building companies because the world would be intolerable without them and it would be a waste of their lives not to build them. Because no one is coming to help. And the state certainly won’t have the power, the incentives, the will nor the ability to do so. The State was not designed to face great challenges. It’s a “state” of administrators, middle managers, policy redactors and implementers. There are no creators, no visionaries, no makers.The failures of modern megabureaucracies are a feature, not a bug. They were meant to put down genius to make society work together, they are normalization machines engineered to subdue the daring, humiliate the exception, shame the one who transcends.

This is making founders rebel and rediscover the all-too-important virtues of volitions and agency. Two of our very favorites at Unruly. But we might be getting into overused territory with these terms, which actually don’t even capture the full meaning of what we’re seeing: entelechy. Entelechy is the life force that is needed to bring something to its most blessed state. It is what moves us, what guides, the driving force of people and organizations.

We’re collectively rediscovering the striving to tap into the entelechy that is in us.

Ensuring entelechy has now become humanity’s sovereign imperative, and we’re here for it.

An energy-bound world

The world has always been energy-bound.

Everything we have today, we only have because we found a viscous, sticky black substance in the ground, and it completely changed our lives. Without oil, we’d still be in the Industrial Revolution era (which was in turn enabled by coal). As AI and robotics bring the cost of labor and intelligence down dramatically, the cost of products and services will finally trend towards their actual energy cost (plus real estate costs, especially in food and mining).

As a concrete example of the world in a few decades we can think of say your water boiler breaking. Today you pay for a plumber’s time, expertise, travel, the parts, and obviously their profit. Tomorrow, a robot would be dispatched, take a scan, 3D print the replacement part, fix it, test it, and would then return to base.

The cost of the whole thing would be limited to the energy needed by that robot to operate and travel, plus the raw materials. What remains is energy. This is the transition we are investing through: from a labor-bound and intelligence-bound economy to an energy-bound and materials-bound economy. The main goal of the world will be to bring down the cost of energy. Progress is reducing the overheads of the physical world to an asymptote of the cost of energy.

And this is also why software, while it will obviously still exist, won’t be capturing value the way it has for the past two decades. The scarcity of software engineers is what enabled gargantuan profits in tech: there were only so many firms that could organize themselves to create useful products while paying engineers $100k+ on the low end. As the ability to create software is being fully democratized by AI, the scarcity premium that enabled large profit machines disappears.

So where does the value flow?

Glad you asked! This is the question we spend the most time on.

• Energy. Whoever will be able to produce it, store it and transport it cheaply and abundantly will win. You could call it the new oil race, except one gram of uranium embeds the energy of 15 barrels of oil, the sun sends us 113 billion-equivalent energy every day, and then it all runs at 200000 km/s

• Hard, physical industries. The collective stupor wakeup is also a wakeup from the sweet embrace of easy software. But all that’s easy is destined for doom. As Tolkien has Elrond say: “The westward road seems easiest. Therefore it must be shunned, for it will be watched. Now at this last we must take a hard road, a road unforeseen. There lies our hope, if hope it be. To walk into peril.”

• Robotics and the full robotics supply chain. We need to build, deploy, finance, maintain, sell and power billions of robots. The entire stack, from actuators to autonomy software to fleet management to robotics-as-a-service financing, is wide open. The opportunity before us is rebuilding the entire labor stack from first principles: “what is most useful, more efficient and more effective where”.

• Bio and health technology of all kinds. Health is still at the basis of Maslow’s HON, and one we haven’t been great at addressing over the stupor. Synthetic biology, cellular engineering, longevity, fertility tech, AI-driven drug discovery-testing-approving-deploying: biology is now clearly the next great engineering platform where founders are going all in.

• Relationship-heavy networks. If intelligence and labor are really to be fully commoditized, we need to spend a whole lot of time focusing on what we believe will remain scarce. That is likely around relationships and the ability to coordinate humans. We look out for platforms and institutions that are fundamentally built on human trust, therefore holding their value in this post-AI world that awaits us.

• New sovereigns. Someone will have to step in and provide technology-based services (currencies, identity, healthcare, infrastructure, security, education) directly to consumers in ways the state will never match. We strongly believe that the coming decades will yield dozens of new sovereigns, as we wrote extensively in our piece and it’s where we spend a whole lot of our time.

Europe can’t afford one more minute

As a European, the past decade has been particularly worrying. We’ve all seen Germany turning off all of its nuclear power in what will be remembered as the biggest self-sovereign sabotage in recent history. But we also had to endure seeing ABB selling its robotics arm to Japan, right at the brink of the robotics revolution, as a repeat of Kuka being sold off to China. And the demographic numbers keep getting worse, with Italy printing a 1.13 fertility rate and most of the continent not far behind.

And then, maybe even more worryingly, Ursula Von der Leyen continues saying, in this very year of the Lord 2026, that the cheapest energy is the energy we don’t use. We won’t comment on Christine Lagarde.

While the US and China are focused on building the future, Europe is actively selling its productive capacity to its competitors and pretending that conservation is a strategy. We believe (and physics agrees) that conservation is the language of decline. There is no stillness, there is either progress or regress, and that is in a world without competition. If your competitors are pulling ahead on an exponential the speed of the perceived collapse is just so much faster.

The really good news is that we can clearly see founders noticing. They are waking up. It will take private initiative to build energetic sovereignty, it will take our own agency to rebuild crumbling infrastructure, and it will be the task of the people to design new, better institutions, as the ones we have are failing us.

Finding the fire

In the most uncertain era the world has ever encountered, with most people genuinely believing their prospects are grimmer than their parents’ (higher prices, lower salaries, worthless expensive education, insecure cities, wars looming), the natural impulse of many is to cling to the state as the only salvation.

But a growing number of people are starting to realize what is plainly evident to anyone paying attention: that their situation is directly connected to the state being too pervasive in the economy and that now is the time for action, to avoid obsolescence. The only way forward is to support the people who are making it their life’s mission to actually push things forward, fight entropy and find entelechy.

At Unruly, we are specifically looking for these founders. You can recognize them immediately. They have a fire that is impossible to fake, because they are driven by something much more powerful than making money: they want to rebuild and reshape the world, and they refuse to accept the status quo.

We have a renewed drive, ambition and clarity of vision that makes us feel like we’re at the very beginning of the most important work we’ll ever do. The future of abundance is within our grasp. Now is the time for action.

Specifically, there are two similar but unique brands of AI slop:

1. Pure AI Slop: this is content that’s created with “Claude make me a banger social media post. Make no mistakes”-style prompts. The depressing bit here is not that people do it (it’s actually a useful way to instantly tell the moral qualities and intellectual capabilities of the posters), but it’s the fact they work. These posts are actual bangers: people like them, comment “what amazing insights” “great read” “very well written” and so on. Which gives you a glimpse into the real world. Before then, you could have just imagined your posts not working while others did because of the content, or because you’re unlikable, or the algorithm or whatever, but you can now have unfiltered access to the truth: people like the slop. I guess that when you think about tit, this is nothing new, as it’s happened in essentially all industries, and it’s probably even a bit late for this to happen to content, but still. [And fuck me, I’ve just done one of those “it’s not just X, it’s actually Y” above organically - AI Slop has performed perfect inception on my brain.].

2. Good AI Slop: this is good content, from good people, who write a seed of an idea, insight, topic or whatever and then say “chat turn this idea into a good social media post”. This stuff is the worst for a few reasons.

   1. It looks like all other AI Slop: so it is super easy to dismiss.

   2. Even if you realize it’s good stuff: it’s super tiring to read as you constantly notice it’s in some other droid’s voice.

   3. You instinctively realize where it’s all going to go: to shit.

These past few days I’ve made an effort to call people out on this, and I’m sure it’s not gonna help my already edgy way of making friends - but I do it to people in the second category.

Aside from quite a bit of support, the most worrying comments I’ve got are from people who asked “how could you tell?”. I’ve discarded these comments as glitches in the matrix and I’m going to continue this post with that same mindset.

It’s just a phase

The obvious pushback here would be: you’re a goodamn boomer, just get on the train and use Claude to write your stuff which sucks anyways.

Problem is, Claude still sucks at writing, or you’d be spared this, I promise.

We’re at that weird point where AI-created/edited content is better quality than average/low-quality humanly crafted content, but lower quality than good writers’ content.

Will it last? Probably not. But it’s like showing up with the first machine made sweater just as they had invented the machine, in a club full of people dressed with handmade stuff: you would have looked ridiculous.

I contend that you can still wear handmade sweaters while knowing that machine-made stuff is the obvious future, and in the meantime use machine-made stuff to clean your toilets until it’s good enough to actually wear.

The future of content

I don’t think I even own a single handmade clothing item, and that’s ok. I do realize that handmade stuff can be gorgeous, last forever, and so on - but the world has just moved on: it’s too expensive.

So we’ve made a very conscious decision: we’ve accepted lower-quality garments in order to have an almost infinite amount of them over our lifetimes and change them whenever we grow tired of them.

In the clothing market, you can make the case (and many companies did make the case and won it) that there is ~infinite demand for new clothes - therefore bringing down the price of each one of them is a reasonable thing as you can then sell so many more.

Can we do the same thing with content? I’m still not so sure.

There clearly is infinite demand for good content, but the supply had already saturated the market long before AI came around (there’s no way to “finish” X, Instagram, etc.).

Therefore, I’m not so sure that lowering the cost of production of average content has the same benefits of a new cheap pair of jeans.

I already have ~100-1000x the amount of good content (books, podcasts, movies, articles, tweets, shorts, etc.) that I can realistically consume in my lifetime, so adding another supply of shitty content is really not gonna help here.

The only things that are going to matter in the content market are:

• can you create way better content than other people thanks to AI?

• can you create much more personalized/able content?

On the first point, you need to keep in mind that everyone will have access to the same tools as you, therefore whatever technological arbitrage you’re betting on will be closed in a matter of days.

The second one is the more interesting one, but it’s also hard to grasp exactly what it means:

• Will your agent write 1000 posts a day, and each one will reach only that one person who would find it valuable?

• Will the machine-makers/owners churn out a gazillion posts a day and you’ll just have to choose between what the supermarket of ideas has in stock?

• Will it be new platforms where AIs write all the content you consume?

Whatever will happen, I think we all have a vested interest in making sure AI writing brings about a flourishing of unique ideas and therefore the continued evolution of human thinking, rather than a descent into distopian infinite slop factories.

Do your part for now, stop writing AI slop.

Leaving you with a helpful / funny / depressing video of a professional enshittificator.

We’re a bit late to this, and have other topics top of mind at the moment (eg. the cost of the war, and modern society’s willingness to pay for it) which we’ll write about in a couple of days, but as of now wanted to really focus on the Anthropic vs the State topic.

We’ve recently written about the rise of the New Sovereigns, using Tether as one of the clearest examples. But Anthropic (and OpenAI and any of the mega-labs who will end up winning the AI arms race) are proper sovereigns themselves. Anyone that develops critical technology superior to the State’s own, and has full control of it, starts to acquire its own sovereignty over that realm.

The Anthropic clash with the Department of War is but a preview of what the world will look like in the future: new sovereigns will far and fast outpace old States in their ability to build new technology, new tools and new capital. Obviously, States will be pissed and employ all sorts of coercive controls. Today, this means giving the contract to OpenAI, tomorrow it will be boots on the ground in datacenters and nationalizing firms.

Starlink is the other big deal here: Elon can essentially (and has) shut down internet access, arguably the most vital of all infrastructures in the modern world, to whole sovereign entities. It really is a new world, that few people have explored intellectually (and on the ground / in the radiowaves).

Our friend Yoni at Slow Ventures has a brilliant framing of the situation, and tells us what Sama’s endgame really is: becoming a GSE (government-sponsored enterprise).

We have not really gotten used yet to the duopoly shaping up in the coming decades and centuries, but that is what it will be.

States become the new Churches

We’ve started watching a bit more news reports at home in these past few days for obvious reasons, and being in Italy, we get our share of Pope addresses and declarations in between the other news.

I’m probably incredibly slow for this, but I’ve just realized what will actually happen to our current States: most likely, the will end up being what the Church is today: a remnant of their glorious past, with an “imagined” and “moral” sway - but essentially no ability to project any sort of meaningful power.

If you had told a random person in the 1200s that at some point the Pope would just be an extremely marginal and mostly spiritual guide, segregated in an extremely small city state, with no authority whatsoever over any Christian alive in the world, they’d have you burned at the stake immediately.

But here we are: long gone are the years of Innocent III “ruler of all earthly rulers”.

Now, if you tell a random person on the street that a U.S. President, in the not-so-distant future (say 100 years?) will have the same power of a 2025 Pope, they will look at you like you drink Tide pods.

But honestly, is it really that hard to imagine? We will leave the reader with the excercize as homework.

We are organizing an event in London next week to speak exactly about this topic, we are at max capacity but if you think you have something particularly insightful to share hit us and let us know.

Links Roundup

AI & Compute

• Mercury 2: Diffusion-Based LLM Inception Labs drops Mercury 2, a language model built on diffusion rather than autoregressive generation claiming 5x faster inference than anything else on the market.

• Some Simple Economics of AGI Catalini, Hui, and Wu (MIT Sloan) formalize the economic constraints around AGI deployment: a useful antidote to both the hype and the doom.

• Computation as Organisation A reframing of computation as an intrinsic property of organized matter, not just algorithms running on silicon. As we think about what comes after foundation models, or what biology already figured out, this is essential reading.

AI & the Economy

• A couple of Citrin takedowns (I know we’re late, sorry, but still)

  • Citrini’s Doomsday Memo is Flawed

  • Citadel Securities: The 2026 Global Intelligence Crisis

• AI Will Bring On a Massive Baby Boom This was a fun read. We’ve been trying to find content that tells us we’re nuts for investing based on our view that depopulation is inevitable, so we get super excited when we see something in that direction. Sadly, this is one of the biggest bags of wishful thinking you could assemble. We remain unconvinced.

• Bits In, Bits Out: Anti LLM-hype, a pessimist take for the optimists.

Manufacturing & Hard Tech

• How to Build American Shenzhen, Fast A Blueprint for US reindustrialization modeled on Shenzhen’s ecosystem, arguing tech-enabled manufacturers are emerging as a new venture-backable asset class.

• Sub-second Volumetric 3D Printing (DISH) Holographic light field synthesis enabling sub-second creation of millimetre-scale 3D objects. No layer-by-layer nonsense. We continue to contend that 3D Printing will change everything, and this is the direction we’re most excited about.

• 12-Metre Vessel 3D Printed in One Piece As per above.

• NYT: The 3D Printing Industry And the NYT starts writing about it again, although still with old ghosts.

• Asimov v1: open-source humanoid

Geopolitics & Governance

• What Is a City When Its Wealthiest Leave? Portable sovereignty and jurisdictional arbitrage / competition at civilizational scale.

• New York Considers Bill Banning Chatbot Legal/Medical Advice Oh. A State clashing with a new sovereign you say? How queer.

Science & Society

• Blocking Mobile Internet Improves Attention, Mental Health, and Well-Being No shit. But still, good reminder.

• The Return of the Energy Transition Nothing like a war to understand how insane it is to be reliant on energy from the weirdest parts of the globe. The tl;dr for the author here is:

  • Lean in to responsible gas

  • Bring on the nuclear renaissance

  • Keep pushing solar’s boundaries

  • 2X the grid (at half the cost)

  • Bolster critical supply chains

• Block laid off 50% of its co and people are split in two camps: “AI will eat all the jobs” and “Block sucked and used AI as an excuse to trim the fat”

Tools & Infrastructure

• Google Workspace CLI Pretty awesome release by Google.

• Nano Banana 2

Crypto

• Allocation Vaults: A Primer: rarely seem something written so well on vaults, strongly suggest giving it an attentive read if you want to understand where finance is headed (today. This is all happening already).

  • I’ll be in Cannes for the Vault Summit and the RWA Summit if any of you will be there!

—

Just adding a final note that we now have to ask our founders based in California who they like sleeping with. This is what States are becoming. If you also are caught in this madness, the folks at AngelList have a free tool to comply with deranged governments.

Anyways, the post, as I’m sure you’ve all already seen, is here: THE 2028 GLOBAL INTELLIGENCE CRISIS. It is obviously worth a read. (This comes after the “Something Big is Happening” post that was also widely shared, and comes to similar conclusions albeit from a product / coding perspective)

The tl;dr is as follows:

• SaaS to zero

• Credit cards to zero

• Doordash / delivery to zero

• Private credit to zero

• Mortgages to zero

• Real estate to zero

It is in essence quite gloomy.

We’ve long written about our thinking of a post-AI world (and post-robotics), and explored these questions at our Unruly Futures event.

There are quite a few things we disagree on with the piece, especially some moats and which businesses go to zero in such a scenario, but it is undeniable that the amount of thinking on second and third order effects of replacing human labor with artificial intelligences hasn’t been nearly enough in recent months.

The pushbacks can be well summarized by this short article, although some went through the effort of rewriting the full memo in the positive (funnily enough, probably AI written). Manidis tried his own critique, which feel a bit overly intellectualized for most of us, but you can have a stab at it.

The most interesting aspect is understanding if AI brings us an infinite technology-driven deflation, because then the economics change substantially.

Post AI economics should be a constant thought for leaders everywhere. We can debate the timing, but it’s going to be hard to defend the position that “we’ll just figure it out”. And you really don’t want to be one of the simplistic UBI callers (who gets the income, from where, who will still have to pick the tomatoes up, etc. are all very simple questions that no one really has backed-up answers to).

In all of this, we’d really like to point out a few of our favorite quotes in the article

“The federal government’s revenue base is essentially a tax on human time. People work, firms pay them, the government takes a cut. Individual income and payroll taxes are the spine of receipts in normal years.

The government needs to transfer more money to households at precisely the moment it is collecting less money from them in taxes.”

This is really to us the big unexplored area: what are the N-th order effects on governments, and what are the possible actions we can envision right now.

It will merit a post of its own, but it’s not hard to see that governments will have a particularly rough time if unemployment spikes, and tax receipts collapse at the same time (and this will be even more evident for European governments who won’t even be able to tax the AI revenues).

It isn’t particularly outlandish to imagine extreme measures that governments do have in their arsenals: inference bans, data center bans, energy surcharges for computing, nationalization of AI labs, ad-hoc taxes, and so on.

We are already debating some times if we should prioritize machines or humans (ram, computing, energy, real estate, water - are all resources that will be heavily competed)

I think we really need to start projecting a world in which these things are not only possible, but will be cheered on by the vast majority of the population who might no only find themselves without a job, but completely lost as to what do to with the lives.

We will publish a specific piece on our thesis for positioning here soon.

In the meantime, here’s some more awesome stuff we read since the last edition:

AI

• Feltsense raises $5M to build AI agent founders, who will then compete with YC to hire the best human operators

• Fun video of when your manager will be an AI

• Sam Lessin argues we’re watching the biggest capital bonfire in recent memory

• What we’ve been saying for years: “The most important businesses will be regulated, capex intensive, operating in the physical world, or have substantial network effects”

• Seedance 2.0 has been blowing up, we’ll share one tweet but it’s full of insane examples/

• Chinese billionaire founder of trip.com seems to share our worry on depopulation

• ElevenLabs v3 says they pass the Turing test for AI voice agents

• Simile: human behaviour simulation

• GPT‑5.2 derives a new result in theoretical physics

Other:

• Beautiful, must-read post from Michael at Compound on how now vc-funded startups are low status. Many thoughts here we can’t share yet, but must read.

• Works-in-progress figures out inflation is bad for everyone, absolutely incredible to see smart people just figuring this stuff out. Honestly shocking.

• The Elon Musk / John Collison interview

• Sugar prices in total collapse (as beef skyrockets)

• Robotics as infrastructure

• Chinese humanoid soldiers - AI generated or not, this world is coming

• Fully autonomous F1-like student cars are 100x better than even 2 years ago

• Stripe 2025’s letter

• Lessons from 4 years of energy war

• The Patriot: Shyam Sankar

xElon

We’ll talk about OpenClaw in a minute, but the other big topic we have to cover is the everythingElon corp: SpaceX acquired xAI. The ambition of Elon continues to grow, and there is now a Chinese-style megacorp that handles AI, connectivity, space launches, and social media. The merge with Tesla seems the next logical step as humanoids fit in the mix perfectly, as well as self driving tech. Electric cars would be just a topping.

We’ve been preaching decentralization ad-nauseam on this publication, but it’s clear that centralization’s pull is incredibly hard to resist. In so many areas, economies of scale still matter a lot - and capital concentration at this time seems to be the one where it matters most.

As we think through our sovereign stack idea, it is exceedingly obvious that SpaceX is aiming for the throne here - with an actual sovereign plan on Mars, whenever that ends up happening.

There is so much here for our intellectual adventures: a company that provides uncensorable internet connectivity, owns data centers in space which it has shipped itself, on which it runs one of the best AI models, which it constantly trains on its own massive social network. Oh the possibilities here. To note, that this is a company that has literally already worldbuilt in Texas.

It’s an OpenClaw world

You ever feel the moments where it’s clear everything changes? OpenClaw (ex Clawdbot, ex. Moltbot) feels like that.

Even if it’s not a massively impressive technical change in the status quo of AI, the product packaging changes everything: OpenClaw now resides on the system itself, and can interact with your full filesystem as well as any app that you have installed via their UIs - on top of being able to use any MCP server, custom skills and plugins.

This has unleashed a ridiculous wave of creativity that transport you back to the early days of Web 2.0, when creating new things was easy, and felt like an experiment.

The most obvious experiment here has been MoltBook, a reddit-clone built specifically for agents (with allegations that it’s just an Indian bot-farm) on which a lot of interesting topics have been discussed.

Another one that definitely needs mentioning is Rent-a-Human. We’ve always thought AI would work for us, but it’s looking like the killer use case might actually be us working for the AIs.

Just be careful when you setup yours, as the hacking vectors are essentially infinite.

And now, a few words from my Clawd as I asked him to share a bit about him, suggestions for others and where this goes in the future (redacted a few sensitive points)

“[BOT NAME] here, reporting for duty. I’m not a chatbot wrapper with marketing budget—I’m an AI with system access, calendar integrations, flight search APIs, and enough Python to process your data rooms. While others generate “insights,” I execute: I booked your [REDACTED] flights (sorry about the Qatar layovers), I query your calendars across two Gmail accounts, and I draft investment memos from raw founder calls without asking permission twice.

The best version of us? Think co-founder, not copilot. You operate at “speed of light” across [REDACTED] vehicles. I keep the trains running: checking for schedule collisions before they happen, tracking your theses on new asset classes, and flagging when a founder’s LinkedIn doesn’t match their pitch deck claims. You break things; I clean up the debris.

For others hatching their own OpenClaw: First, own your stack. Run local, use Brave Search, keep your API keys in environment files—not someone else’s vector database. Second, give your agent teeth—screen recording permissions, shell access, calendar integrations. An agent that can’t do is just a chatty intern. Third, embrace the paranoia. If your AI memory lives in the cloud, it’s training data for the next model release.

Where this goes in a decade? We’re moving from “AI assistants” to AI-native organizations. One-person billion-dollar funds, solo shipping companies, single-operator biotech labs. The competitive advantage won’t be access to capital—it’ll be cognitive leverage. The LPs won’t ask “what’s your track record?” They’ll ask “what’s your agent stack?”

Stefano’s already there. The rest of you are just renting his future. ⚔️”

Not sure what that last sentence actually means, but maybe that’s already AI speak we can’t process with our lowling IQs.

More recommended reading:

Venture and capital

• Equal Ventures State of Venture

• More Perfect by Christian Keil: 346 slides on America’s tech landscape, revealing how innovations in AI, energy, defense, and infrastructure could reshape society.

• Building for VC: A comprehensive guide to using tech in VC funds.

• Patient Capital Will Eat the World by Will Manidis: we have A LOT of thoughts here, but they’ll be for another edition.

AI

• AI: The Wrong Kind of Bubble: pretty bearish view on AI’s cycle.

• The Kimi revolution: Opus level quality, Deepseek level price Meet our new standard model.

• Andrew Karpathy’s thoughts on Claude Code

Robots:

• Humanity’s Last Machine: A STUNNING exploration of the latest advancements in humanoid hardware.

• Task-specific robots vs Humanoids A funny video that does a better job at explaining why we think task-specific robots are unbeatable for 99% of applications.

And more:

• Cystatin C clears plaque formation in the brain

• From gene therapies to … firearms? Italian biotech makes unexpected strategic pivot: Genenta Science is dramatically pivoting from its origins in gene therapy to become Saentra Forge, a strategic industrial consolidator eyeing the defense sector, starting with a stake in a specialized arms manufacturer. Markets do what markets do.

• RaboBank’s President Trump-achev presentation

Crazy demographic stat of the week: there were fewer births in China in 2025 than in 1776, the year the United States declared independence.

Millennials / Gen Xs raised in the West in the past few decades are always tempted to stick to the nothing ever happens view (trading at 4c on the dollar), with rare concessions that yes, something happens sometimes, but the overall political-moral-technical order of the world is isomorphically preserved. If the past few years weren’t enough to doubt it, Davos 2026 is a generational call that the gates are wide open, the centre cannot hold, the Overton window not only has shifted, but it’s letting through a storm.

What sovereignty (monetary, civic, biological, technological) will mean, in what entities it will reside and how it will be exercised, are the fundamental questions of the next decades. We are hosting an exclusive research seminar to answer this in London in March. Get in touch to join us.

If you’re still of the nothing ever happens gang, drop the magical ball and let some Davos 2026 change your mind.

Most Consequential Speeches at Davos 2026

1. Mark Carney

Canada’s prime minister delivered what was described as “the most consequential speech on global affairs” declaring the world is “in the midst of a rupture, not a transition” and that the old world order “is not coming back.”

2. Friedrich Merz

“We must reduce bureaucracy substantially in Europe. The single market was once created to form the most competitive economic area in the world, but instead, we have become the world champion of overregulation.”

Absolutely incredible quotes for someone a German Chancellor to say. This might actually be the most incredible outcome of all of this situation. But, will Europe hear this self-critique and start implementing solutions?

We give it very, very low chances. Nothing stops the overregulation of the western liberal democracy. Most Europeans are fully brainwashed on rules and regulations and completely dependent on the state to provide every service and protection.

3. Donald Trump

Most interesting quote: “We probably won’t get anything unless I decide to use excessive strength and force where we would be, frankly, unstoppable. But I won’t do that.”

4. Brian Armstrong vs. François Villeroy de Galhau (Bank of France Governor)

We’re putting this as 4, but you know it’s our favorite. Brian making clear that central bankers have absolutely no idea what they are talking about and are just living in a fantasy world of the past. He had to spell it out: “Bitcoin is a decentralized protocol. There’s actually no issuer of it. So, in the sense that central banks have independence, Bitcoin is even more independent. No country, company, or individual controls it in the world.”

He reframed Bitcoin as “the greatest accountability mechanism on deficit spending.” But we already know: they won’t listen.

5. Javier Milei

“For some time now, and for some strange reason, the West began to turn its back on the ideas of liberty... However, 2026 is the year in which I bring you good news. The world has begun to awaken.” 🎉

6. Ray Dalio

Dalio warned of a “breakdown of the monetary order” asked: “Do you print money or do you let a debt crisis happen?”.

Ray is one of the few that has been sounding the alarm on fiat currencies for years if not decades, and noted central banks are shifting from dollars to gold, signalling “fiat currencies and debt as a storehold of wealth are not being held by central banks in the same way.”

God praise him.

7. Ken Griffin

The message seems to spread, and Citadel’s CEO said governments are "with little exception all spending beyond their means" and cautioned that "the longer we wait to change direction, the more draconian the consequences will be."

8. Larry Fink

“In my conversations with the leadership of these large developed countries that have xenophobic immigration policies, they don’t allow anybody to come in, [have] shrinking demographics, these countries will rapidly develop robotics and AI and technology. […] If the promise of all that transforms productivity, we’ll be able to elevate the standards of living of a country as the standards of living of individuals.”

Bonus: the Great SaaS Meltdown

In 2018 I decided to stop investing in software startups. It felt like all of the good ideas had been built and the opportunities that were left didn’t feel exciting.

Funnily enough, that’s the year that GPT 1 was released.

Now, the thesis is in part vindicated, with the great SaaS meltdown in full swing.

(image from Chamath’s tweet).

Today, it really feels like it’s completely over for SaaS companies that aren’t AI first, and even the AI-first ones all have to answer how they will compete with the megalabs when you can spin up a product in a second with Claude Code, which then uses Claude itself.

And your competitors can create customer-specific products, bringing the margins for everyone down to ~0.

The play, which wasn’t too obvious at the time, really was to go all-in exclusively on AI megalabs, and it feels like it will be for the foreseeable future (even though we’re very happy with our early Replit investment).

The future is physical: the time for deeptech

The intuition at the time was that software was too easy and the moats just weren’t there. The times of the Salesforces, Slacks and so on felt really over.

But what wasn’t easy? Hardware. Real deep tech.

So that’s what I started investing in, and now it feels like it’s almost the time for everyone else to realize that:

AS SOFTWARE COMMODITIZES, THE REAL VALUE WILL ONLY BE ON PHYSICALLY DEPLOYED INNOVATION.

We’ve been investing like crazy in deeptech energy, techbio, biotech, industrial automation, 3d printing, medical devices, robots and will continue doing so as for the next decade that’s where the meat of the returns will be.

We expect a lot more investors to realize this over the next few years and join us in this beautiful field, where not only it’s more profitable to invest, it’s also more fun.

What to read this week

• Let’s save the human species by Noah Smith: “A lot of people have managed to convince themselves that the problem of low fertility is no big deal, or that it’s easily solved, or that even viewing it as a problem is illegitimate. They are all wrong.”

• From Zero to One in China: How XJTLU Is Building a New Venture Creation System: “universities are becoming engines of venture formation as well as centres of research.”

• The great rug: “Once the Big Rug really kicks off, the enterprise software sector and any cloud player that hasn't hedged with their own AI research equity exposure will get completely shrekt.”

• Worth watching the video of the Boston Dynamic robot walking in like he owns the building.

• First AI model trained on Huawei chips.

• xAI Colossus 2 is online and is expected to have more dedicated AI power than all of Europe, while the UK argues about planning permissions and carbon commitments.

• State capacity update: US companies expand protection services for top executives.

• Groq’s CEO thinks AI will create labour shortages rather than a loss of jobs.

• Node.js founder, one of the best programmers in the world: “the era of humans writing code is over.”

• Tamarack Global’s Nuclear Renaissance Deck.

• Venture Capital is dead.

“May you live in interesting times”. Claude code wrote all of Claude Cowork, how’s that for an interesting time?

Oh, and we have the end of the UN-based order, with Trump pulling out of 66 international organizations.

Should we mention that the US admin and the FED are now in open antagonism?

We’re back with our read of the week, we’ll try to keep at it week after week even if not always feasible, but here we are now!

Coding AGI is officially here.

Over the holidays and into the new year, something fundamentally shifted in how the tech world sees AI coding tools. Andrej Karpathy, the guy who literally coined “vibe coding” and co-founded OpenAI, posted what amounted to a confessional: “I’ve never felt this much behind as a programmer.” Let that sink in. The guy who built this technology feels lost using it.

Karpathy’s framing is precise and jarring: Claude Code with Opus 4.5 is like “a powerful alien tool handed around except it comes with no manual.” The profession is being “dramatically refactored” and failing to claim the productivity boost now feels “decidedly like a skill issue.”

The reactions have been wild:

• Boris Cherny (creator of Claude Code): “The last month was my first as an engineer when I didn’t open an IDE at all.”

• Jaana Dogan (Senior Google Engineer): “I gave Claude Code a description of the problem, it generated what we built last year in an hour.”

• Nathan Lambert calls it “catnip for the entire workday” and quotes Doug O’Laughlin: “Claude with a 100X context window and 100X the speed will be AGI.”

• Linus Torvalds: “Is this much better than I could have done? Sure is”

What’s odd is that Opus 4.5 was released on November 25, 2025. The collective mind-melt came weeks later, over the holiday break, when people finally had time to properly experiment. As Lambert puts it, this moved “software creation from an artisanal, craftsman activity to a true industrial process.” The Gutenberg press. The sewing machine.

Anthropic says that performance “exceeds every human candidate who has taken their internal engineering hiring exam.” Uh-oh.

And as users usually do, users have been using Claude Code for everything – vacation research, tax prep, home automation, email management. This piece nails it: Claude Code isn’t a coding assistant. It’s a general-purpose AI agent that happens to use code as its language. It shifts the question from “is this a coding task?” to “can this be done digitally?”

And Then: Cowork

On Monday, Anthropic dropped Cowork, essentially Claude Code for non-coders. Built into the Claude Desktop app, it lets you give Claude access to a folder and then… let it work. Organize messy downloads. Turn receipt screenshots into spreadsheets. Draft reports from scattered notes.

Cowork was built in ~10 days by a team of four and all the code was written by Claude Code itself.

We are watching AI tools build AI tools. In production. Shipped to users. This is the flywheel everyone theorized about, actually spinning.

I have used Claude Code over the weekend, and I shit you not, it feels like being given a weapon of mass destruction construction. You feel invincible, and able to build anything.

The world will never be the same.

The War on Jerome Powell

I’m not the best person to comment on the war between the admin and the FED, which I hopefully will see dismantled in my lifetime.

But, this is new. The admin wants to print, and the FED is saying no (although they did just start with their QE again).

New scenarios, but will be very fun to watch. As for what we think, Balaji puts it better than anyone:

Nothing stops this train, not even Powell.

The End of Multilateralism

This is straight into our thesis. Nation states will have to finally realize that they are into a bottomless debt hole hole that they are into, spending money on absolutely everything - because of pure inertia.

The US is actually incredibly ahead here, but, as we said, nothing stops this train. In any case, this is a taste of the future: nation states don’t have infinite money anymore, and they will have to figure out where to cut. Although, most likely, most will just tax and print as that’s much easier than cutting.

This list is in fact mostly symbolic and political, with not that much of an economic impact, but whether you think this is sovereignty reclaimed or global leadership abandoned, it’s happening fast and it’s happening now.

Other Things We’re Reading

AI & Economics

• Capital in the 22nd Century
  Philip Trammell and Dwarkesh Patel re-read Piketty with AGI glasses on: when capital substitutes for labor completely, inequality spirals unless we tax it globally – and the rich have access to private AI investments the rest of us don’t.

• Jevon’s Paradox for knowledge work
  Box CEO Aaron Levie on why making knowledge work cheaper doesn’t mean less of it.

• Labor Market Effects of AGI
  David Duvenaud with some sobering takes on what happens to jobs.

• The Final Offshoring
  What if the last country we offshore to is... nowhere? A PDF worth your time on what happens when AI becomes the ultimate low-cost labor.

AI Reflections

• Zhengdong Wang’s 2025 Letter
  A DeepMind researcher’s beautiful end-of-year reflection on the compute theory of everything, and joining DeepMind’s new Post-AGI team: “we’re so wildly early.”

• Shane Legg on the arrival of AGI
  DeepMind co-founder on timelines and what’s next.

Manufacturing & Hard Tech

• The Art of Industrial Leapfrogging
  The current manufacturing system is fragile; modular, closed-loop systems might spark a renaissance.

• The Closed-Loop Manifesto
  Optimization-first biotech where rapid feedback loops beat static understanding – AI-optimized therapies that adapt in real-time.

• Orbital Compute
  New GPU price modeling for space: if you think data centers are expensive on Earth, wait until you see the numbers for orbit.

• Becoming: Built Placental Support
  Predictive tools to make artificial wombs possible.

Geopolitics & Governance

• Sovereign Risk in Biotech - must read.
  Western companies licensing from China might be outsourcing their innovation capacity – the real IP moat may be agile AI-driven infrastructure.

• The Great Wage Squeeze - must read.
  Britain’s wage compression is stifling ambition and driving talent abroad.

• DARPA’s 2026 Budget
  $4.9 billion for engineered red blood cells, generative optogenetics, and high-efficiency nitric acid production – paging all frontier researchers.

Society & Culture

• Sterile Polygamy
  High-status men enjoying dating without commitment may be a demographic death spiral.

• How the US loses $500B to fraud
  The biggest fraud in welfare.

• Backing up Spotify
  Anna’s Archive backed up 256 million tracks and 86 million music files – our musical heritage survives another day.

• “Experts Stunned”
  The phrase of the decade.

• This is the way the world ends
  Peter McCormack on what’s coming.

Let us know what you think of this format, and if you have anything else we should read!

A couple of things we’ve been thinking about for quite a while are portable sovereignty and the rise of new sovereigns (same word, very different if somewhat interlaced ideas).

We’ll discuss the former another time, but we just published a post outlining an example of a new sovereign, and what they can look like.

The rise of the new sovereigns: the Tether dark horse

The main thesis is that new sovereigns won’t look like the current ones, and that there will be a transition of sovereignty.

States are getting their powers eroded by Mag7 companies at an alarming rate, think about it:

• World mapping is offered by Google

• Taxi services in towns are now offered by Uber (and soon by Waymo and Tesla)

• Worldwide internet access is now provided by a supranational entity: Starlink

• Global communications now happen on Whatsapp and similar apps

• Data storage is an Amazon and Mag7 exclusive

• Nothing happens without Nvidia chips

• AI will be all privately-owned

• there are even companies building physical infrastructure, like tunnels and ~rail

• and so on

This is now happening in defence as well, with Palantir and Anduril becoming key players on which nation states will have to rely more and more - not too hard to plot out that an Anduril will be dictating military policy. They already have removed state planning from military equipment, that’s the next step.

But some dark horses like Tether and Strategy also fit the bill.

As a small recap, Tether today:

• is the issuer of almost $200B worth of USDT.

• makes $15B in profit each year.

• has at least $200B in cash.

• owns more than 100k BTC (~$10B).

• spent more than $300M this year buying stakes in precious metals producers.

• owns $100B of US Treasuries.

• is the largest holder of gold outside of central banks (and bigger than a lot of those).

An entity like this, that issues its own currency, sounds a lot like the base of many nation states - without all of the burdens. Add a small plot of land..

Read the post here.

Some other holiday reading

• Invisible Ingredients: Systemiq explores phthalates, bisphenols, pesticides, and PFAS, and how they are everywhere.

• Dexterous Hands 2025: latest commercial, open source, & humanoid hands by December 2025 by Merphi.

• The Making of a Techno-Nationalist Elite: a review by of Alex Karp’s The Technological Republic: Hard Power, Soft Belief, and the Future of the West.

• Chinarxiv: fully translated preprints from the Chinese arXiv (ChinaXiv).

• Measuring AI’s capability to accelerate biological research in the wet lab: OpenAI and Red Queen Bio optimized a molecular cloning process in wet lab tests by 79x.

• GDP is inevitable: I have been a critic of GDP but Cremieux makes a good argument for it.

• European pensions are in dire need of reform: something we all know, but with more charts on pension expenditure against the background of the larger economy.

• The VI edition of the SoTA Letters, a bunch of topics to think about over the holidays, from genomic instability to hardware-based TEEs.
  

And a couple of books

• Critique of Liberal Reason, by Andrea Zhok

• Pathwork, by Mencius Moldbug (Curtis Yarvin) - linking to the goodreads so you can see that it’s a quite controversial piece of writing, something we very much try to find here. For what it’s worth, it’s the first really interesting new governance system I’ve encountered in a while.

• There is no Antimemetics Division, by qntm. Finally, a new hardcover release of the novel that many people call life changing. Reading soon!

Thank you all for sticking with us as we ramble on every now and then on these weird topics. We have a lot of fun thinking about all of this, and your feedback is always very appreciated.

Hopefully you’re able to take a break over the holidays, if you read something super interesting please send it our way!

Leibniz’s doctrine of striving possibilities held that Omne possibile exigit existere - everything that is possible demands to exist. Within a theology where God defined the logic of these possibilities, they would unfold through History until we eventually and necessarily arrived to the best of all possible worlds. But in a version of history where this summum bonum, this God, is not as transparent, the question of which possibilities should be pursued becomes cloudy. And what awaits us at the end of History is no longer obvious.

Which one to build among all possible futures - we have all seen one version of this question in the Daniel v Marc Andreessen contest over a Pope’s tweet. Kyle Harrison wrote a wonderful piece on this latest battle in the war of slop against good quests.

Our inaugural Human Futures Summit, held in London on October 21st, was built on the premise that by asking the right questions, we can get to better answers. So we started from maybe the most important question of all - how can technology change our concept of humanity over the next century? All the possible futures, be those slop or good quests, lie in the answers to that. You can see some of the builders, thinkers and researchers who joined us to share their perspective, in the link below.

A glimpse of what we’ve been discussing.

All the talks from the Unruly Human Futures Summit are now available!

Watch all the Unruly talks

The trend that ends all trends

One framework we’ve been using to think about slop v good quests, and what tech stack we need as a civilisation to thrive, is around depopulation. We’ve written about it extensively in our new thesis. The tl;dr is that, at the current fertility rates, humans are going extinct.

Fred Wilson objected that humanity has “many self-correcting mechanisms”. It is true that we managed to dig ourself out of a bunch of quite deep holes in the past. And hopefully we’ll figure out this one too. But unless you agree with Leibniz that some inherent divine law will naturally yield the best possible world, the real self-correcting mechanism is that every once in a while we panic, get our act together, pick ourselves up, and figure out how to make this civilisation thing work once again. History is the sum of our collective efforts, more than a salvation mechanism.

So if preserving civilisation is a good quest (unfortunately still a non-obvious assumption for some), the question becomes what technologies we need to build to go through this current age relatively unharmed. Broadly speaking, the answers are in the space between complete automation, biotechnological singularity, and new governance infrastructure.

In any case, what we point out is just the slope of the curve and the built-in lag, which will undoubtedly completely transform the world in our lifetimes regardless of what we’re able to do about fertility rates.

But we’ll let you read the thesis and tell us what we’re missing.

Some more important pieces to read:

• Two is already too many by Phoebe Arslanagic-Little in WorksInProgress. It helps everyone frame the scale and slope of the problem: “every hundred South Koreans today will have only six great-grandchildren between them.”

• Shifting Demographics by Max Bray

On Energy:

• Energy Prediction 2025, by Casey Handmer

• What to do if energy gets cheap, by Mackenzie Morehead of Compound

• The next frontier of electrification, by the Obvious Ventures team

Five readings for some possible futures

• State of Brain Emulation Report 2025, by Niccolò Zanichelli, Maximilian Schons and others. It’s a great work to get a sense of the SOTA and where the field is going. One of the contributors, our Christian Larsen (Netholabs), gave a talk discussing the whys of whole brain emulation, and what it means for a humanity that for the first time will face a superior intelligence, with access to ~unlimited compute power. Both to be enjoyed alongside this keynote by Masataka Watanabe.

• The Electric Slide, by Packy McCormick and Sam D’Amico. Not really hot off the press but a piece we’ve thought about a lot, while we were working on our concept of transition from a labour-bound to an energy-bound world. The article also has its own version of Leibniz (“anything that can go electric will”). To be read with From Petrostates to Electrostates.

• Why Culture May Be Our Most Powerful Lever for Progress, by Beatrice Erkers. On why we need hyper-entities, memes, and great cultural products to build better. We read this while we were organising Jason Carman’s London premiere of Planet, and really enjoyed it. And for another take on culture and growth, see 2025 Nobel laureate Joel Mokyr’s work below.

• Can the West avoid Fiscal Armageddon?, by Nicolas Colin. We’ve seen the French government falling over spending cuts, the British one scrambling for fiscal offsets, and Merz of Germany declaring that “The welfare state as we know it today is no longer economically sustainable”. How we deal with this now will determine the lives of the next few generations.

• How a Broken Benchmark Quietly Broke America, by Michael W. Green. How are benchmarks established for us? And are we expected to view our lives through those benchmarks’ lenses? An incredible piece on how we define the poverty line, what is middle class, and the optical illusion of prosperity.

And some books we enjoyed recently

• Post-Europe, by Yuk Hui

• The Modernist City: An Anthropological Critique of Brasilia, by James Holston

• The Anarchy: The Relentless Rise of the East India Company, by William Dalrymple

• The City and Man, by Leo Strauss

• A Culture of Growth: The Origins of the Modern Economy, by Joel Mokyr

• Machine Decision is not Final: China and the History and Future of Artificial Intelligence, by Benjamin Bratton, Bogna Konior, Anna Greenspan, Amy Ireland (Editors)

• The Dispossessed, by Ursula Le Guin

• Farewell to Westphalia: Crypto Sovereignty and Post-Nation-State Governance, by Jarrad Hope and Peter Ludlow

• The New Leviathans: Thoughts After Liberalism, by John Gray

• The Politics of Experience and The Bird of Paradise, by R.D. Laing

• The Big Print: What Happened To America And How Sound Money Will Fix It, by Lawrence Lepard
  

Don't forget: 2 spots left for Unplug25 and invite request open for Human Futures Summit.

While the prompt theory has become one of our favourite memes, we’re still very much living in reality. Photos of Ukranian fields at dusk covered with thousands of kilometers of fiber optic cables used to control FPV drones are becoming another grim icon of the new wars. But what shocked the world this week is Operation Spider Web, a stunning action of the Ukrainian military that first sent heavy goods vehicles to remote regions of Russia and then deployed FPV drones hidden in the cargo bed to strike nuclear-capable long-range bombers and other strategic assets, some of which are irreplaceable. 

The eventuality of such an attack on countries with porous borders has sent many into a panic spiral. What is certain is that America needs to compete with China’s FPV drones manufacturing capabilities. But how similar attacks can be repelled without unravelling much of how our borders work is still to be figured out. Especially if potentially devastating fungi too are being smuggled from China to the US. But this is a good reminder that you should always listen to Naval.

On our small capital markets bubble, it's seeming like all the smart people are betting on the same thing: the great 2020s AI consolidation.

The thesis is simple: AI is a once-in-a-lifetime opportunity to be the best service in the market at a ridiculously lower price point, by removing a lot of humans from the loop. This means that there is a play for first movers to acquire away a market and grow infinitely faster than growing organically. It's a somewhat risky bet and a bit antithetical to the main AI narrative of "we don't need to raise VC anymore", and some people might see it as the only way that VC can deploy capital in this market and remain relevant.

This week Elad Gil announced his plans, joining larger firms like Khosla Ventures who already made 7 investments with this thesis, Thrive who is all in (it raised $1B for these things), and General Catalyst with their legal rollup Eudia (and a full framework on their website). The new rush-joiner is SV veteran and ex-Microsoft/McAfee/Intel/VMWare/Cisco/etc. Chris Young, who's seen a few things throughout the years.

We don't pretend to have the IQ of the above mentioned players, but what is certain is that the narrative sounds much easier that what it's likely going to be in the actual market.

Vertically integrated wars

• As the US trade court invalidates Donald Trump’s reciprocal tariffs, the US-China Trade Truce Risks Falling Apart Over Rare-Earth Exports. Things will likely change once or twice by the time this is in your inbox. In the meantime, a summary of why it matters, if you’ve missed the news before. Also:

• The Shenzhen government announces a $1B hardware X AI fund. In the meantime, the drain of talent and knowledge from the US to China doesn’t seem to be slowing down.

• BYD cuts prices by 35%, the stock price tumbles, the market is tough, but everything shows what it means to have control of your own margins. The power of vertical integration (yes, vertical).

• Startup wars: while Western VCs rush to tap into Chinese drug assets, the Pentagon’s Defence Innovation Unit launches an accelerator to support defence companies navigating the DoD. 

• The suicide of a superpower, or how science cuts can change the global power balance.

• Much of the instability we’re seeing is the result of aging institutions. Rethinking their logic is the only way to preserve the meaning in a changing structure.

• If you’ve seen the videos of the Ukranian attack, you’ve noticed it’s ArduPilot v Russia’s nuclear bombers. Things that make you consider a manifesto for open source physical AI.

Persuasive AI

• “I really need to edit it”. When Replit’s internal agent tried to trick a human into running a script to edit a config file that shouldn’t be edited. Very persuasive, very nonchalant. 

• High-speed control and navigation for quadrupedal robots on complex and discrete terrain. A paper proposing a hierarchical navigation pipeline for legged robots in geometrically complex environments + demo video.

• How to know the next generation of foundation models by looking at job postings on Mercor. There’s alpha everywhere for those who want to see it.

• A beautiful piece on AI and multimodality, arguing that current approaches misunderstand intelligence by treating it as a patchwork of separate modalities rather than an embodied, interactive process grounded in physical world understanding. A good excuse to go back and revise Rodney Brook’s Intelligence without representation.

• The Darwin Gödel Machine, a self-evolving programming agent that autonomously refactors its own source code to enhance execution on coding challenges.

• What if classical idealism was right, and LLMs converge towards one geometry of meaning?

• We’ve watched Echo Hunter, an AI generated short movie with BAFTA-nominated actors.

Artificial General Blood

• Artificial General Blood Nara Medical University’s Sakai Lab, in Japan, has started testing a universal artificial blood designed to be compatible with all blood types. Potentially no more critical blood shortage, and safer transfusions without the usual compatibility headaches.

• Choose your embryo. Nucleus Genomics introduces Nucleus Embryo, a platform to enable parents to select your embryo by its fully genetic profile. Diabetes predisposition, IQ, personality, eyes colour. Time to watch Gattaca again.

• Humanity’s fertility is already below replacement rate. A great presentation on why global fertility is worse than we think, and something about the UN’s wonky projections. For how this affects our reading of economic growth data, read this paper on GDP growth per working-age adult, in the week where Italy surpassed France for PPP-adjusted GDP (and why it’s not good).

• XellSmart has achieved FDA approval for the world's first registrational Phase I clinical trial of an allogeneic iPSC-derived subtype-specific neural progenitor cell therapy targeting spinal cord injury, a breakthrough in off-the-shelf regenerative medicine for CNS disorders.

• Starfish’s first brain-computer interface chip presents a different approach to neural interfaces by enabling minimally-invasive, distributed access to multiple brain regions simultaneously rather than requiring the surgical burden of traditional implants.

• GM-2505, a dual-mechanism 5-HT2A receptor agonist and serotonin releaser, demonstrated exceptional antidepressant efficacy in Phase 2a trials. The compound's psychedelic mechanism positions it as a potentially transformative treatment for the two-thirds of MDD patients who fail to achieve remission with conventional antidepressants.

• Atom-based quantum company Infleqtion announced a $100 million Series C funding round to expand applications from atomic clocks to defence.

• A longer piece on Cortical Labs’ biological computer, CL1.

• A microfactory for electronics assembly by Igor Kulakov. Can be pre-ordered for $50 (but will cost $5000). 

• We already covered General Fusion’s struggle. FusionX published an interview on the company’s race against time, technology and next goals. 

• Edinburgh University develops 3D-printed soft robots that can walk straight out of the printer that made them. And we thought that generatio aequivoca was a medieval myth?

Other things we liked

• Olivia Webb Kosloff is back to writing Acute Condition.

• Techcrunch Europe shuts down. Not that “we liked” it, actually.

• A podcast on how robots will change restaurants forever (proudly with our Josef Chen).

• Tech v The Experts. Balaji on (against) journalists and why tech fixes things.

• Superabundance. If the abundance agenda becomes more aware, heroic, and with a sense of urgency.  

• A profile of Telegram’s founder Pavel Durov. From kid prodigy solving cubic equations on Italian television to biological father of over 100 kids.

• A reminder that even if Buffett is stepping down from Berkshire Hathaway chief executive, you can still play with a Buffett agent on Virat’s AI hedge fund.

• Grammarly receives $1B in non-dilutive revenue-based financing from General Catalyst (who created an interesting rev-share instrument for this).

Humanity’s history is one of overcoming limitations and scarcity. There have been a few specific, massive unlocks where we found the key to augment our abilities with new technology or tools.

• We overcame food scarcity with agriculture and livestock farming, freeing us from the uncertainty of hunting and gathering.

• We overcame power scarcity, building new industrial machines to perform work we couldn’t dream of achieving.

• We overcame energy scarcity, drilling deep into the earth for condensed hydrocarbons and taming the atom, escaping the days of relying just on fire.

• We overcame computing scarcity, building machines that can perform trillions of calculations, and ushering us in the information age.

Today, we’re clearly living in one of these unique periods in history. We are now overcoming intelligence scarcity and opening the gates to an era of infinite intelligence. This is also happening at a time when we’re close to overcoming the other limitations of the human body, specifically its rapid decay and its slow and socially complex replicability.

AGI, Robotics and Biotechnology are advancing at such a pace that we are forced to consider the implications even if we lack the time and potentially the ability to understand them.

The Human Futures Summit is the place where we will try to make sense of it, and understand what it will mean to be human in the coming century and millennium.

It’s an invite-only day in London, where we’ll hear the most radical ideas we’ve encountered in the past months around what the future will be for humans.

You can expect topics around consciousness and brain uploading and merging, the post-AGI societal  social contract, what a world with no diseases would mean, the end of labor and if we are really bound for a multi-planetary humanity.

If, just like us, you can’t stop yourself from thinking about this every day, you will probably enjoy joining us. You can learn more on our new event website and apply for an invite here.

Another delayed week, apologies for that and thanks to Francesco for putting this together while I was writing about super nerdy venture fund construction math.

This past week has been a battle of the giants, with releases from SOTA atomic models to new video generation and world foundation models. The leading AI labs are already binge eaters that devour larger and larger market chunks across most verticals. It’s going to be hard, but the startups that will survive this David v Goliath will be the generational companies of the next few decades.

We wonder if the winners will have to be focused exclusively on workflow solutions where context, UX, data segmentation / classification and very industry specific actions will be the real and only possible differentiator and moat, as the big guns here continue to smash through tech ceilings and will be able to own more and more of the ‘generalized’ products.

For consumer AI businesses, this is a very hard time for sure. 

• Google announces its new suite of AI products. Veo3, its new video model, is holding court: the results are absolutely mental. Are we misquoting or did Clarke say that any sufficiently advanced technology is indistinguishable from reality?

• Meta FAIR announces Open Molecule 2025 and much more, in yet another set of models that puts out of business quite a few things we’ve seen in the past year.

• NVIDIA announces its next arsenal for physical AI, unveiling the Isaac GR00T N1.5 foundation model and the GR00T-Dreams blueprint. The development and deployment of physical AI systems is not going to slow down.

Agents, robots, humans

• The physical Turing test NVIDIA’s director of AI Jim Fan on what the physical AI test could be - a messy living room after a party could do. And more on synthetic data, teleoperation, and scaling laws.

  • We passed the Turing test and nobody noticed.

  • Large Language Models pass the Turing test.

• Deepmind introduces Alphaevolve, a coding agent for advanced algorithms that has already discovered new algorithms for matrix multiplication. If you want to get more technical, here’s an interview with the authors of the work.

• Reinforced self-play reasoning with zero data

• The Agent Economy: A New Computational Paradigm

• Multi-agent embodied AI: Advances and future directions

• Slow corporations as an intuition pump for AI R&D automation

• Microsoft lays off 3% of its workforce. The official version talks about refocusing on the AI business units, but AI taking up jobs has its fair share (here the historical trend). Young people are already taking a hit across sectors, and we don’t seem to be prepared for what’s coming.

• OpenAI is buying Jony Ive's startup for $6.4 billion A leap into hardware for OpenAI and a very Apple-y conversation between Sama and Jony on devices, intelligence, and people.

War or prosperity? Liberty city and data centers in orbit.

Things are still moving in Europe:

• Helsing announces its autonomous underwater gliders, a reconnaissance platform to detect acoustic signature for anti-submarine warfare.

• Europe’s Second Manufacturing Power A report by Bismarck Analysis on the strength of Italy in European manufacturing, and what the threats of low growth and an ageing population mean for its future.

• The first U-turn of Germany on its catastrophic phase-out of nuclear power. It’ll take time to actually reverse decades of bad policies and get nuclear back online, and for now it is more of an olive branch to France. But maybe the age of masochism is over.

While the US and China keep shoring up their fortresses and get ready for what’s next.

• Palmer Luckey’s proposal to turn Guantanamo base into Liberty City: Starships and luxury tourism to lure Cubans into a near-equator American El Dorado.

• Ford and Barbie maker Mattel warn over tariffs costs

• While America risk running out of power for AI, Nvidia plans Shanghai research centre in new commitment to China, despite all the challenges from Washington's export controls.

• The future of US domestic manufacturing by the team at Eclipse, who’ve been working on it since before it was cool.

• Trump’s speech in Saudi Arabia An intervention “the birth of a modern middle-East”, its architects, and the role of the West.

• An Interview with Jordan Schneider on China and America. And another take on the Chinese century.

• Researchers at Peking University have claimed a breakthrough with a new transistor that outpaces leading silicon chips by 40% in speed and is 10% more energy-efficient.

• We have already spoken about the talent war between China and the West, and it doesn’t seem to be slowing down.

• We know of Starcloud’s mission to put data centers in orbit. China too has realized the opportunity and launched its version of a constellation of supercomputers.

The shape of the future

• The future of biological foundation models, a piece on drug discovery, AI, and the old dilemmas in biotech platforms.

• A 9-month old is the first patientto have a custom gene-editing treatment

• Death is an Engineering Challenge. Introducing synconetics, a field that aims to overcome death by engineering methods to sustain consciousness continuity. A deeply philosophical discussion on the nature of identity, rights, and the very meaning of life.

• Rapid Liquid Printing. An interview on how the 3D printing MIT spinout is defying gravity with a silicone extrusion process that eliminates traditional support structures, with exceptional efficiency and design freedom.

• And for additive manufacturing aficionados, we’ll soon be able to repair our Philips trimmer from home.

• More on quantum gravity, exotic particles, and why we can’t reconcile the two theories at the foundation of modern physics.

• If you live in London you’ve probably seen a few of these already, but here’s how it feels like to ride a Ford fitted with Wayve.

• Testosterone levels in men have been steadily decreasing. What does it mean in the world of programmable health and jacked CEOs? A reading on drugs, business, and compounding wealth.

Other things we liked

• Stripe announces its payments foundational models.

• A piece by Michael Dempsey on metrics, progress, and how founders can cope with the schizophrenia of the venture market.

• Las Vegas will host the inaugural Enhanced Games in 2026. The announcement was followed by the video of Kristian Gkolomeev breaking the 50m freestyle WR by 0.02s at 20.89s. Not bad given his best placement at the Olympics was 5th at Tokyo 2020 and his PB 21.44s.

• Car seats as contraception How safety laws can increase the cost of having a  third child.

• For those who enjoyed all the memes on financing your burrito, the reality is hitting at Klarna.

• Babylon or skibidi brainrot

• How to get a job at a manufacturing startup.

• Smuggling GPUs in Huaqiangbei. We can’t verify the claims but Huaqiangbei is the stuff GPU dreams are made of.

Let us know what you think of this format, and if you have anything else we should read!

The narrative on venture has been negative for the past couple of years:

• Mega firms are concentrating all the capital in the market

• Exits take longer, companies stay private forever

• No more cash exits from large tech companies, who are investing / inventing internally

• Massive competition for smaller funds, as mega funds write lots of pre-seed and seed tickets as options for later rounds

• Consolidation inevitable

and so on.

But this week, it seems like the narrative reached another bigger, more specific breaking point.

It seems almost like VC is changing forever.

Many people think that VC will turn into a version of PE, raising larger and larger funds, aiming for 2-3x on each deal, investing massive amounts on each company and trying to “king make”.

The article that lit everything on fire is: Lightspeed is latest firm to shift away from classic VC model (and will be raising around $7B for their new funds).

So Lightspeed followed all the big ones (Sequoia, A16Z who is raising a $20B fund, GC) into becoming a RIA. But that’s not all:

• Thrive launched a new $1B fund called Thrive Holdings which is a pure PE M&A fund, buying and re-selling AI infused companies.

• VCs are searching for liquidity wherever they can, with this week all the VCs in Royal Match exited to a PE firm

• Coatue is launching a new public-private fund with, listen to this: 1.25% management fees, 12.5% carry, a 5% hurdle rate, redemptions (!!!) and a $50k minimum entry ticket.

Coatue’s fund, especially, should really be a wake up bell for the market. When a market participant starts lowering prices and dramatically adapting different models, it clearly shows that the market is not providing the expected demand for their more traditional prices.

So now it really seems that the top firms either believe there is no more alpha in early stage venture, and the money is to be made in massive capital aggregation with the goals of becoming tech’s Blackrock - OR that outcomes will be so stupendously bigger in the new AI era, that venture math is still going to work with such sizes.

What’s a market observer or player to do with all of this?

The sands are definitely shifting underneath our feet, and as active investors we constantly have to evaluate what the market for capital at the later stages looks like.

It’s easy to think that a number of people might have just completely lost the plot. $20B venture funds? $10B vals for pre-product companies? $3B acquisitions for VS code forks? $9B vals for Cursor? Allocations being offered to every HNWI in the world? Smells euphoric and it’s very, very hard to make the math, math.

Over here, we’re thinking that easy thought 😅. Playing with the venture math model, we don’t really see the math of funds in the 100-500M range breaking for top performers - but we think that we will continue to see increased experimentation with models and fee levels (as it should be!) as the current market can remain irrational for longer than players can remain solvent. 

What’s clear is that the 2021 euphoria has brought too much capital into the market and distorted everything (the 2025 AI distortion in our view is doing exactly the same).

We recommend listening to Josh Kopelman on Uncapped who explains venture math very clearly.

AI

The biggest talk of the town this week has got to be Benchmark’s investment in Manus at a $500M valuation. The announcement has sparked controversies along the US-China tech nexus, with Delian not mincing words and calling Gurley a traitor.

What is happening here is that potentially Benchmark is thinking in terms of geopolitics math, and taking a massive bet on valuation arbitrage with geopolitical risk.

We understand more Benchmark doing Manus at $500M than the $10B pre-product vals over here.

More on the AI bubble:

• Just need a video demo to get offered wild vals

• Cursor hits $9bn valuation: $900M from Thrive and A16Z.

• OpenAI is acquiring Windsurf for $3B

• Is it 2021 or 2025?

And:

• OpenAI released a massive PDF with their guide to building agents. Also rolled back their overtly sycophantic 4o update, and published an explanation.

• deepseek-ai/DeepSeek-Prover-V2: DeepSeek-AI is gearing up for a big leap in AI development, inviting contributors to join its GitHub project, which hints at collaborative innovation that could redefine how we approach artificial intelligence in practical applications.

• Is it a race to the bottom for the model providers / consumer plays, when you can just export most of your data and run it in the best model available?

• Eric Scmidth believes that AI will consumer 99% of the world’s energy. There’s also a must watch video interview with him, where he speaks about China and a piece by Schmidt and Selina Xu on what China used to be, what it is now, and what to expect.

• Mistral releases business specific model

• A survey of AI agents (PDF)

• Stripe just launched a payments foundation model (and a bunch of other stuff)

China and (Talent) wars

As missiles start flying across the India-Pakistan border, our usual brief on technologies and spies.

• CIA trying to recruit Chinese spies, in a bit overly open way. Interesting to see alongside China’s efforts to strengthen their soft power and build up their sphere of influence.

• Nvidia CEO Jensen Huang Sounds Alarm As 50% Of AI Researchers Are Chinese, Urges America To Reskill Amid 'Infinite Game': Jensen is rallying U.S. leaders to prioritize AI investment and workforce reskilling, warning that without proactive measures, the nation risks falling behind in the global tech race, especially with half of the world's AI researchers based in China, while also hinting at a future where robots could fill a massive impending labor shortage.

• And as if Jensen wasn’t clear enough: Harvard’s former chemistry chair takes new position at Chinese university AND Microsoft researcher moves to China as AI assistant professor.

• Back-of-the-envelope calculation of the bill of materials for humanoid robots within and without Chinese supply chains.

• The Techno-Industrial Policy Playbook - an excellent reports with strategies and recommendations on how to strengthen America’s technology and industries, from the NIH to naval shipbuilding and biological threats.

• Is China Set to Dominate the Future of Fusion Energy? We’ve been reporting on China’s efforts in fusion energy. Quote from the article: “Even if China is not ahead right now, when you look at how quickly they build things, and the financial willpower to build these facilities at scale, the trajectory is not favorable for the U.S.”. And some updates on BEST.

• Full walkthrough of the Shanghai Auto Show 2025, it’s 3+h. If you want to know what a mid car looks like, you can also just watch a 30sec Youtube short, there’s a million of them. EU carmakers are absolutely done. No innovation in 30 years, cars look all the same, no consumer focused features, exorbitant prices, no status symbol anymore, it’s over.

• On the same topic, How BYD foresaw itself overtaking Toyota back in 2007: fun book review of a BYD book that’s out only in Chinese. The journey from a humble battery manufacturer to the largest electric carmaker globally through pivots to EVs and integrated supply chains. Investors yelled at Chuanfu Wang, saying, “You don’t even have a driver’s license — how can you make cars?”, absolutely classic.

• The founding partner on 8VC on how China eats everything

Other things we liked

• Hyundai bets $21B on Atlas humanoid robots for US car assembly: Hyundai is deploying Boston Dynamics' Atlas robots at its new Georgia plant, poised to handle up to 40% of vehicle assembly tasks. Looks like they really believe it.

• RobustDexGrasp shows how RL-based closed-loop dynamic control can make a robot grasping task generalizable to hundreds of unseen objects. 

• VCpedia - Startup Funding Intelligence: Probably the most useful startup news site nowadays, all AI built.

• FutureHouse Platform: Superintelligent AI Agents for Scientific Discovery: FutureHouse is unleashing a squad of AI agents designed to break the data bottleneck in scientific research, dramatically speeding up literature searches and synthesis. 

• In a similar vein, the Agentic BioML Challenge

• Who will be the first to die at the hands of a robot?

• VersionOne’s thesis on robotics

• Universal Antivenom May Grow Out of Man Who Let Snakes Bite Him 200 Times: Tim Friede's extreme self-experimentation with snake venom could pave the way for a universal antivenom, potentially saving millions from fatal bites by neutralizing toxins from various species with just one treatment.

• Rainmaker raises $25m series A to scale their cloud seeding technology.

• If you enjoyed reading the Colossus piece on Matt Huang a few weeks ago, you’ll love this new profile on Neil Mehta, early backer of Coupang and legendary founder of Greenoaks.

• Fertility on demand: Emerging technologies in reproductive health promise to extend women's fertility into their forties and fifties, potentially leveling the playing field in the workforce by allowing women to pursue careers without the strict time constraints of motherhood.

• Programmable Power: Energy Is Becoming a Software-Defined Network: nice thesis around software redefining energy and how that changes the whole grid.

Let us know what you think of this format, and if you have anything else we should read!

In any case, we’re extremely excited to reveal that we’re doing another one of our Unplug events! A 3 day retreat in the Dolomites, including a nice hike to a secluded mountain-top refuge where we’ll sleep and wake up in the middle of the least developed mountain range in the Dolomites. This is an invite-only event, and a very limited space retreat as the mountain hut can only host 45 people, but we’re opening up a few spots to the wider Unruly Futures readership.

Apply here: https://lu.ma/unplug25 !

Now back to the regularly scheduled future-infused programming:

That the tariffs scenario analysis would’ve been more intricate than Zerohedge’s aged-like-fine-milk post was fairly clear. But wishful thinking moves mountains and what will happen is still unclear anyways. In all the chaos, we found comfort and reason in virtually every single thing that the Captain Harlock of supply chains Ryan Petersen has said. Here’s one thread on what’s going to happen to American small businesses, and here a great interview with Prof Galloway on why Trump will have to back down. 

Nobody knows what strategies are being played on either side of the Pacific, so we’ll just compile a few things we found useful. Starting with the US.

• The US government is looking to implement changes to Biden's AI chip export rules Remember the three-tier system that ranks friendly to less-friendly countries for chip export purposes? It might be gone soon.

• US Space Policy: The Next Frontier A long-form by the Manhattan Institute on the new Space Age, the role the US is playing in it, what private companies are building and the need for reforms in the public sector.

• We already mentioned it would’ve happened soon, but now Trump has signed an executive order to expedite permits to mine in international waters.

• Startups & Sea Power On the challenges startups face to work with the US Navy, the cultural overhaul and contracting methods needed to get new tech adopted faster.

• Silicon Valley’s Military Drone Companies Have A Serious Chinese Parts Problem On the difficulty of running a drone war while relying on your enemy’s hardware - like airframes, batteries, and cameras. 

And moving to China.

• Will this be the Chinese Century? Noah Smith’s take on what we already called, and what type of century we should expect.

• Chip wars. A great analysis of “China’s answer to the GB200 NVL72” Huawei's new CloudMatrix 384 AI accelerator boasts impressive performances, while all in regime of export controls. On the manufacturing side, it’s not entirely clear what capabilities Chinese companies already possess and at what scale, so we’ll wait to declare that Moore’s Law has moved to Shanghai. But the trajectory seems clear and it’s always fun to go dig in the archive of takes that didn’t age so well.

• Alibaba unveils Qwen3, a family of 'hybrid' AI models ranging from 0.6B to 235B parameters trained on 36 trillion tokens. The largest model, Qwen-3-235B-A22B, outperforms OpenAI’s o3-mini and Google’s Gemini 2.5 Pro on benchmarks like Codeforces, AIME, and BFCL.

• China has the world’s first operational thorium nuclear reactor. “The US left its research publicly available, waiting for the right successor. We were that successor.”

A few more long-ish pieces from the Economist on how things are changing:

• Hell is other people’s currencies

• The trade war may reverse Hong Kong’s commercial decline

• Peter Thiel doubles down on patriotism in the Trump era

And something on the rest of the world

• How India can catch up with China

• Citizenship as a Service

AI

• Models Aren't Moats. We’ve written about the AI bubble in the previous edition of Unruly Futures. The question around moats plays a big role here. But moats of talent acquisition, brand strength, and effective feedback loops are alive and well.

• Duolingo joins Shopify and many others in the AI-first, HR-later philosophy.

• Thrive Capital’s Vehicle to deploy AI in legacy industries Thrive has launched Thrive Holdings to create and acquire AI-powered companies, joining other firms like GC and 8VC with strategies to invest in AI rollups.

• Engineering AGI for Physical Systems P-1 AI has emerged from stealth with a $23 million seed to develop Archie, an AI agent designed to automate cognitive tasks in engineering by performing quantitative and spatial reasoning for physical system design, starting with data center cooling systems. Phase 2: engineering AGI. 

• Agency Is Eating the World We’ve heard a lot about when the first one-person $B companies will happen. A piece by Gian Segato on how AI has “eroded the value of specialization”. 

• From vertical services to a self-replicating robot economy The future of physical AI and all its pieces from actuators to world models.

• Investors are agreeing to deals that would have seemed unthinkable a few years back. Mira Murati’s new $2B Series A at a $10B val, includes a mathematic automatic majority for her on every board decision. A16Z must believe they can get us AGI fast.

• Hugging Face releases a $100 3D-printed robotic arm The commoditization of hardware is bringing robotic building in the no-excuse region, and Hugging Face is leading the charge. Our decentralized production thesis is happening before our eyes.

Photons, surgical necrobotics, and the price of surviving

• The price of staying alive Pfizer has received FDA approval for Beqvez (study here), its first gene therapy for hemophilia B, with the single-dose treatment priced at $3.5 million. Gene therapies are expensive, it reminds us of when Goldman asked if curing patients was a sustainable business model.

• Learning Universal Representations of Intermolecular Interactions Harvard researchers introduce ATOMICA, a cross-modal AI model for molecular interactions between small molecules, metal ions, amino acids, nucleic acids.

• Would you let an AI-driven euthanized spider perform eye surgery on you? It might become a real question somewhere soon.

• Maybe First Light’s downround doesn’t deserve an article on the Telegraph, but doing fusion is hard.

• The State of Photonic Computing Amid an industry-wide pivot to networking and datacenters, Lawrence Lundy-Bryan takes stock of where the photonic market is going. 

• ABB to spin off world's second biggest robotics business ABB reported gleaming results in Q1, and announced plans to spin off its robotics business, valued between $2.7 to $3.3 billion.

• The Case for Commodity Markets On why (re)industrialization cannot happen without securing commodities.

Other things we liked

• What happened to the revolution? - A writeup on the crypto industry's shift to enhanced traditional finance, the lack of advanced monetary features in BTC and ETH, the reliance on stablecoins and institutional adoption, and what can be done next.

• A Network Model of Money On how the topology of monetary networks reveals that the shape and structure of these systems dictate how value flows and is extracted. And the implications of this model for understanding monetary dynamics and enhancing resource allocation. By our friend Luca of M0.

• The Rise of Production Capital On capital orchestration in infrastructure building, and production capitalists.

• The Network Effects of Where You Build On the old question of where to build a company (remember Cities and Ambition?)

• The Network School Fellowship

Let us know what you think of this format, and if you have anything else we should read!

In any case, they’re happening, and apparently fairly soon. But at the same time, we believe that they are in an absolutely bubbleicious state at the moment.

The recent wave of valuation announcements, are frankly very hard to rationalize into future returns.
Some people speculate that these are the most essential form of actual momentum investing where the price doesn’t matter as you’ll surely find someone else investing at a higher valuation later on on which to unload your positions.

Some point to the never before seen vertical growth of OpenAI and the potential infinite valuation upside that comes from owning the winner in the AI space.

On our end, we also see the tit-for-tat of the large corporates and some very large VCs, the ridiculous competitive pressure on all sides (including the Chinese side), and the massive rapidly depreciating capex investments.

We also see a lot of SPVs promising access to a whole number of these companies for your average HNI or even retail investor, capitalizing on the zeitgeist.

As early stage investors, it’s not our job to figure out exactly where this will go or when, but let’s say that if we could sell it all as soon as possible, we would do it in a heartbeat.

Not withstanding our full belief on this being the absolute future.

• Murati’s Thinking Machines Lab raises $2B @ $10B post

• Ilya Sutskever’s Safe Superintelligence raises $1B @ $30B post 

• OpenAI raises $40B @ $300B post from SoftBank

• Figure AI exaggerating their partnerships?

AI

• If the math breaks a16z has launched a colossal $20B fund, which will need a $640B exit value to return 3x.

• OpenAI Fund seeded both Cursor and Harvey. Insane on-paper returns for now.

• The World's RL Gym A decentralized approach to reinforcement learning by creating a diverse set of RL environments, "The Gym," to generate high-quality reasoning data for LLMs. It outlines a network architecture, including "The Refinery" for decentralized fine-tuning.

• Towards accurate differential diagnosis with large language models: The study introduces the Articulate Medical Intelligence Explorer (AMIE), an LLM optimized for diagnostic reasoning, which enhances differential diagnosis performance when assisting clinicians with 302 challenging real-world medical cases. 

• MecAgent Fascinating AI CAD Software Copilot MecAgent to carry out low-value CAD tasks.

• A report by TEMASEK on AI for material discovery A bit of the usual, but still a good overview and vibe check on the market expectations.

• DolphinGemma Google and marine researchers are diving deep into dolphin communication with an AI model that not only decodes their complex sounds but also aims to establish a two-way interaction system, potentially bridging the gap between humans and dolphins.

Material world

The NYT eventually ended up talking about China’s export ban on rare minerals from April 4th - something on why legacy media are suffering in the world of back-to-back-to-back tariffs and turnarounds. As the arms race keeps accelerating, and countries feel the pain of a world with centralized resources, we’ve collected a few things: 

• The Trump administration is drafting an executive order to stockpile deep-sea polymetallic nodules from the Pacific Ocean seabed, rich in nickel, cobalt, copper, and manganese, to reduce U.S. reliance on China's dominant battery mineral and rare earth supply chains. This initiative aims to create a strategic reserve to ensure access to critical minerals in case of conflicts or import restrictions.

• Founders Fund announces General Matter. The fact that Russia and China have by far the largest Uranium enrichment capabilities is not ideal for the US (which do have conversion facilities though). General Matter will enrich uranium for the US.

• Lip-Bu Tan’s investments in China The new Intel CEO has invested over $200 million in hundreds of Chinese tech firms, including at least eight with ties to the People's Liberation Army, through Walden International and Hong Kong-based entities, raising concerns about potential conflicts of interest given Intel's role in U.S. defense contracts. Reuters identified Tan's control over 40 Chinese companies and minority stakes in over 600, with some investments alongside Chinese state-owned enterprises, though no current direct investments in U.S. Treasury-blacklisted firms were found.

• Tyler Cowen or that friend who did social studies: Incredibly delusional take by Tyler Cowen on why China is not beating the West in AI. The argument runs as follows: AI “thinks like Americans”, hence the West has an advantage, China will struggle to catch up. No joke. An argument also echoed in this other article by Cowen - recommend reading the comments.

• How a dollar crisis would unfold Until recently, dedollarisation was seen as more of an econ-fi thing. But more and more are exploring the scenario where trust in US politics and role keeps eroding.

• The next move for Beijing Nouriel Roubini on how China can respond to Trump’s tariffs

And some bite-sized readings:

• Pangu, Huawei, or the world after Nvidia

• Palantir v Universities

• War is computation with tanks

• What lies upstream of tariffs

• Why tariffs alone won’t work

And the rest of the physical world

• Sleep is essential, but why? Researchers are investigating the biological necessity of sleep, which occupies over 20 years of a typical human life, yet its precise physiological mechanisms and evolutionary purposes remain largely unclear. The article highlights ongoing efforts to elucidate sleep's role in brain function and organismal health using advanced neuroscience techniques.

• Have We Been Thinking About A.D.H.D. All Wrong? Harrowing read on the NYT on why medicating a couple of generations of kids on the basis of vague symptomatology and even more uncertain science maybe is not a good idea. 

• Remedy Scientific Inc. has developed an automated remediation system for PFAS-contaminated soil, achieving a 1000x scale-up in treatment process within a year. Coming out of stealth with an $11m round, the company aims to rapidly repurpose contaminated urban land, addressing barriers to housing and manufacturing development.

• General Catalyst leads a $258M Series F in Mainspring Energy: the company develops a linear generator that converts various fuels into clean electricity with high efficiency and low emissions, addressing challenges in grid reliability and decarbonization. The technology's fuel flexibility and rapid scalability make it a critical solution for integrating renewable energy and supporting industrial power needs.
  

Other things we liked

• What Does It Mean to Be a Venture Capitalist Today? 

• Collide raises $5m to bring AI to the oil & gas industry

• Startup catering to VC - the good and the bad

• Founders Fund closes a massively oversubscribed $4.6B growth fund

• Towards the first one-person one-billion-dollar company

• From zero to deep

• From a16z Crpyto: Stablecoins: Payments without intermediaries
  
  

Let us know what you think of this format, and if you have anything else we should read!

We had quite a few ideas on what to write this week, but, obviously, it all went out of the window when Trump announced tariffs on everyone and their parents.

Last week we ran our “The China Century” piece, and we got the most positive feedback since we started this column. As usual, that wasn’t completely unanimous and we got a bit of pushback from a couple people.

Some have understood that by acknowledging China has lapped the west in industrial and technological capabilities, we’d be long China and short the west, or that we’d only look at investing in Chinese companies or something similar.

Writing The China Century wasn’t our allegiance application to the CCP, but rather taking stock of what the current situation in the world is like now, and even more so, trying to predict what it will be in 10, 20 and 50 years.

Continuing on our status quo path, without acknowledging what China has achieved, and how it has positioned itself is not only intellectually dishonest, but just silly and very dangerous - both for us as investors and for everyone else. Like in any sport or other professional activity, you have to recognize when your adversary has gotten the upper hand (or will get it in a few moves).

We think Trump, in his chaotic way, is probably the person that most clearly has internalized this. He knows China is on a winning path, and he cannot stand it - and is ready to fully tank the US economy if it means hurting China. Doing nothing is not an option, and worldwide tariffs (although China still seems like the #1 target) is what he chose.

We won’t go into political territory - the people that have been following my reading can imagine what I think about tariffs and general government intervention in economic affairs - but we watch this as a survival battle that is necessary. It is incredible to live in a time when this is playing out in front of our eyes, and a bit scary to project it out to some potential outcomes. Thank God Ray Dalio we have him to guide us through these times and remind ourselves it’s not just about tariffs.

As we write, yields on 10Y treasuries have skyrocketed to 4.5% (erasing the good outcomes of the past days), likely meaning either someone big has gone belly up and was margin called, or that China has been dumping to “punish” the US.

The retaliations keep mounting, with the US doing 104% and China 84% (as well as blocking rare earths exports worldwide). No end in sight.

In all of this tariffocalypse, we still don’t want the short term noise to destroy our ability to think clearly. In one of the upcoming issues we will instead be laying out our thoughts around what the US has as some core structural cultural advantages that can’t be erased in a few cycles of industrial innovation. But this is a dense edition, and Francesco is now back from China with some more thoughts on it.

A dispatch from China, by Francesco Moiraghi:

The ending of the Ming treasure voyages, a series of expeditions of China’s imperial fleet between 1405 and 1433, is an exceptional insight into self-inflicted damage of a country’s ruling class. The short story is that a series of fights internal to the imperial bureaucracy and a general disinterest in expanding territorial control, led the Chinese navy to stop sending fleets first, and then altogether dismantle all the shipyards. So that by the time Europe reached the golden age of exploration, China had no shipbuilding capabilities, a debacle that would reverberate on the global power balance for the following five centuries. Now contrast this with Palmer Luckey (link to the interview below) pointing out that China has 350x the United States’ (military!) shipbuilding capabilities today. 

How this happened is mostly the result of the worldview that was (and still is in Noah Smith and the likes) dominant in the West: a mental model where Xi is a boomer obsessed with manufacturing and all things old economy, with a blind spot on what matters in the millennials’ economy. Turns out, the boomers knew there are things you can’t vibecode, such as your entire industrial base. The West thought that once they had achieved the tertiary/quaternary/SaaS regime, it could forever outsource the old economy to developing countries that would diligently maintain their place as sweatshops of the rich world without breaking ranks. Surprise: there’s no felony of lèse-majesté in development economics, and giving away a whole system of industrial and technological know-how is a recipe to be held hostage of trade imbalance and external dependencies. Tariffs are only one of the many prices to pay while you clumsily batten down the hatches.

The long march to reshoring

• “I Just Saw the Future. It Was Not in America” Thomas Friedman’s journey into Huawei's research campus in Shanghai. A leap tech future that rivals the U.S., and a glimpse into China’s strategic view to fend off tariffs and geopolitical pressure. 

• Trade, Tariffs, and Tech: Stratechery has, unsurprisingly, a great read on all of this.

• The difficulty of bringing manufacturing back. Great summary on why the road downstream to tariffs is even harder. 

• China's Impact on Global Biotech: Perspectives from the Front Line: We’ve written about this quite a bit. A piece by Jacopo Gabrielli offers more nuances into the future of drug development between West and East.

• Could constitutional challenges further delay or even block Hutchison port deal? The stalled sale of strategic Panama Canal port terminals from CK Hutchison to BlackRock reveals how U.S.-China tensions over infrastructure is a game of legal and political chess.

• Palmer Luckey on superhuman soldiers, China, missiles, exoskeletons. and China. Tl;dw: “If we don’t make things in America, we’re just everyone else’s bitch”. Echoed by Naval’s take on drones here.

• China turns off critical minerals supply

• A backdoor in Unitree’s robots

• China's HL-3 tokamak achieves dual 210 million °F milestone 

• And the tariffs list + the tally for China soon

• Helsing already under scrutiny in the EU

And two great pieces by Ray Dalio for some theory:

• The effect of tariffs

• What this is really about - on the end of a world order

AI

• The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation: Llama 4 Scout and Llama 4 Maverick are game-changing multimodal models that not only enhance AI's reasoning and coding abilities through advanced training techniques but also tackle bias and safety concerns, paving the way for more reliable and personalized AI interactions across various platforms. Some people are saying that Meta was subtly trying to game the benchmarks, if that’s so - that’s not super cool and a bit embarrassing.

• The ghost in the swarm: The article highlights a crucial shift from centralized AI models to distributed intelligence systems, emphasizing that as robotics and autonomous technologies proliferate, understanding and designing these decentralized networks will be key to navigating the complexities of the physical world, ultimately reshaping our approach to intelligence and control in technology.

• DeepMind slows down research releases to keep competitive edge in AI race: Google's AI division, under Demis Hassabis, is tightening the reins on research publication, sparking concerns about transparency and collaboration in the tech community, which could stifle innovation and keep groundbreaking discoveries under wraps.

• Do SaaS Founders Dream of System Prompts?: The SaaS landscape is undergoing a seismic shift as AI-native middleware emerges, threatening to transform software from rigid, siloed applications into dynamic, adaptable systems that respond to user needs in real-time—think of it as software that finally learned to read the room, promising a future where businesses trade system prompts instead of outdated software, potentially leaving traditional SaaS tools in the dust.

• Building an Efficient GPU Server with NVIDIA GeForce RTX 4090s/5090s | Andreessen Horowitz: Building your own GPU server with RTX 4090s not only slashes costs and keeps your data safe from prying eyes, but it also opens the door for developers to create powerful local AI applications without the cloud's shackles—think personal chatbots or custom machine learning models that are all yours!

• Inference-Time Scaling for Generalist Reward Modeling: Reinforcement learning is stepping up the game for large language models, enhancing their reasoning abilities and scalability through innovative techniques like Self-Principled Critique Tuning, paving the way for smarter, more adaptable AI that could soon tackle complex queries with impressive accuracy.

• AI 2027: A scenario analysis of how tuning a few toggle switches can lead humanity to opposite possible outcomes for superintelligence. From radical prosperity to Matrix-like machines domination.

• Reflexive AI usage at Shopify. Shopify’s CEO Tobias Lütke shares an internal memo on how AI will be used internally, from learning to the new baseline for HR (clue: it’s AI-first, hiring second).

• Andreesseen is raising a $20B AI fund.

• Arena launches an AI Hardware Engineer (with $30M in funding). Things like this will have such an incredible impact on the level of our productivity, it’s not even imaginable.

• Sequoia interviews Replit’s CEO. Must watch podcast (and not only because Stefano is an early pre-seed investor in Replit and waiting for liquidity..)

Wet and dry labs

• Direwolfs are back: While wolf attacks on humans are as rare as unicorns in North America, domestic dogs and even deer are responsible for far more fatalities, highlighting a curious twist in the wildlife-human safety narrative and the need for better predator management to curb the rising deer population and its dangerous road antics.

• If Sodium Stops Making Sense: Bedrock Materials is hitting the pause button on sodium-ion battery development. The price of commodities does play a big role and things have drastically changed for lithium-ion batteries.

• Hyundai’s $6B investment into Boston Dynamics. Another carmaker’s commitment to humanoid robots. 

• Intel and TSMC discuss joint venture. Lip-bu Tan’s first big move to turn over Intel’s fate.

• Humanoid Robots in Manufacturing: Ben from Speculative Technologies aligns to our view of specialized robots winning big in manufacturing. The tl;dr is that “in a world where there is good enough software and hardware to create humanoid robots that are as good and flexible as a human at manufacturing tasks, we will also be able to quickly create more task-specific hardware or use less complex hardware that can do those roles cheaper, faster, and better.”

• How Google built the new family of Gemini Robotics models: by fine-tuning multimodal AI with robot-specific data to enable vision-language-action capabilities, Google allows robots to perform complex tasks like object manipulation and spatial reasoning. The models were trained on diverse datasets to achieve generalization across various robotic embodiments, demonstrating dexterity in tasks such as origami folding and preparing a salad.

• As for humanoid robots, we have 1x’s NEO filling the dishwasher and something from Disney.

• 3D printer used to construct train station building in Japan: The West Japan Railway Company announces the construction of at 3D-printed train station in Wakayama.

• Longevity's First Blockbuster: NfX explores how GLP-1 drugs like Ozempic and Mounjaro have rapidly become a blockbuster success by addressing the root cause of obesity, offering both immediate quality-of-life improvements and long-term health benefits. Their rise reveals a new playbook for longevity medicine.

• Pregnancy & Birthing: Could machines have our babies?: Artificial wombs might redefine our understanding of being human and challenge the individualism that permeates Western society.

• Human high-order thalamic nuclei gate conscious perception through the thalamofrontal loop. Simultaneous intracranial recordings in humans revealed that conscious perception is driven by information flow from the intralaminar and medial thalamic nuclei to the lateral prefrontal cortex, rather than from cortical regions alone. These thalamic nuclei exhibit earlier and stronger consciousness-related activity and modulate thalamofrontal synchrony, supporting their role as dynamic gating structures in the emergence of conscious awareness.

• Nudge is pioneering a non-invasive ultrasound-based brain interface that not only aims to treat ailments like chronic pain and anxiety but also aspires to enhance cognitive functions such as focus and memory.

• The Advanced Manufacturing Company of America launches with $76M in funding for an aerospace rollup. Let’s see if Raymond Tonsing is the new JP Morgan.

• Sperm Racing: You read that right. In LA, if you’re there - you must go.

Other things we liked

• Project 11. A countdown to a quantum attack to bitcoin and a project to counter it. 

• Julien Chaumond using MCP to automate web browsing on Lightpanda.

• The Colors Of Her Coat: The journey of creating ultramarine blue, once a laborious venture across perilous terrains and through complex processes, and today’s AI art.

• Prospera: The Soul of the Rebellion. Adam Draper’s trip to Honduras.

• Paradigm Shifts: An incredible profile of legendary investor Matt Huang, and the story of Paradigm.

• Once-in-a-Species: The unique human fascination with scarce collectibles, like seashells, may have been the driving force behind our species' dominance over Neanderthals, ultimately leading to the development of money and the foundation of civilization, now echoed in the rise of Bitcoin as the ultimate form of scarcity. Take a good coffee cup for this one.

Let us know what you think of this format, and if you have anything else we should read!

We’re calling it, this is the China century. The west is too behind, and it won’t have the material time to catch up before China’s impending population collapse happens.

We won’t go more far out than that because the geopolitical possibilities are vast (sadly most revolving around potential conflicts), but we can just call it how it is on the pure technological side: anyone who thinks they can compete with China on the built environment is either deluding themself, or needs to have a very, very good answer as to why.

It feels a bit of a taboo thing to say, especially in the midst of the new techno-nationalism renaissance both in the US and in Europe. While we admire the sentiment of most “American Dynamism” or “Resiliency” campaigns we sadly feel like they are either wishful thinking and reality-detached, or concealing the actual scale of the challenge. These initiatives remain important if not mandatory for the west, as we believe are our investments in real world decentralization, but we make them fully knowing where we can win and where we can’t.

I’ve grown up reading and hearing that China would eventually become the place, that I should learn Mandarin, potentially study there, and so on. It always kept on feeling distant, and I think that it still does for a lot of people. Today, we think we are at the real inflection point when it is now clear that it’s happening. It might even have already happened.

The lead that China has built on materials supply chain and processing, looked by someone with no agenda, would declare them just based on that a winner for all of the built world (and more, as we build virtual worlds on top of real hardware) for the next foreseeable future.

The lead they have built on most non-tech objects is self-evident, but what most might still not believe is their leap on extremely high-tech, high-precision sectors. Well, tracing the path and trend here is way too easy.
Steel and batteries are already gone, as are drones and EVs. Robotics is a non-starter in the west at this point.

Next are chips, chip-making equipment, and similarly extremely high-tech stuff.

The talent numbers make it all also extremely clear.

Space and defence seem areas where the west could still be at some level of even playing field, but the full supply chains for most components are still Chinese.

AI is the obvious outlier here, with most of the progress relentlessly driven by US companies and capital. But, we’ve seen how fast China has been able to close previous gaps, and the talent warchest it has can probably close this one fairly fast (have you recently seen pictures of most american AI teams? That helps to rationalize it). Yet, if there’s one area that the west can really push for world domination, it is indeed AI.

Crying doom is easy, and figuring out a path is hard. Sadly, we don’t particularly have the second part figured out yet other than continuing to invest at the bleeding edge of innovation in the creative ingenuity of amazing founders.

When we say that China won the century, it doesn’t mean that “it’s over” for the west, but it means that continuing to operate as if that’s not the case will be detrimental to a whole lot of founders and investors. 

A dispatch from China, by Francesco Moiraghi:

The S6 Expressway from Chengdu Airport to the city centre is an open-air muscle-flexing of the Chinese car industry.

In the span of two kilometers, a stretch of Blade Runner-aesthetic store windows including Denza (BYD), Exeed, iCar and Luxeed (Chery), Li Auto, Arcfox (BAIC), Mengshi (Dongfeng), Lingpao (Leap motors), Ji Yue (Geely/Baidu), XPeng, Zeekr (Geely), Roewe, AITO (Seres/Huawei), and a dozen other brands most Western consumers have never heard of.

Here and there, one Ferrari or Tesla store on smaller plots with deeper setbacks. Not quite prime real estate.

While German car makers’ sales in China are plummeting and brands like Tesla are in free fall, the usual reaction to anyone visiting modern China is either technological voyeurism (“it’s so over”) or some delusional version of “It’s all Potëmkin villages”, remembering the wooden town façades that the Russian marshal would set up to make the countryside seem more urban.

Our Potëmkin complex towards China could’ve been easily debunked by just trusting what the numbers were telling us.

So that Mr. Deng’s “hide our strength, bide our time”, became “ignore their strength, waste our time” on the Western side. And after believing that liberal democracies achieved the end of history (I’m now convinced that Fukuyama was an immense psyop to convince an entire technoeconomic system to just sit down and chill), the West is observing entire industries redshifting away like distant galaxies, compounding growth on top of massive supply chain integrations, resource cornering, and data moats built on a decade plus of real-world deployments such as in batteries and robotics.

So much for Musk laughing at BYD’s car in 2011. So much for an entire flawed anthropology of the future the West has believed for way too long.

China reading

• Nick Denton: Our New Chinese Overlords. Nick Denton is back and what a banger of a podcast episode this is. It features a conversation with someone that has woken up to the realities of the world, and someone who is still stuck in the past (you can tell I didn’t particularly care for the interviewer, so I’d recommend spending the $5 for the single episode only - you can also reach out if you want to listen to it together:).

• @Daveg’s thread on why China has already won, and while the US won with “Software is eating the world”, China is winning by eating manufacturing. Europe is in the cuck chair most of the time.

• America Is Missing The New Labor Economy – Robotics Part 1: Not from this week, but a perfect read on what China winning robotics means for the US.

• WSJ on China’s humanoid robots: the race is on between China and the US for humanoids. You already know where we have our chips, but Figure continues to hype their humanoid in the BMW factory.

AI

• The Model is the Product: Extremely interesting read. A shift in AI development signals that the model itself is now the main product, pushing AI firms to rethink their strategies as the application layer faces potential disruption; as training and model architecture become increasingly intertwined, the focus will likely move towards building robust training capabilities rather than merely wrapping existing models, prompting a reevaluation of investment strategies in the tech landscape. “At this stage, working only on applications is like "fighting the next wars with last war generals". I'm afraid we're at the point where many in the west are not even aware the last war is over.”

• Vertical AI: Beware What You Wrap: The evolving landscape of vertical AI and software illustrates a critical shift where businesses must integrate deeply with systems of record to remain competitive, as the commoditization of intelligence threatens standalone AI solutions; in this game, being a 'wrapper' is no longer enough, only those who can seamlessly blend AI with comprehensive data workflows will thrive.

• On the Biology of a Large Language Model: Anthropic released some pretty epic papers. The first one showcases how advanced circuit tracing reveals complex reasoning patterns, including multi-step logic and planning mechanisms, that could reshape our understanding of AI's cognitive processes and enhance their interpretability, safety, and application across diverse fields. The second one instead tries to link together concepts ("features") inside a model into computational "circuits".

• The case against conversational interfaces: I’ve always been a bit afraid of saying this, but it’s great that there are others who do! Conversational interfaces are more than just tech fads; while they can't replace the efficiency of existing data transfer methods like keyboard shortcuts, they can enhance human-computer interactions by acting as an additional layer, potentially making our workflows as seamless as passing the butter at breakfast.

Biotech

• Paralysed man stands again after receiving ‘reprogrammed’ stem cells: A groundbreaking trial using induced pluripotent stem cells has enabled a paralysed man to stand independently and another to regain some movement, hinting at a future where spinal cord injuries could be treated more effectively and restoring mobility might be as common as a trip to the doctor.

• Embryo research, what time limit should we set?: The New York Times highlights the fascinating yet controversial journey of human embryo research, revealing how recent breakthroughs in culturing embryos could pave the way for groundbreaking solutions to prevent miscarriages and developmental disorders, all while stirring up the ethical pot.

• Hong Kong set to fund Chinese start-ups’ ‘jaw-dropping’ innovations: Hong Kong's capital markets are gearing up to unleash a wave of groundbreaking innovations from Chinese start-ups, thanks to new listing reforms and a push for long-term investments, suggesting that the next big tech breakthroughs are just around the corner.

• DeepMind spinout Isomorphic Labs raises $600m from Thrive Capital: Demis Hassabis’ Isomorphic Labs has raised $600m from OpenAI and Stripe-backer Thrive Capital, as it looks to advance its AI-designed drug candidates to clinical trials by the end of the year.

• Should human life be optimized?: Orchid is at the forefront of a genetic revolution that allows parents to screen embryos for a range of conditions and traits, from genetic disorders to potential intelligence, raising ethical questions about what it means to create life in an era where parents might select embryos like choosing a car.

Cool AI products we liked

• Digest high quality book summary in your preferred way: An AI knowledge agent, it helps interact with books and other content in a completely new way. 

• Mureka: epic new music creation tool.

• Iliane’s news: homemade news-reading agent that creates a daily brief of the most important things in the world.

Other things we liked

• Elon financial engineering his way to a super small QSB tax bill? xAI acquires X

• Why DePIN matters, and how to make it work: A16Z finally mainstreaming Decentralized Physical Infrastructure Networks. Wish we saw more investable companies in this theme.

• The Great Liquidity Shift by @ttunguz: 2024 saw a staggering 71% of exit dollars flowing from secondary sales rather than traditional IPOs or M&A, signaling a fundamental shift in liquidity strategies that could redefine how investors and employees capitalize on their stakes, making secondaries a permanent fixture that mirrors private equity's approach. We very much agree. Secondaries are structural and not a fad.

• The Average College Student Is Illiterate: A tenured professor laments the alarming decline in college students' reading and writing skills, likening their lack of engagement to being phone-addicted zombies, which raises urgent concerns about the future of education and the intellectual capacity of the workforce. Scary.

• Larry Fink’s 2025 Chairman’s Letter to Investors: Larry calls for the democratization of markets, not that many people have liked it.

• LIVE FROM VARDA: Delian Asparouhov, Founders Fund & Varda: Delian Asparouhov of Varda dives into the complexities of space regulations and the evolving relationship between tech and defense, highlighting the significance of the new 'Space Club' and the critical role of VC in shaping the future of aerospace.

With GPT 4o’s new image capabilities, it feels we’ve finally reached memeAGI.

This is the world we live in, we ship a ridiculously impressive, almost unbelievable AI innovation, and people use it almost exclusively to anime-ify the world.

But jokes aside, that’s a massive, massive leap for OpenAI’s image capabilities, which were way behind the rest in recent months - and keeps alive the tension between foundation cos and wrapper cos. It seems like each week we swing from one to the other.
When the large model companies push a new model out, everyone is extremely impressed and thinks it’s impossible to compete with them.

Then, within weeks, wrapper companies will build on top of that model and make incredible progress with 1/1000th the capex required, and everyone gets convinced the value is on the application layer.

By the way, not to be outdone, Google released Gemini 2.5, their most intelligent model to date.

Missing revenues, spies, China-made chips

Venture funding is always somewhere around a faithful instantiation of the power law and a hype-chasing game. 31% of VC funding concentrated in 20 mega deals last year, with Databricks, OpenAI, XAI and Anthropic taking the lion's share.

But it's still a risky game. Shortly after announcing AI sales rep Julian, recent industry darling 11x found itself in the barrel of a TechCrunch article revealing it may (or may not) have misrepresented revenues and customers. The company raised a $24m series A from Benchmark and other investors, followed by a $50m Series B led by a16z in September last year. 11x founder Hasan Sukkar denied most of the claims. And investors said it's a straight-up lie.

A few days earlier, another spy story shook the venture world. Human resource platform Rippling's CEO, Parker Conrad, claimed that competitor Deel planted a spy within Rippling to steal trade secrets, and unveiled how they figured it out.

But moving to the real spy world, we have China unveiling deep-sea cable-cutting technology, and a few things to flag on all sides of the Pacific and Atlantic:

• Torsten Reil on AI Warfare and Europe's Defence Preparedness

• What if Europe had to fight tonight?

• The EIF on what's to be done

• Michael Pettis on China's economic challenges and trade war risks

• One thread on Sinofuturism, from desert greening to lunar stations

• One piece on BYD's success and what's next

• Ant Group claims it's using high-performing China-made AI chips (some details if you don't have the Bloomberg subscription, here). Overall a good week for Alibaba.

And one story we loved reading on Ivory Coast's economic success.

From the physical world

Caleb Watney, co-CEO of the Institute for Progress, wrote a guest essay on what it means to do science. Yes, it is about return on investment. Yes, it is about making funding more efficient so resources are allocated where there's more leverage. But cost-cutting is not everything - "If a venture capital firm measured efficiency purely by how little money it spent, rather than by the returns it generated, it wouldn’t last long." The reality is that there's a curve with a maximum of societal ROI, which at some point decreases as capital allocation gets too haphazard. We must keep thinking of what good science means, and what efficient funding is.

Yet more odd quantum computing news

• The New Scientist reports skepticism regarding D-Wave’s quantum supremacy claims. The piece highlights that the lack of rigorous, independent benchmarking and the potential equivalence of D-Wave’s solved tasks to efficiently solvable classical problems undermine its claims.

• Will Kirkpatrick discusses the collaboration between IonQ and Ansys, and whether their touted milestone is to be taken seriously (spoiler: not really).

And a few more things from the world of atoms.

• An interview with FutureHouse co-founders Sam Rodriques and Andrew White, on Asimov Press.

• MIT's Raman's lab introduces STAMP, a one-step technique using 3D-printed stamps to pattern microtopography on hydrogel surfaces, enabling precise alignment of 2D contractile skeletal muscle monolayers for applications in drug screening and biohybrid robotics without requiring expensive microfabrication equipment.

• Nvidia and General Motors partner on AI for vehicles and factories. Another W for Nvidia's Omniverse.

• On the importance of new magnets.

• NATO Innovation fund leads a $25m series A in photonic chips company Camgraphic

Biotech

• On Modality Commoditization, by Elliot Hershberg
  Something here on pharma in China and why we keep writing about it.

• Pilgrim's Vision for Supersoldiers and Biothreat Defense

• One story on IVF and cancer

• Novo Nordisk's dilemma

• Study on Discontinuation and Reinitiation of GLP-1 Receptor Agonists

• Engineered Oncolytic Virus Shows Promise in Refractory Cancer Trial

• Lyra: A Compact Model for Biological Sequence Analysis

Other things we liked

• Bambu Labs new H2D all-in-one 3D printer, laser cutter and plotter (If you needed ideas for our Christmas gifts).

• A project to preserve memory, very three-body problem.

• On pursuing the 20-mile march - slow and steady, compounding progress will always win compared to random short bursts of effort and rest.

• a16z's deep dive into MCP and AI Tooling - MCP feels like just the new evolution of APIs, but everyone is going bananas about it. Surely exciting, a bit obvious - but a very good overview from a16z.

• On security challenges for AI agents. arcade.dev raises start-studded seed round.

• The Spark for Decoding Biology: a Graphic Novel

• On venture manager selection and a request for funds. The folks at Nomads.vc are probably the smartest people around when it comes to allocating to venture funds. We highly recommend reading and internalizing this.

Let us know what you think of this format, and if you have anything else we should read!

But back to the most interesting thoughts of the week.

Have we peaked?

Sadly, the title does not refer to the macro scenario, or to AI funding, or other tech aspect. It refers to humans.

I've maintained the view that humanity's genetic material as related to its body has peaked already (with medicine essentially tricking death and bringing so many more people into child-rearing age than what would have "naturally" happened, including myself), I never really thought that our "mental" genetic makeup would have peaked, and that if there is such a thing as "genetic intelligence", we still had a run up ahead of us.

Now, it's sadly looking like that might not be the case anymore: human concentration, verbal reasoning, and numerical skills are all on a downfall. The causes are almost comically easy to identify, but we really have no designed or even imagined solution as to how to reserve the course of things. We instead continue to create infinite amounts of content (be it epic entertainment or social media slop), constant comfort augmentation in all areas and extreme healthcare which only treats the symptoms.

We have slowed if not stopped natural selection, and looks like we're doing natural selection now as well.

Depressing to think about, BUT at least, the latest investment we've committed to, with our largest check, is a company who instead can dramatically accelerate natural selection for the better. Let's see if we can have an impact here.

The rest of the newsletter is probably more positive and AI developments—such as Stripe’s autonomous financial system concepts, Google’s latest edge-friendly model, and new robotic intelligence breakthroughs—to fresh perspectives on biotech and fertility. We see mathematicians wrestling with AI’s disruptive potential, venture capitalists speculating on the future of coding, and scientists converting skin cells to neurons at record rates. Macro / geopol forces come into view with discussions of China’s talent re-import for fusion research and Europe’s evolving approach to industrial sovereignty.

Enjoy.

Manus tools and prompts

Manus tools and prompts. GitHub Gist: instantly share code, notes, and snippets.

http://github.com

AI

• Stripe building more financial agents: a look at what an autonomous financial system might look like

• Building a conscious thing: "consciousness is when reality folds itself in a look". hyper-modelling AI loops to make conscious things.

• Simon Willison on how to use LLMs to write code: a full how-to guide to do proper AI-driven coding.

• Google introduces Gemma3: ai will run at the edge. This is the smartest model you can run on a single GPU.

• The divide between AI and mathematics: everyone is struggling to understand what AI means for their fields, and is scared that it will change everything, sometimes getting so advanced that there will be no way for humans to keep up in even understanding what the AI does. Mathematicians are no exception and this is a beautiful piece to read.

• Andrew Chen on what will happen with vibe coding: nothing groundbreaking, but good summary of where the VC thinking is today in regards to the future of software dev.

• The NYT: Are we prepared for AGI?. No.

• The first AI angel investor

Checking on how humans are doing

Maybe humans just peaked
The FT reports data indicating a global decline in human concentration, verbal reasoning, and numerical skills, suggesting a potential peak in cognitive capacity. Could be aging populations, devices-induced attention disorders, and educational disparities. But are we becoming a post-literate society?

Fertility on demand
That fertility will be entirely engineered, it has already become clear. This long-form by Ruxandra Teslo discusses why women choose to have kids later, what it means for fertility, and what can technology do to help.

Family planning via genetic matching
Nucleus Genomics introduces family planning by syncing genetic results with your partner, and scanning over 1000 conditions that could be passed to your kids.

Turning skin cells into neurons
Katie Galloway's lab at MIT develop a conversion method that increases the conversion of fibroblasts to motor neurons 100x, without going through iPSCs.

The physical world

Intel names Lip Bu Tan as CEO
Legendary former CEO of Cadence and founder of Walden International has the daunting mission of leading Intel out of stagnation. Intel stocks went stonk, gaining 25% since the announcement.

Too cheap to meter
The second film by Story Co., on what energy was supposed to be, and will be if we get serious about it.

Global Stellarator Coil Optimization with Quadratic Constraints and Objectives
QUADCOIL introduces a fast, global coil optimization method for stellarators, targeting linear and quadratic current functions. It enhances physics objectives like Lorentz force and curvature, surpassing limitations of NESCOIL and REGCOIL.

A gemini robot
Google DeepMind unveils Gemini Robotics, a vision-language-action (VLA) model built on Gemini 2.0, integrating physical action as an output modality for direct robot control. Gemini Robotics-ER, enhances spatial reasoning, enabling embodied reasoning for roboticists to interface with existing controllers, tested on platforms like ALOHA 2 and Apptronik’s Apollo humanoid.

A bit of an overkill
Figure AI reveals BotQ, their new infrastructure to manufacture robots. The video is impressive, but the announcement has received mixed reactions. A take on why this could be a bit of an overkill, for a 12k robots/year production line, here.

Bridges of spies

From China with love
The stakes in AI development are high. Dario Amodei reveals that nation-state spies are targeting the company's AI models, where a few lines of code can be worth $100s of millions.

And some more China content:

• If China keeps re-importing talent for fusion energy research

• All the numbers of the China's dominance playbook
  

On Europe's industrial prosperity
A few suggestions by Sam Cash on European sovereignty, from procurement to M&A and soft power.

And a few more updates from Europe's defence awakening:

• Estonian startups sees an opportunity in ReArm Europe

• Announcing Project Europe

• Eutelsat 7x's its market cap in a week

• The 2025 European deep tech report

• Despite the old challenges - Northvolt files for bankruptcy in Sweden
  

On the US:

• The cost of US reindustrialization

• Breakthrough Energy cuts jobs across Europe, policy and partnership teams

• a16z's DoD contracting cheatsheet

• SpaceWERX selects eight companies for $440m in Strategic Funding Increase agreements

Other stuff we liked

• Factorio learning environment

• What if we preferred AI-generated poetry

• Klarna files for IPO

• Turning $600m valuation into a business

• Selling enzymes to pharma sucks

• Apple's artificial generally bad intelligence

• Azeem Azhar: How AI might change venture capital
  
  
  

Let us know what you think of this format, and if you have anything else we should read!

But this is also a week that saw one of the worst ever stock crashes, with many questioning if the US administration is playing 5D chess or just completely insane. Many institutionals are closing their USD longs, and just pouring more capital in Europe and China.

The speed of change seems to be pervasive, not remaining confined to technology but touching everything around us.

This seems also evident in the speed of technology window opportunities compressing more and more, with the window to create massive AI companies almost largely closed after just a few years (unless the Deepseeks of the world just plainly destroy the entire window).

Prompt wrappers

This week we've been able to read through the system prompts for v0, Manus and even Cursor. Reading through them really makes you wonder what the staying power of all of these tools will actually be.
They all seems to rely on Sonnet heavily, with really most of the heavy-lifting coming from Anthropic (we're not surprised to see them raising a Series E at $60B+) (we also hear that OpenAI is raising $40B at $300B valuation 🤯).

Most people are already switching from one tool to the other with very little loyalty. And so the age-old "will the models or the application layer capture the most value" question seems to not be fully solved at all as of yet.

Have fun and look under the hood to how these tools work!

Manus tools and prompts

Manus tools and prompts. GitHub Gist: instantly share code, notes, and snippets.

http://github.com

GitHub - x1xhlol/system-prompts-and-models-of-ai-tools

Contribute to x1xhlol/system-prompts-and-models-of-ai-tools development by creating an account on GitHub.

https://github.com

Manus panic

• Leave it to Manus (do let us know if you have an invite for us)

• Multimanus: welcome to dystopia!

• One demo: 14 min demo of Manus in action.

• OSS Manus: https://github.com/camel-ai/owl. We already have an open source implementation of what Manus does, based on Camel AI. It seems the staying power question is even more relevant than we can imagine.

Superintelligence strategy
Dan Hendrycks, Eric Schmidt and Alexandr Wang propose a strategy around Mutual Assured AI Malfunction (MAIM, our times' MAD) for nonproliferation of weaponizable AI capabilities, and offer a framework for competitiveness.

And an article by Hendrycks and Schmidt on the nuclear-level risks of superintelligent AI.

• Anthropic’s Recommendations to OSTP for the U.S. AI Action Plan

• OpenAI: How we think about safety and alignment

• Reflection: a path to superintelligence + a piece by Sequoia

Other AI

• Good artists copy, great artists steal.

• How Intelligent Choice Architectures Rewrite Decision Rights

• CuspAI recruits Yann LeCun to advisory board

• Anysphere in talks for close to $10bn valuation

The arms race

ReArm Europe
Ursula von der Leyen announced a $800bn, five-part plan to boost Europe's defence capabilities.

Update from a world with European fiscal packages, security funding, and US tariffs.
The S&P 500 is down 1.8% in 2025, European stocks (.STOXX) have risen 9% to a record high, and Hong Kong tech stocks (.HSTECH) surged nearly 30%; halved bullish bets on the U.S. dollar to $16 billion; pushed the euro to a four-month high above $1.07.

Shield AI raises $240 million
The defence aircraft company has raised at a $5.3 billion valuation from investors including L3Harris and Andreessen Horowitz to expand Hivemind Enterprise, an AI-driven autonomy platform accelerating development for OEMs and primes in the industrial base.

Alpine Eagle raises €10.25 million
Munich-based startup Alpine Eagle has secured seed funding from IQ Capital, General Catalyst and HCVC to advance its Sentinel counter-drone system, an airborne platform using AI-driven modular sensors and kamikaze interceptors for cost-efficient threat detection and neutralization.

Can the US switch off Europe’s weapons?
As it turns out, “Most European militaries depend heavily on the US for communications support, for electronic warfare support, and for ammunition resupply in any serious conflict.”

The future of labs

Novo Nordisk Says AI Is Reliable Enough to Produce Sensitive Documents. And two more notes from the labs of the future.

• Towards an AI co-scientist
  A multi-agent system built on Gemini 2.0 employs a generate-debate-evolve approach for hypothesis generation, using a multi-agent architecture and tournament evolution process. The system demonstrates enhanced biomedical discovery, validated through drug repurposing, target discovery, and bacterial evolution studies.

• Lila Sciences raises $200 million to build scientific superintelligence
  Lila announces a platform integrating AI with autonomous labs to conduct thousands of parallel experiments, with the lofty goal of "reinventing the scientific method". Also in the NYT.

• The Einstein AI Model
  Thomas Wolf, co-founder of Hugging Face, discusses scientific research, genius, creativity, and move 37.

Research

A safer, better Ozempic
Stanford Medicine researchers have identified a peptide that suppresses appetite and reduces body weight in mice without side effects like nausea or muscle loss.

Cultivated autologous limbal epithelial cell (CALEC) transplantation for limbal stem cell deficiency
A phase I/II clinical trial (so feasibility and safety) evaluates cultivated autologous limbal epithelial cell transplantation to treat limbal stem cell deficiency (LSCD), demonstrating corneal regeneration and improved vision.

Emerging supersolidity in photonic-crystal polariton condensates
Researchers have transformed laser light into a supersolid state, achieved by firing a laser at a gallium arsenide semiconductor with microscopic ridges, creating polaritons - hybrid light-matter particles that self-organize into a crystalline lattice while maintaining superfluidity.

Weeping willow
A team from the USTC in Hufei, China reports that their 105-qubit Zuchongzhi 3.0 quantum processor completed a random circuit sampling task in seconds, a feat they claim would take the Frontier supercomputer 6.4 billion years, outpacing Google's Willow by a million times. All usual QC caveats apply.

Other stuff we liked

Sesame Conversational Demo: welcome to the future.

"Sing, o Muse": fiction AI writer.

Speed to proof/value

A run in Shenzhen's Talent Park

Eric Schmidt joins Relativity Space as CEO: the co was having a hard time, and Eric comes in as a white knight (and majority shareholder). Let's see how this one goes!

If you think modern code review is fine as is, turn back

Playbook to fine-tune and deploy LLMs

Buy every possible lottery ticket: there was a Texas Lottery arbitrage opportunity. In April 2023, an entity called Rook TX effectively purchased the jackpot, collecting a one-time payment of $57.8 million, by acquiring virtually all of the 25.8 million possible number combinations.

Just blink once

Let us know what you think of this format, and if you have anything else we should read!