I am at the tail end of a transpacific flight, crossing 14 time zones in about 13 hours after two shorter flights earlier in the day, and strangely enough, I feel fantastic. Despite trying to travel less, I still take on average 1-2 flights every week, often with time zone changes. Not as much as a decade ago during the crypto boom, when Sean and I were building First Block Capital and Hut 8, and there were weeks when we spent three nights sleeping on airplanes. There were even times when I had to cut business trips short and fly home because I was physically exhausted. I probably still take too many flights to be considered healthy, but what is interesting is that despite being a decade older, I actually feel better than I did during those days. As I sat on this flight, feeling surprisingly energetic ten hours into an intercontinental journey, I started wondering why. Does life get better after 50? For me, it seems to.

What has changed?

Earlier in the flight I was reading two Nature papers on biological clocks and the use of principal component analysis to calculate biological age. The idea is that aging is incredibly complex, but certain “master dials” explain much of that complexity. That made me think about my own life. What dials have changed over the past decade that might explain why I feel so much better today?

Time management

Interestingly, I still have a huge lack of time today, but the difference is that I manage it very differently.I have become much more selective about meetings than I was in the late 2010s. No more coffee meetings, lunches, or dinners without a clear purpose. Instead, I prioritize spending time with friends and family. I try to keep Zoom meetings short, ideally 30 minutes, although that is not always possible. I rarely do in-person meetings anymore unless I am attending a conference. Depending on where I am in the world, my first call is usually at 7 or 8 a.m., and I try to finish my last one before 10 p.m. No more calls before dawn or late into the night.

AI and X

AI has helped enormously as well. In the past, I would spend 1-2 hours every morning catching up on news across all the sectors I follow. That meant opening dozens of browser tabs and scrolling endlessly through Twitter. Today, AI agents prepare summaries while I sleep. By the time I wake up, I have a good overview of the most important developments in AI, biotech, crypto, financial markets, and other topics that interest me, along with the most important discussions happening on X. In 15 minutes, I know what happened overnight.

I have also built a personalized dashboard on X with curated lists focused on niche topics that I want to understand deeply. That has been a real game changer, because I don’t miss any important news anymore, and I have one place where I can find all the latest developments on topics that interest me.

A state of flow

But the biggest difference is something harder to quantify: For the past few months, I seem to have been in a near-constant state of flow. Looking back, flow for me is not about being relaxed, because my days are still packed. It is the feeling that the things consuming my energy are also giving energy back. Ten years ago, almost everything on my calendar cost me energy. Today, many of the things on my calendar recharge me.

My to-do lists used to feel endless. No matter how much I got done, there was always more work waiting. These days, I regularly see the light at the end of the tunnel, even if I never quite get to zero.

Having a purpose in life

In many ways, that shift began when Sean and I started Helixion Therapeutics a couple of months ago. I feel like I am finally working on something that is deeply meaningful to me. Designing personalized cancer vaccines and making them more accessible feels directly connected to improving people’s lives. That gives me a different kind of energy than many of the projects I have worked on in the past.

I have always enjoyed being ahead of the curve, and building an AI-first biotech company at a time when many organizations are still figuring out how AI fits into their workflows is intellectually exciting. Every day I learn something new. I spend hours reading biology and immunology papers every week together with Claude, exploring ideas, asking questions, and trying to understand where the field is heading. Instead of draining me, it gives me energy.

Alcohol

The changes did not stop there. Because I now have regular early-morning and late-evening meetings, I stopped drinking alcohol during the week. Over time, that reduced my desire to drink on weekends as well. I still enjoy a few glasses of wine with friends on my boat or over a nice dinner, but that is about it. In airport lounges and on planes, I now drink sparkling water or occasionally I splurge on a Diet Coke. I do not miss alcohol nearly as much as I expected.

A decade ago I drank almost every day. Before an overnight flight eastbound, I had a routine: 2 beers in the lounge, a steak and 2 glasses of California Cabernet Sauvignon on the plane, then I popped melatonin and went to sleep. Today I still take melatonin for jet lag and I may order a steak on the plane, but for the rest it’s just sparkling water when I am in the air. Part of this change may be a consequence of the flow I am experiencing, but I also suspect that reducing alcohol actually reinforces it. Even 1 or 2 drinks noticeably affect my sleep scores. I still enjoy a drink every now and then, but hangovers are a thing of the past.

Diet

I had already started intermittent fasting years ago and still do it today. More recently, I have focused on eating more protein and vegetables and fewer carbohydrates. I also serve myself smaller portions because I tend to finish whatever is on my plate. I still enjoy chocolate or chips occasionally, but I no longer feel the need to snack nearly as much as I used to.

Exercise

Exercise has become more consistent as well. Since starting Helixion, I have tried to get to the gym almost every day before dinner. I typically run five to eight kilometers and do some weight training. Nothing extreme, but enough to feel that I have accomplished something. I used to listen mostly to podcasts while running. I still do that sometimes, but more often I listen to upbeat music. I created a few playlists that are perfect for 30-45 minute runs, and they seem to pull me even deeper into flow. Instead of filling my head with new information, the runs give my mind space to wander. I regularly come up with new ideas while running and I occasionally stop to make notes on my phone before I forget them.

Supplements

I have also become much more consistent with my supplements. I take around 15 of them daily. In the past I would often skip a day or 2 when traveling or during busy periods. Now I place them on my hotel desk or kitchen counter where I cannot miss them. If I forget to take them during the day, I take them before bed. It may not be the ideal timing, but it is certainly better than skipping them altogether.

Sleep

All of this has contributed to better sleep. In my forties, I often struggled to fall asleep and frequently woke up during the night. Today I usually fall asleep within 10 minutes and sleep through the night. I try to be in bed by 10:30 p.m. with a book or podcast rather than working late into the evening. I still probably do not sleep enough. Most mornings I wake up naturally after about 6-6.5 hours, but I feel rested and ready for the day. That is very different from ten years ago when I relied on an alarm clock and often wished I could stay in bed longer. Whenever possible, I also take a 20-30 minute nap after lunch, which really improves my afternoon productivity.

Living in the tropics

Another factor may simply be where I live. Spending most of the year in the tropics means warm temperatures and sunshine almost every day. I spend much more time outdoors than I did when I lived in Vancouver, Europe, or China. It feels great to open the curtains and balcony doors in the morning, feeling warm air coming in, and stepping outside without having to think about the weather.

Taken together, all these changes have created a continuous sense of flow that has now lasted for months. I feel more productive, healthier, and happier than I did a decade ago. My personal relationships are stronger, although there is still room for improvement there. Looking back, I do not think there was a single breakthrough or silver bullet. The improvements came from dozens of small decisions that reinforced one another over time.

I reset my biological clock by 11 years

Recently, I was invited to submit my biomarkers for Professor Brian Kennedy’s LinAge2 biological clock. Those were the same Nature papers I had been reading earlier on this flight. I honestly did not know what to expect. Part of me worried that 30 years of working too hard would have left visible scars. But he results surprised me: At 53 years old, my measured biological age came back resembling that of someone who had not yet turned 42! Of course, biological age models are imperfect, and I would not base major life decisions on a single number. But the result was still encouraging, it suggested that the changes I have made over the past few years may be having a meaningful impact.

Conclusion

If there is one lesson I take away from all of this, it is not that purpose alone matters, nor that supplements, exercise, sleep, or nutrition are the answer by themselves. It is the combination of all of them.

For me, the biggest shift was finding work that I genuinely care about. That seems to have created a positive feedback loop that improved many other aspects of life. But the reverse is also true. Better sleep, better health, less alcohol, and more exercise make it easier to do meaningful work.

If you can find something that excites you, whether it is your job, a side project, volunteering, building a business, creating art, or helping other people, it may be worth pursuing. AI has dramatically lowered the barriers to starting new projects and learning new skills. Many opportunities that were inaccessible only a few years ago are now within reach.

Life can be amazing, but you have to make it amazing yourself. Small improvements compound over time, just like investments do. Start early and keep working on it, it will pay off eventually.

Interested in doing the LinAge2 biological clock yourself? There is no customer-facing website yet, but you can get in touch with Avely@beyondclock.com

I posted a thread on X about a month ago laying out why I think the market has World (formerly Worldcoin) wrong. At the time WLD was trading around $0.25, down from nearly $10 at its 2024 peak. It’s up roughly 32% since then, but that barely registers against the size of the opportunity I see here. The more I use AI day-to-day (in my work at AI-first Helixion Therapeutics, in research, or in just navigating the internet), the more obvious it becomes that what World is building will be needed by everyone, and sooner than most people realise.

Let me clear, this isn’t a shill for a crypto token. I own a small amount of WLD, genuinely insignificant in the context of my portfolio. I’m writing about it because I think it’s one of the more misunderstood projects in crypto right now.

The problem nobody has solved yet

Sam Altman put it pretty direct in a recent conversation with Cosmo Jiang from Pantera Capital: AI has advanced faster than even the people building it expected. We’re rapidly heading into a world where AI-generated content outpaces human output, where you can no longer trust that the person on the other end of a video call, a dating app, or a customer service chat is actually a person.

Spend a few minutes on X and you already see what this looks like. Bot armies and AI-generated images passed off as real photos are everywhere. And we’re just at the very start of this curve!

The solution has to be a way to prove you’re human. This has to be done cryptographically without giving up your privacy in the process, and that is exactly what World ID is.

The iris scan

The iris scan that set World apart is where most people get stuck, and I understand why. Walking up to an Orb to have your eyes scanned sounds scary or even dangerous, if you don’t know what’s happening underneath. I did it myself almost 3 years ago, and did a lot of research before doing the scan.

What happens is that the iris is used to generate a one-way cryptographic hash. The hash proves you are a unique human without storing identifiable biometric data anywhere. You can then prove you’re a real person to any service that integrates World ID without revealing who you are. It’s privacy-preserving in a way that traditional ID systems simply can’t match, and a lot more privacy-preserving than handing over your passport to yet another centralised database.

If you do the deeper dive, the iris concern is largely a non-issue, but it is clearly part of why the market is discounting this so heavily.

The first use cases are already live

What surprised me when I dug in is how much is already happening, the public partnerships alone are quite big. Tinder started to use World ID to fight bots pretending to be humans, Zoom is integrating it for verified human meetings, and the governments of Taiwan and Malaysia are exploring it for citizen services

But the bigger picture Pantera laid out in a YouTube video goes far beyond these integrations:

Digital advertising is roughly a half-trillion-dollar industry. Advertisers don’t want to pay for bot impressions, and that problem is about to get exponentially worse. Capture even 5–10% of that industry through verified-human impressions and you have a major business on its own.

Government services like social security distribution, voting, or welfare. The fraud savings alone are enormous, and governments are actively looking for solutions here.

Agentic commerce is the one with the biggest potential. As AI agents start doing real economic activity on our behalf (things like booking travel, making purchases, executing trades), every transaction will likely need a human in the loop somewhere for accountability and trust. I think World ID is the natural solution for that.

Credential storing Less important for now, but a huge opportunity in the longer run. Once you’ve proved you’re a unique human, you can append verified credentials on top: driver’s license, education, employment history. It becomes the base layer for digital identity.

The math

Pantera’s thesis is that if World captures a meaningful share of these markets at $5–10 of revenue per user per year, this becomes a $50 billion free cash flow protocol. There are roughly 4 billion smartphone users globally, that’s the addressable population, and the infrastructure is being designed to scale to that level.

Now apply the token math. Fully diluted, with every future unlock through 2032 priced in, there will be a maximum of 10 billion WLD tokens. Mature, successful tech companies trade at 25× free cash flow or higher. Apply that to $50 billion FCF and you get a $1.25 trillion market cap, which works out to roughly $125 per token on a fully diluted basis.

WLD is sitting around $0.34 today. Even with all the dilution baked in, the math points to something like a 300–400× return over the next several years if the business model works. That’s maybe not Bitcoin-in-2013 territory, but it isn’t far off, and the comparison feels appropriate to me. Investing now in World means you are early, just like BTC in 2013. Investing now is far outside the consensus, far outside the Overton window, just like BTC 13 years ago. And the use cases for World are not obvious until you think about it for a while, just like with BTC.

The OpenAI IPO

The near-term catalyst I’m watching is the planned OpenAI IPO, because Sam Altman is (deeply) involved in both companies. When OpenAI goes public, World could benefit as well. Retail investors who discover OpenAI through its listing are going to find World as “the other thing Sam built.”.

The fact that Elon lost his court case again OpenAI (mostly because he filed too late), was a good thing not only for OpenAI but indirectly also for World.

The risks are quite big though

A couple of things about the risks, because there are some real risks here. First of all the dilution schedule is heavy and runs through 2032. Dilution gets reduced in July this year, but if token usage doesn’t go up significantly this could keep pushing the price down. Next to that, the regulatory environments are still figuring out biometric ID, and at least some jurisdictions will push back. They simply don’t understand it, just like many still don’t understand Bitcoin (I always see that as an opportunity). It also means that adoption could be much slower, or that adoption would stop. There is a lot of execution risk between here and a half-trillion-dollar revenue base. But I feel that once adoption goes up most of the execution risk will be less important.

Finally, Sam Altman can be a risk. He has become a very controversial figure over the past years, and although his involvement helped World a lot, it could also become a problem. Hard to judge how this will develop, but keep it in mind.

None of this is financial advice, just my personal opinion, and my opinion is often very different from mainstream investors. I have been wrong on many crypto bets and I’ll be wrong on plenty more. But when I look at where AI is heading, and at what is needed to keep the internet usable for humans, World is the best answer I have seen so far.

I had a long conversation this weekend about what AI-driven deflation actually does to a portfolio. We kept coming back to the same uncomfortable question: if AI and robotics really compress prices the way we think they will, does the old fiat-era playbook still make sense?

In my March post on the AI abundance economy I argued that AI could break the standard “productivity goes up, stocks go up forever” story. But I didn’t spend enough time on what deflation itself actually is, and I didn’t really challenge one of the most important beliefs of almost every investor I know: that cash is trash.

After thinking this through, I don’t believe that this is automatically true anymore, because it depends on what kind of deflation you are in.

Two very different kinds of deflation

Most people, especially people trained in modern economics (like myself), think of deflation as one thing: a dangerous spiral where demand collapses, debts crush borrowers, and central banks have to rescue the system. I think that can happen, but it is only one type of deflation (‘bad deflation’), and it is the only one fiat economists are trained to worry about.

There is another kind, and that is the kind I think we are about to experience ourselves. When entrepreneurs produce more with less, and when AI and robots collapse unit costs, prices fall because production becomes genuinely cheaper. That could actually be a good thing, because in that case: Productivity goes up → lower per unit costs → falling prices → higher real purchasing power. I call this good deflation.

The 19th century industrial boom under the gold standard looked a lot like this. Real wages went up, the economy grew, and nominal prices fell. I feel we rarely talk about it because the post-1971 world (when the US abandoned the gold standard) has trained everyone to fear any price declines as if the world would come to an end.

AI deflation looks much more like the 19th century version than we have seen in the 1930s, and that means it will likely have a positive effect.

What happens to revenues and profits

For most production companies, nominal revenues will almost certainly decline. If a robot-powered factory produces 10x the output at a tenth of the labour cost, competition drags prices. This is great for consumers, but for most companies revenues will go down, unless volume explodes enough to compensate the lower price per unit.

The picture for profits is more interesting: A small minority of companies will redeploy capital into AI-based production faster than their competitors. Their unit costs collapse faster than their prices, so their margins increase and profits rise. These are the AI-first players, the early adopters.

The majority of companies won’t adapt in time, however. Legacy costs and processes will stop them from acting fast enough. I believe they are doomed to fail.

I think the total profits in the economy might go up, but they will get concentrated in very few companies. Profits for the average firm will fall. The slow movers get killed.

Index investing in times of deflation

Stock prices are discounted future cash flows. In good deflation, broad indices like the S&P 500 probably decline in nominal terms. Revenues go down for most companies, while multiples go down because investors price in lower growth.

At the same time, AI and robotics leaders could do extremely well in real terms. But watch out: because AI changes so extremely fast, even today’s winners can become tomorrow’s losers. If you don’t keep up and don’t keep investing in AI (=laying off most of your workers), others will pass you and it’s over and out. That is why I believe that all of a sudden we will see a huge increase in unemployment. Companies are forced to lay off most people in order to survive.

I keep coming back to the point I made in March: even the winners risk getting squeezed by taxes, once governments realise the labour tax base has evaporated and only the AI giants have cash flow left to bleed.

So broad market indices like the S&P 500 probably do not “moon”. Only ownership of the few winners will make you money, but you have to pick the right ones, and be willing to exchange them for better ones if you feel they are getting left behind. This is exactly why I am concentrated in AI and robotics stocks (both in the US and in China), rather than holding broad ETFs.

Holding cash

In a pure innovation-driven deflationary world, cash gains purchasing power. If prices across a broad basket of goods fall 10-20% per year, cash effectively earns a 10-20% real return without risk. That is the exact opposite of the world I have operated in for my entire investing career. In the fiat era, “cash is trash” has always been my slogan. Inflation ate your cash, so you had to be invested in the market.

But under good deflation, cash is no longer trash. It’s then an asset that becomes more valuable over time, when prices go down (with every dollar you can buy more next year than you can now). But investing in equity will still make you a lot more money. If you hold only cash, you preserve wealth but you miss the real wealth creation.

Cash will be is a reasonable short-term parking asset, but it is still suboptimal for the long run if you are an investor. Selective exposure to AI leaders beats cash, while broad index exposure probably does not. Stock indices might actually underperform cash in nominal terms. In our current economic world that is a strange sentence to write, but it is where my logic lands.

This is the opposite of the current fiat-era playbook. Under inflation, you have to be invested or you lose. Under good deflation, you can sit in cash without bleeding, but the best move is still to allocate capital to the narrow set of real winners. The “hold the whole market” advice from most financial advisors was written for an inflationary world that may be ending.

Bitcoin

And Bitcoin? Bitcoin has a fixed supply of 21 million. In a world where productivity gains are not diluted by money printing, a scarce, non-sovereign savings asset is exactly what you want. Purchasing power of BTC should rise with the deflation rate, on top of any adoption-driven appreciation. In Austrian economics terms, BTC is what honest money looks like in a world where the state has lost its monetary monopoly.

In a prosperity boom where people save more, demand for superior savings instruments will go up. Fiat cash will give you more purchasing power, but bitcoin will go up more, because it has the same monetary properties plus a fixed supply.

The nominal price of BTC measured in a deflating fiat might not “moon” the way people hope, because everything denominated in that fiat is falling. But the real value, what a Bitcoin can actually buy, will explode.

Still, there is a short-term risk: if broad markets crash, BTC often gets sold with everything else. But long term, it is the asset most aligned with a deflationary regime. That is why even though I don’t have a lot of Bitcoin exposure anymore, I still think it should be part of every balanced portfolio.

Open issues

There are still a couple of things I am not sure about. The first one is how much governments will distort this. If governments keep printing money the way they do right now, it may offset the deflationary effects. In that case scarce assets like Bitcoin would be very good to hold.

Taxation is another issue that is unclear. If everyone loses their jobs, someone has to pay for them to survive (I wrote about this in March as well). If the AI giants become the de facto tax base for a universal basic income, the after-tax returns on investing in these companies could be much lower.

Finally, it’s unclear who large the group of winners will be. If it is 10 companies globally, picking the right ones is extremely important and getting it wrong is catastrophic. If it is a hundred, diversification starts to work again. My guess is closer to ten, and they will all be AI or robot-related.

Closing thoughts

The fiat-era investor assumes three things: inflation is permanent, cash loses, and investing in an index always wins in the long run. If AI delivers the productivity revolution I think it will, all three of those will be wrong. Inflation won’t be permanent, and we will be heading into structural deflation in many categories.

Cash does not necessarily lose anymore, because in times of deflation it can earn a real return by doing nothing. Lasty, index investing may not be a smart strategy anymore. The index will be mostly a portfolio of slow movers, that get outcompeted or taxed into irrelevance.

The right portfolio in this world looks nothing like the 2000s or 2010s playbook. It is concentrated in hyper-winners, it holds Bitcoin as the non-sovereign savings asset, and it can hold cash more comfortably than before.

As always, this is not investment advice. It is just how I am thinking about it right now, and I am still figuring this out myself.

In my last post of 2025 I wrote about my plans for the new year and said I would go all-in on AI. That’s what I did, and last week I shared that Sean and I had started biotech company Helixion Therapeutics and built a model to rapidly generate personalized cancer vaccines. If you had told me three months ago we would have a working cancer vaccine model by the end of March, I would have said that’s impossible. Well, it turns out it is very possible. It’s not perfect yet, for sure, but it’s unbelievable how fast AI has developed over just the past couple of weeks.

I have a degree in economics, so I have no medical or a technical/science background (although I read a LOT of scientific articles), and I’m certainly not a PhD. Despite that, Sean and I managed to build this in an incredibly short time without spending large amounts of money. That alone should tell you something about where we are with AI and science.

The power of cross-pollination

Over the past couple of months, and especially the past couple of weeks while working on the Helixion model, my mind got flooded with new ideas related to AI, genetics, medicine, and DNA. We’re focused on personalised cancer vaccines right now, but I believe we can eventually make this much broader (Sean will try to stop me when he reads this, he is much better at focusing, and he is right of course, but I like to dream).

Drug repurposing, for example, is low-hanging fruit. You can build AI models that look at the existing literature (and not just Western like most people do) and come up with drug applications that people haven’t considered. It’s really not that hard if you understand how AI works and if you know what to look for in the literature. This is not a new idea, I know others are working on this as well, but I am not sure if they are AI-first as well and if they are taking the same approach.

Here’s what I think is happening: a PhD goes incredibly deep into one field. That depth is valuable, but AI changes the equation. You no longer need to go as deep into a single field. It’s now much better to go fairly deep into several fields and then find the overlaps. That’s where the novel ideas live and where new science will be created. AI can help you with that, because AI is a PhD in every field and can explain it to you like you’re 5 years old (try it, it’s fun if you prompt it to do that sometimes).

I didn’t know a thing about biology, genetics, cancer vaccines, proteins, peptides, or neoantigens a couple of months ago. I didn’t know how mRNA vaccines work (except for the basics) or how you create them. I’m certainly no expert yet and won’t become one, but I now know enough to come up with new ideas. Maybe even faster than some PhDs, because I connect dots across fields instead of digging deeper into one.

I’ve done this before: The FBC Bitcoin Trust (that became a publicly listed Canadian Bitcoin ETF years before the first US one was listed) came from combining knowledge of traditional finance and Bitcoin back in 2017, when nobody was thinking about it. Bitcoin mining was similar: two different fields, which allowed us to raise money for Bitcoin mining in public markets with Hut 8. This meant we could keep (‘HODL’) our mined Bitcoin, instead of selling it to pay for service and power. At the time that was revolutionary. AI makes this kind of cross-pollination dramatically easier and faster.

AI as the ultimate research partner

One reason we were successful with our cancer vaccine model is that AI helped us find the right datasets fast. Some were license-only or not for commercial use, so we asked AI to find us free alternatives (we are financing this ourselves for now). If they were available AI would eventually find them, reformat the data, and get it into our model. These things literally weren’t possible until even a few months ago.

AI finds every paper you’re looking for and it explains or summarises these papers for you. It’s patient, always on, always enthusiastic. You give it a huge task and it’s like an eager intern who genuinely wants to do the work (“Yes, good idea!”). I make mistakes constantly, and I ask AI to explain basic things to me over and over. It doesn’t care, it just does it over and over again without judging me.

I have four different AI instances open on my laptop at all times (ChatGPT, Claude, Gemini, Grok) and I go from one to the other, testing what one tells me by discussing it with another. I have preferences for certain tasks and I know each model’s strengths and weaknesses pretty well by now.

What universities should understand (or learn)

This brings me to a harder point: AI makes me realise how slow science is. Scientists are great people, don’t get me wrong, but they’re not entrepreneurs. Their speed is different from our speed. In medicine the traditional path of Phase 1, 2, and 3 trials is a real bottleneck. We need to come up with alternatives, and maybe AI can help us get there.

I was recently looking at my master’s thesis that I wrote full-time during a couple of months. I believe that anybody with good knowledge of Claude Cowork could now do the whole process (defining a problem, collecting the data, building spreadsheets and models, running statistical tests on them, drawing conclusions, and then writing 40 pages about it full of graphs) in a weekend, maybe even in one day. Claude comes up with ideas, finds the data for you (that took me weeks in 1995), formats it, runs regressions on it and comes up with conclusions. I don’t know if universities realise it, but I can’t imagine that smart students who are AI-native spend more than a few days on writing a full master’s thesis.

If you take that a step further, I think that one good AI-first PhD can now do the work of 10, maybe even 20 PhDs who just use AI as a more expensive version of Google. That’s a huge opportunity if you’re a PhD or researcher, but it means PhDs need to embrace AI fully, right now. If they don’t spend serious time learning what it can do, they will be left behind by people who understand AI and can work with it.

If I’d run a university right now I would completely change its curriculum. Become AI-first before AI takes over the world right under your nose. This doesn’t take much time, because you can do that with AI as well! It’s actually an easy and a fun exercise: feed the syllabus and the required reading into Claude and let it design a 6- to 9-week course for each subject fully built around AI, or as a next step even a personalised course for every student. Maybe even more important, force your PhDs and professors to become AI-first, and literally check every week how many tokens they used. I can guarantee you that the number of publications would skyrocket.

The acceleration math

2025 was the year of AI coding and vibe coding, experienced developers coding together with AI, or noobs like me coding apps that they couldn’t create just a few months earlier. The best coders don’t code anymore, they just give AI the direction and check the output.

2026 is the year of agentic AI. Multiple AI agents running tasks for you simultaneously, while giving tasks to subagents. As an example, you could literally ask an AI agent to redesign university courses and make the AI-first, give it all the information and the next morning you have a new curriculum. If you haven’t tried agents yet, do it. I can guarantee you it’s life-changing, you will suddenly understand where this is all going.

Here’s how I see the acceleration: this year, we will get scientific breakthroughs that would normally take five years. Next year, that doubles, we’ll achieve in one year what would otherwise take ten. By the end of 2027, we could be where we would normally have been in 2040. That’s how fast this is going.

Coming back to medicine, I strongly believe that within a few years we can solve most diseases. I’m now starting to think we will be able to turn aging around, so we will actually age backwards, which means we could eventually live far longer than anyone currently imagines. The science is pointing in that direction, and AI is the accelerant.

The double-edged sword

But it’s not all bright, because most people will lose their jobs in the coming years. For them, this will be terrifying. The polarization we already see in the world will only get worse with AI. We may face serious problems (economic disruption, social unrest) that are genuinely hard to solve.

Unfortunately our governments still live in the old world, they don’t see the tsunami that will hit us. The old science world too (universities, research centers, regulatory bodies) still want to use frameworks designed for a much slower pace of discovery. That needs to change if we want to capture the full potential of what AI makes possible.

Where this is going

The future is scary but will eventually be bright, unless AI kills us (which is a real possibility, but that’s for another time). I am convinced that science will see a great acceleration starting this year already and it’s only speeding up. Within two years we’ll be at a level where we would normally be by 2040. AI companies will soon (this year?) release specialised versions for law, accounting, medicine. These will be tools that don’t just assist but actually replace entire functions with full accountability. I now realise that these AI companies will become the biggest companies in the world before the end of the decade, they will simply take over the world.

What we are seeing now is just the beginning, things will only go faster from here. Don’t get left behind.

I’ve been thinking about longevity since at least 2017. Back then I wrote about it as an investor thesis, the idea that extending human healthspan would become one of the defining technological challenges of our generation. It was somewhat abstract at the time, interesting to think about over dinner and easy to file under “future trends.”

Then I lost two friends to cancer. That changed a lot for me, because something abstract suddenly became personal very fast.

When someone close to you dies from cancer, you go through the usual stages. But at some point, if you’re the kind of person who builds things, you start asking different questions. Not “why did this happen” but “what is actually being done” and “where are the gaps.” I started reading a lot about cancer. First books, then online articles, and eventually I ended up with papers about cancer immunotherapy, personalized medicine, and neoantigen vaccines. I talked to people in the space and then I found a gap that shouldn’t be there.

The cancer vaccine opportunity

What did I learn? Your immune system is designed to find and kill cancer cells. Cancer cells have mutations, and some of those mutations produce tiny protein fragments (called neoantigens) that get displayed on the cell surface like a flag. In theory, your immune system should see those flags and attack them. In practice, however, it often misses them.

A personalized cancer vaccine works by sequencing a patient’s tumor, finding the mutations, predicting which ones will produce flags the immune system can actually recognize, and putting the best ones into a vaccine. The vaccine then trains the immune system to look for those specific flags. BioNTech and Moderna are both running clinical trials on this right now, and some of their early results have been remarkable. For example, patients with pancreatic cancer, one of the deadliest cancers, are now showing durable immune responses.

But there’s a bottleneck. A patient’s tumor might have hundreds of mutations, but only a handful will produce neoantigens that actually trigger an immune response, while the vaccine can only carry about 10 to 20 candidates. So the question is: which ones do you pick? I realised this is a computational problem and that the current tools are not solving it well.

The gap in the market

The standard tool used in clinical neoantigen prediction is called pVACtools. It was developed at Washington University and published in 2020. It’s a good tool, it’s open source, well-maintained, and widely used. But it ranks neoantigen candidates primarily by just one thing, how strongly the peptide binds to the patient’s immune presentation molecules.

But of course it’s not the whole picture, because there are many biological factors that pVACtools doesn’t account for. Is the gene actually turned on in the patient’s tumor? If not, the mutation can’t produce a neoantigen, no matter how strong the binding. Does the mutation make the peptide look sufficiently “foreign” to the immune system compared to the normal version? Is the mutation in a position where T-cells can actually see it?

Published research has shown these factors matter. The TESLA consortium (a group of 36 research teams that has nothing to do with Tesla, TESLA stand for The Tumor Neoantigen Selection Alliance) identified five key parameters governing neoantigen immunogenicity. Multiple academic groups have demonstrated that machine learning models incorporating these features outperform binding-only ranking. Yet in clinical practice, most vaccine trials still use binding affinity as the primary or only ranking signal. I saw an opportunity to build something better.

Sean Clark, Helixion Therapeutics and Elyra

Sean Clark has been my business partner on all my ventures for the past 10 years. Together we started a number of successful businesses in the past, among others Nasdaq-listed Hut 8, a Bitcoin Trust that turned into a TSX-listed ETF, First Coin Capital that was sold to Galaxy Digital as part of its public listing process, and a number of private companies and funds. We now founded Helixion Therapeutics together in Vancouver. Sean is a builder, someone who just gets things done. He is not a career academic, but someone with a deep technical ability who can take a complex problem and ship a working solution. We complement each other well, I bring the strategic vision and network, he brings the engineering execution and is one of the best sales people I ever worked with.

We built Elyra, a neoantigen prediction engine that takes tumor sequencing data and produces a ranked list of vaccine candidates. The scoring model uses five biologically meaningful features: gene expression, binding affinity, agretopicity (how much stronger the mutant peptide binds compared to the normal version), peptide length, and the position of the mutation within the peptide.

The most important finding during development was that gene expression is the single strongest predictor of neoantigen immunogenicity, even more important than binding affinity! It accounts for 36.3% of the model’s predictive power, compared to 29.5% for binding. This makes biological sense: if the gene isn’t active in the tumor, nothing downstream matters. But pVACtools doesn’t use expression for ranking. And that’s the gap we exploit.

External validation on the TESLA benchmark

In this field, everyone claims their model works. The only way to know for sure is to test on truly external data, meaning testing it on patient data that the model has never seen during training.

The TESLA benchmark is the field’s gold standard for this. It was created by 36 research groups and comprises 918 peptides across 9 patients with melanoma and non-small cell lung cancer, with experimentally confirmed immunogenicity labels. Basically, if your model works on TESLA, it works in the real world.

Elyra achieves a Global AUC of 0.812 on TESLA. That beats pVACtools (0.797) and matches the TESLA consortium’s own published composite model. Precision@10, which measures how many of the top 10 candidates are genuinely immunogenic, improved more than three times compared to binding-only ranking.

To be transparent: the margin over pVACtools is real but not enormous. We’re preparing a preprint with a full statistical analysis. But the direction is clear, and the model has room to improve.

Why now?

I believe the timing is right for three reasons.

First, the underlying science has matured. Cancer neoantigen vaccines have gone from theoretical to clinical. BioNTech, Moderna, and others are running Phase II trials. The results so far are encouraging enough that more trials are coming. Every one of those trials needs neoantigen prediction, and the current tools haven’t kept up with what we now know about immunogenicity.

Second, the open-source ecosystem has reached a tipping point. All components needed to build a serious neoantigen pipeline, are now all publicly available under permissive licenses. A few years ago, building this would have required either a large academic lab or a well-funded biotech with proprietary tools. That barrier is almost completely gone.

Third, and this is the one most people in biotech have not fully absorbed yet: AI has changed what a small team can build. AI allows you to dramatically accelerate the implementation of ideas. The bottleneck used to be that even if you knew what to build, the engineering took months or years. That’s no longer the case. The bottleneck now is knowing what to build, things like domain insight, judgments about which features matter, which datasets to trust, and which validation to run. That’s the hardest part, and that’s what Sean and I bring to the table.

The vision

Elyra’s scoring model is one component of a larger pipeline. The full vision is an end-to-end service: a research partner or biotech sends us tumor sequencing data, and we return a validated, ranked neoantigen report with structural analysis and expression profiling. Think of it as neoantigen prediction as a managed service, the computational biology department that small vaccine companies don’t have and can’t afford to build.

We’re looking for a scientific advisor

This is where I need help from my network. We are looking for one or more computational biologists or immunologists to join Helixion as a scientific advisor.

The right person knows the neoantigen prediction space from the inside. They’ve worked with tools like pVACtools, NetMHCpan, or MHCflurry. Ideally they have hands-on experience with T-cell assays, cancer vaccine clinical trial design, or immunogenomics pipelines. They may have published in this area.

But just as importantly, they’re entrepreneurial. This is a small company at an early stage. The company is officially based in Vancouver, Canada, but we work from all over the world (I am mostly based in Singapore for example). We are pre-revenue and we are building something from near-zero. If that excites you, we should talk. If you need a big team and a guaranteed salary, this probably isn’t the right fit.

Starting as a advisory role with stock options in the company at a very low strike price. If there is chemistry, we can discuss something bigger.

If you know someone who might be interested, I’d appreciate an introduction. And if you’re that person yourself, reach out to marc@helixiontx.com, DM me on X (@marcvanderchijs), or here on LinkedIn.

A personal note

I started this post talking about losing friends. I want to end there too. The reason I care about this is not because it’s a good investment thesis, although I think it is. It’s because cancer is personal for almost everyone. If Elyra can help put even slightly better neoantigens into a vaccine, that translates directly into a stronger immune response in a real patient. Maybe that patient gets a few more years. Maybe the cancer doesn’t come back.

I don’t know if Helixion will become a large company. I don’t know if Elyra will become the standard tool in this space. But I know the problem is real, the science is ready, and the tools exist to build something meaningful. I’d rather try and fail than watch from the sidelines while the technology catches up to what we already know.

If you’ve read this far and any of it resonates, get in touch.

This week I did an X Spaces (audio podcast on X) conversation with my friend @innerdevcrypto.

We live in different parts of the world so we don’t often meet face to face, but it’s always good to catch up with him and talk about life, the state of the world, AI, Robotics, and crypto. During this podcast I among others explain my new investment strategy and how I selected the assets. If you don’t want to spend a hour listening to the Spaces, this is an AI-generated summary with time stamps.

Innerdevcrypto is a (semi-)anonymous, extremely successful crypto trader, but unlike most crypto whales he is more focused on Inner Development and helping others, than on creating more financial wealth. Check out his X feed for some very interesting and sometimes unexpected content.

Marc van der Chijs (@marcvanderchijs) & Tim (@innerdevcrypto)

9 March 2026, 10 PM Singapore time

---

1. Geopolitics and the Middle East conflict

~0:55 – 6:00

Main points

• The conversation opens with discussion about the escalating conflict involving the U.S., Iran, and the broader Middle East.
• Marc believes war is always a mistake, and thinks the U.S. leadership underestimated Iran.
• There is uncertainty about the direction of the conflict:
  • Could become a long war similar to Vietnam
  • Could escalate into a larger global conflict
• Marc mentions that:
  • Friends in Dubai and Abu Dhabi are mostly safe, though tensions are felt.
  • Bombing is heard in the region but daily life continues normally in the UAE.
• Concerns discussed:
  • Potential oil infrastructure attacks
  • A 1970s-style oil crisis if oil supply is disrupted
  • Oil prices above $100 often trigger recessions
• Marc suspects Trump expected a quick operation, similar to earlier regime-change attempts.
• However, Iran is described as:
  • A large, historic civilization
  • Unlikely to collapse quickly under pressure.

Marc’s investment stance during geopolitical uncertainty

• He is not trading around the news.
• Markets are too headline-driven.
• Strategy:
  • Hold existing positions
  • Avoid reacting emotionally to geopolitical volatility

---

2. Marc’s new investment strategy

~6:00 – 11:30

Tim asks about Marc’s portfolio reshuffling over the last year.

Major shift in portfolio

Marc explains:

• His crypto exposure is much smaller than before, although he remains bullish on Bitcoin.
• His portfolio is now heavily focused on AI and robotics.

Key holdings mentioned

• Tesla (large position)
• Galaxy Digital (AI data centers + crypto exposure)

New focus: Robotics

Marc believes:

• Robots will eventually replace most human labor
• Robotics is the next major technological wave after large language models

His strategy:

• Invest in “picks and shovels” for robotics
• Example components:
  • Vision sensors
  • Robotic actuators
  • Motion systems
  • Robotics hardware suppliers

---

3. Increasing investment exposure to China

~7:45 – 10:30

Marc explains he is diversifying geographically.

Previous allocation

• ~90% of investments in the United States

New target allocation

• ~60% United States
• ~40% outside the U.S.

Focus areas outside the U.S.

Mostly Chinese companies listed in Hong Kong, particularly in:

• Robotics
• AI infrastructure
• Data centers
• Solar and energy systems

Reasons:

• China may become a major AI power alongside the U.S.
• Provides a hedge against U.S. geopolitical or economic risks

Marc believes the future AI landscape will likely be dominated by two superpowers:

• United States
• China

---

4. How Marc used AI to build his portfolio

~10:00 – 11:30

Marc describes using AI models to help construct his investment strategy.

Tools mentioned:

• Grok
• ChatGPT

How he used them:

• Discussed investment goals with the models
• Analyzed robotics companies
• Compared:
  • financial metrics
  • industry positioning
  • growth potential
  • historical performance

Outcome:

• Selected 8–10 robotics-related stocks
• Chose direct investments rather than ETFs

Reason:

• Robotics ETFs had mediocre returns
• Individual companies had stronger growth potential

Strategy:

• Hold these companies long-term
• Expect robotics to become a major economic sector

---

5. Robotics and humanoid robots

~11:30 onward

Discussion shifts toward humanoid robotics, particularly:

• Tesla Optimus
• Chinese robotics companies (like Unitree)

Marc argues:

• The robotics revolution will be bigger than many investors currently expect.
• Robotics may become the next trillion-dollar industry after AI software.

His thesis:

• LLMs transformed software and knowledge work.
• Robotics will transform physical labor.

Implication:

• Eventually most labor could be automated.

---

Core Themes of the Conversation

1. Geopolitical instability

• War risk
• Oil shocks
• Economic consequences

2. AI-driven economic transformation

• AI software revolution
• Next phase: robotics

3. Investment strategy changes

• Less crypto concentration
• More AI infrastructure and robotics

4. China vs U.S. technological race

• Dual AI superpowers
• Geographic diversification of investments

5. AI as an investment research tool

• Using LLMs to analyze markets
• AI-assisted portfolio construction

Full conversation https://x.com/Innerdevcrypto/status/2029775982509322471

Innerdevcrypto X feed: https://x.com/Innerdevcrypto

Note: This conversation and this summary are not investment advice, just my personal opinions and my views on where the world is heading and how you can prepare for it.

I’ve noticed something about the AI conversation: it’s still mostly framed as a stock-picking game. Which companies win, which ones lose, and how much upside is left in the winners. That’s the fun part, and it’s the part investors like to talk about, but I think it skips the more uncomfortable question, the one that actually determines what equities are worth in the long run.

Because if AI and robotics really deliver what people are promising, if we really move into an abundance economy where the cost of producing many goods and services collapses, then we’re not just talking about a new wave of productivity. We’re talking about deflation, and deflation has a way of rewriting rules that investors treat as laws of nature.

My working assumption is blunt: Over the next couple of years, most white-collar work starts getting destroyed at scale, and in the five years after that the same happens to blue-collar work as robotics catches up. So call it ten years until “most people” are structurally out of the labor market, at least in the way we currently define labor. We can argue about timelines, but the direction seems obvious to me: cognition gets automated first, then physical work, and then you have a society that can produce almost everything without, or with far less, human input.

The upside is of course abundance. But the downside is that abundance has a price too, and that price might show up in places investors are not modeling at all.

Abundance can be deflationary, and deflation rewrites the rules

If AI and robots take over production, the cost curve collapses. That means we should see an era of mass deflation, where a broad basket of goods and services gets cheaper each year. Not the 2–3% kind of “central bank target” story, but potentially prices going down by 20% per year in important categories, or maybe even more, for a long time. It won’t be uniform across everything, and some categories will stay sticky, but the direction is hard to escape: when marginal production costs approach zero, prices tend to follow.

And the moment you accept that, you run into the first big investing problem: “Cost goes down” does not automatically mean “profits go up.”

Why “the winners” might still get squeezed

Even the companies that survive the AI transition may face a different kind of pressure than investors are used to. Deflation does not just make consumers happy, it changes what pricing power means, and it turns competition into something closer to a knife fight.

Take Tesla, just as an easy example because it forces the question into the open. Tesla talks about selling the Optimus robots in the $20k–$30k range. Fine. But in a world where robots build robots, where factories become more automated, where supply chains get optimized by AI, and where competitors also have robots and AI, why would the long-term selling price remain $20k–$30k?

If Tesla can eventually build an Optimus for $5,000, that does not mean Tesla gets to keep the rest as profit forever. In a competitive market, prices drop. They have to drop, because everyone’s costs are dropping too. It becomes less about “can you build it” and more about “how much of the consumer surplus can you keep before a competitor takes it.

Or take Amazon: it has scale, logistics, distribution, cloud, and an ecosystem, but a lot of what makes Amazon powerful is operational excellence and coordination, and AI makes coordination cheaper for everyone. If AI turns a bunch of smaller players into “mini Amazons” in specific niches or if AI agents threaten to disrupt Amazon’s model, pricing pressure rises, and even the giant gets pulled into a more competitive equilibrium.

So yes, costs collapse, but prices collapse too. And that is not the story most DCF models are telling.

My “90% die” thesis, stated more defensibly

I used “90% of companies could go bankrupt” as a provocative way of saying: I think the majority of public companies are structurally unprepared for the regime we’re moving into. Here’s the more precise version of that claim.

A large chunk of listed companies exist because they sit in one of these positions:

1. They are a bundle of humans performing repeatable cognitive work (analysis, support, coordination, reporting, marketing, sales ops, basic legal work, basic finance work).
2. They are a thin layer of distribution that used to be expensive (middlemen, aggregators, brokers, certain marketplaces).
3. They are a branded wrapper on commodity production and their differentiation is mostly marketing and shelf space.
4. They are a regulated monopoly or oligopoly whose margins depend on the fact that entry has been hard, not because the product is inherently hard.

AI attacks (1) directly. It attacks (2) by lowering transaction and discovery costs. It attacks (3) by accelerating competition and by making “good enough” alternatives easier to create. It even attacks (4) over time, because regulators eventually have to respond to public pressure when the gap between “what something costs to produce” and “what you pay” becomes absurdly visible.

That’s why I think a lot of companies won’t survive, not because they all literally file for bankruptcy, but because their economic role gets competed away. They become irrelevant, acquired for scraps, or slowly melted down.

And that leads to the next idea: if most companies are disrupted, then a small set of companies becomes dominant. And that creates a different problem.

Universal basic income changes the tax base, and that hits equity returns

If most people don’t have jobs, the income tax base collapses. In many countries, income taxes are a major source of revenue. If labor income disappears, the state has to fund society some other way, especially if it’s paying universal basic income at scale. That money has to come from somewhere.

It’s obviously not going to come from the unemployed, so it will likely come from capital. And more specifically, it will come from the small set of companies that survive and dominate production, because that’s where the cash flow is.

This is where the stock market story starts to feel naive. If my “many companies die” assumption is even partially correct, then the remaining giants effectively become the tax base. And once you realize that, you get forced into a scenario that almost nobody puts in their spreadsheets: corporate profit being taxed at levels we haven’t seen in modern times.

Pick a number. 60%? 80%? I’m not claiming we’ll literally see 90% corporate taxes, but I am saying the direction of political pressure is obvious. If society wants stability and most people don’t earn wages, “tax the winners” becomes the most natural move in the world, especially when those winners are visible, concentrated, and globally powerful.

We’re already seeing the Overton window shift on wealth taxes, capital gains taxes, and new forms of “solidarity” taxation across jurisdictions. It’s not hard to imagine a future where the largest AI and robotics firms are treated like utilities, expected to fund the social contract.

That’s the key point investors miss: the more the winners win, the more they become the funding mechanism for everyone else.

So what does this mean for stock prices?

Stock prices are supposed to reflect discounted future cash flows. In an AI-driven abundance economy, you can get a strange combination:

• Revenue pools shrink in real terms because prices fall.
• Margins compress because competition accelerates.
• Taxes rise because governments need a new base to fund society.

If those three forces show up together, then the standard “AI will boost productivity therefore stocks go up forever” narrative is incomplete at best. You can still have a handful of dominant firms, and you can still have massive output growth, and yet equity returns could be lower than people expect because the surplus is competed away or taxed away.

This is why I suspect the market is complacent. Most investors are still debating “which companies benefit from AI.” Far fewer are asking: “What happens when AI pushes prices down so hard that profit pools shrink, and governments treat the winners as the new tax base?”

Crypto enters the picture, awkwardly but logically

I’m not trying to turn this into a crypto sermon, but the logic is hard to ignore. In a world where governments need revenue and will go after visible pools of capital, assets that are easy to tax become politically convenient targets: public equities, property, ETFs, custodial accounts.

Bitcoin and other crypto assets are different, not because they are untouchable, but because they are more portable and can be held in ways that are harder to control. Of course governments can tax ETFs, on-ramps, corporate treasuries, and realized gains. Of course they can make life difficult.

But the enforcement friction is different. And in a world where the state needs to extract more from capital, the structure of an asset starts to matter more than most investors want to admit.

It’s also not crazy to imagine new forms of taxation aimed at foreigners, withholding taxes, or other measures that treat global capital as a funding source for domestic stability. The world can change faster than our mental models.

The uncomfortable truth: we don’t have a model for this

What unsettles me is that I don’t think anyone has solved this puzzle. We don’t have a good historical template for “automation eliminates most labor income” combined with “abundance-driven deflation” combined with “a handful of mega-corporations dominate production” combined with “governments must fund a population at scale. We can hand-wave it as “we’ll figure it out,” but investors need to price it.

And if this regime is even directionally correct, then today’s stock market could be mispriced in both directions. Markets may underestimate how high the winners can go during the transition, because the next few years of AI-driven leverage could be insanely profitable. But markets may also underestimate how hard the end-state can be on margins, taxes, and therefore long-term valuations.

So yes, we could go much higher first, because markets rarely price second-order effects until they have to. But something has to give eventually.

The questions I’m still working on

How deflationary will it really be? Some things may remain scarce: land, possbily energy, certain raw materials, regulation-protected services. If those dominate the cost of living, deflation is weaker and UBI requirements rise.

What level of UBI is actually needed in an abundance economy? If basic goods get dramatically cheaper, maybe $500 a month buys a decent life. If housing and healthcare remain captured, it doesn’t.

How high do taxes go, and on what exactly? Corporate profits, wealth, land, consumption, financial transactions, data, robots, compute? The choice changes everything.

Do capital markets accept a new regime, or does the risk premium change? If equity becomes an increasingly political claim on profits, investors will demand a different return, or they’ll look for different assets.

Do we get fragmentation or coordination across countries? If one place taxes aggressively, capital moves, unless the winners are immovable or governments coordinate.

Conclusion

I still believe the next decade will be dominated by AI and robotics, and I still believe we’re moving toward abundance. But abundance is not automatically an investor utopia. Abundance can be deflationary, deflation can compress profits, compressed profits change valuations, and a society with mass unemployment changes the tax regime in ways that most investors are not thinking about at all.

The weird part is that I can hold two opposite views at once: For my investments I can be bullish on the transition and uneasy about the destination. And for society I am very worried about the transition (riots, civil wars), while I am bullish on the outcome (abundance).

I don’t like instability, and I don’t like that the future feels much less predictable than it used to. But it is what it is. The only rational response is to think harder, and accept that “normal” may be a temporary phase we’re already leaving behind.

There’s a lot of talk right now about programmers losing their jobs because AI can code. But that’s the wrong framing, this is much bigger than that, coding is just the most visible part. If I look back at my last corporate job, at Mercedes-Benz regional headquarters in Beijing (2000–2002), I realize almost all of it can now be done by AI. And this was not an intern role or so, I was Senior Manager, Controlling, for Daimler’s operations in Northeast Asia (Daimler was the mother company of Mercedes-Benz). In hindsight, that title was partly job inflation, it was more of a solid middle management job. But I was good at it, I had real responsibility, and Mercedes-Benz paid a lot of money for me to do it.

What I actually did all day

The job was simple, in the way many corporate jobs are simple once you strip away the endless meetings and politics.

I communicated with joint ventures, factories, and sales organizations in the region. They sent me spreadsheets with their financial results. I analyzed them, consolidated them, and turned them into reports and presentations. I made budgets: monthly budgets for the next year, and  annual budgets for the next three to five years. Then every month I compared budget vs actuals, explained variances, wrote it up, and presented it.

There were other tasks, of course. There are always other tasks, especially for a young, ambitous guy who was still eager to quickly climb the corporate ladder. But most of the job was variations of collect numbers, make them consistent, explain them, and tell a story.

The expat package was wild

When I moved to Beijing I had just turned 28. I don’t remember my exact salary anymore, but including expat ‘hardship’ allowances it must have been somewhere around USD 100k to USD 150k per year at the time. This was a real salary in those days, and in Beijing it went very far. For some reason headquarters in Stuttgart thought life in Beijing was hardship, but for me it was actually much easier and relaxed than life in Germany. You could live like a king, and I did!

On top of that I got a housing allowance, which was USD 6000 per month (that number was so astronomical to me at the time that I still remember it). I had a very nice company car, and I was flying business class. Even the vacation flights to Europe for me and my girlfriend were business class and paid for by the company.

Mercedes spent a lot of money on this role. And the crazy thing is: the role was basically a workflow, a workflow that is now completely automatable.

AI can do my entire workflow

AI can:

• write and format reports
• build and improve presentations
• analyze numbers, spot anomalies, find drivers
• reconcile messy spreadsheets and inconsistent formats
• translate and communicate in multiple languages
• ask follow-up questions when inputs are missing
• chase people politely when they are late
• run the same process every month without getting bored

I didn’t speak Chinese well back then. I spoke English and German fluently (and Dutch, which was useless for this job). AI can communicate directly with factories in Chinese, Korean, Japanese, English, German, whatever is needed, and it can keep the tone consistent.

Even the spreadsheets I received can now largely be produced by AI agents inside those factories and sales orgs. A lot of financial reporting is simply formatting, mapping, validating, and explaining. In other words, both the inputs and the consolidation layer are automatable.

This is not about AI getting smarter anymore

For a long time we could say: “Yes, AI helps, but you still need people.” But that argument is getting weaker by the week. Over the past weeks AI has improved so much that you don’t need the people anymore. AI agents can now schedule tasks (for example, make a weekly report every Monday morning, create a powerpoint presentation for headquarters every Friday afternoon), and make decisions about what to  independently from their human.

From now on it’s not about whether AI can do the job, but it’s about whether companies are willing to let it do the job. Are they willing to adopt AI? It requires organizational courage, a lot of risk tolerance and of course the eternal internal politics. The fact is that cutting staff is painful and messy, and nobody wants to be the first manager to say: “I can replace my entire team with software.” But they can, and eventually they will.

One person could do twenty of these jobs

My old job is a perfect example of leverage. If you give a capable person the right AI tooling, they can run the reporting and budgeting process for far more entities than one human ever could in 2001. If a company like Mercedes-Benz wanted to, they could centralize a lot of controlling work. One small team at headquarters could cover what used to be dozens of regional controllers. They won’t do it overnight, but they will do it over time. Quietly, with “reorganizations” and “efficiency initiatives”.

If you are in that job today, you should pay attention

Somebody is doing a job similar to mine right now. Somewhere in Stuttgart, Singapore, Beijing, Dubai, wherever Mercedes-Benz has regional structures. That person is at serious risk. They might be excellent at their job, but their job is a process, and processes are what software eats.

The emotional part, for me

I was proud of that role at the time. I didn’t love corporate life, but I was good at it. It felt like a serious job. A “real” career path, a big salary and huge condo, all while in my late 20s. I felt like I had made it. (As a side note: I still eventually decided to leave that cushy job, to study Chinese and become an entrepreneur, something none of my colleagues at the time understood).

Realizing that the whole thing can now be done by AI is a revelation. It makes the changes feel very concrete. It’s not a thought experiment anymore, or a theoretical idea. It’s my own life, my own résumé, one of my own chapters, and I can see how it gets automated.

What happens next?

If you are a 20-something today in a white-collar job, you probably feel this already. You see what the tools can do, because you use them, and you know the direction. If you are in your 40s or 50s, you might still be underestimating it. Not because you’re not smart, but because your mental model of “what a job is” was formed in a world where the work required humans by default.

The scary part is the speed, and the fact that many people will not have an obvious “next job” to transition into. If you lose your job and you know you can find another one in a few months, you can survive. If you suspect your whole category of work is being automated across the economy, that’s a different psychological reality.

If I’d be in that job right now, I would spend all my waking hours learning about AI and vibecoding and I would automate my own job. I would show top management in both the regional headquarters and in Stuttgart how they could save tens or hundreds of millions of dollars right away. My colleagues would likely hate me for it, but they would simply be shooting the messenger, because if you don’t adapt you die. Maybe I would be fired for doing it (Mercedes-Benz was a very conservative organization in those days, they resisted change, especially if initiatives came from middle management, which was one reason why I left the corporatate world), but my job would have been gone eventually anyway.

We will need a new social contract

I don’t see a way around some form of universal basic income, or what Elon calls “universal high income”. I believe it’s a stability requirement, without it the world as we know it will fall apart. If people lose their jobs and don’t have a way to provide for their families, plus no expectation of finding another job anymore, they will revolt.

We are heading toward abundance in production and scarcity in employment. That mismatch creates political pressure, resentment, and instability. We can either design for it or crash into it. I’m not optimistic that governments will handle this gracefully. But I also don’t think we can stop the technology anymore. And honestly, I don’t want to stop it, because the upside is enormous. The question is whether we build a world where people can still live with dignity while the machine does more and more of the work.

Final thought

If my Mercedes-Benz “senior manager” job from 2000 can be done by AI now, then a huge portion of white-collar work can be done by AI. Many people believe it may happen in 5-10 years, but reality is that we are there already.  

AI doesn’t need to get more capable anymore, in its current form it’s good enough already. What we are seeing now is the slow realization inside companies that the old org chart is outdated. Once that realization spreads, things will move faster than most people expect.