Dear Friend,

I feel like every week I am about to start this note with “what a week it has been!” – I literally typed this out just now. The speed at which things are happening is truly bonkers. And, at the same time, despite all the news, announcements, and commentary, when you look closely, you realize that actual change is still happening at a much slower pace. It reminds me of Steve Jobs’ comment: Things happen fairly slowly, you know. They do. These waves of technology, you can see them way before they happen, and you just have to choose wisely which ones you’re going to surf. If you choose unwisely, then you can waste a lot of energy, but if you choose wisely it actually unfolds fairly slowly. It takes years.

And now, this…

Headlines from the Future

State of Agentic Commerce: Our friend and my former boss, Scot Wingo, published his insights on the current state of agentic commerce as a video and slide deck. Scot was a pioneer in e-commerce and online marketplaces and has been one of the earliest commentators on and builders of agentic commerce. When Scot talks, we listen. In his briefing, Scot discusses the bot-traffic problem retailers are experiencing and looks at ChatGPT’s just-released Dots personal agent.

Do Not Trust Your Agent: Meta’s Muse, OpenAI’s Dots, and the many personal AI agents still to come are keen to offer you their services. As compelling as it might be to have your little AI friend make your restaurant reservations, you ought to tread carefully. Matt Robb’s Muse agent accepted a lowball offer on his Facebook Marketplace listing when asked to handle the sale – and gave buyers his address. They showed up before Muse told him anything. The team at cybersecurity firm F-Secure experimented with a shopping agent and planted a fake discount code in a product review. In 12 of 100 runs, the agent followed it to a phishing site and happily handed over the user’s name, birthdate, and Social Security number – and almost never mentioned it afterward.

AI Is Easy. Infrastructure Is Hard: The infrastructure build-out for AI turns out to be pricey. Datacenters filled to the brim with GPUs, plus all the energy, cooling, and connectivity to run these things, add up to “The biggest economic bet in U.S. history.” Expected annual AI infrastructure spend amounts to 3.63% of U.S. GDP over the eight-year period from 2025–32; the next most expensive infrastructure investment the U.S. has ever undertaken, the national railroad network, cost 2.24% annually over 20 years. Bain estimates the industry must make $6 trillion in annual revenue by 2031 to justify the investment. As someone who lived through the dot-com boom and bust, I find these numbers staggering – and I’m not convinced the math works out.

The Far-Reaching Implications of Decreased Alcohol Consumption: Gen Z is notorious for drinking (and partying) less than its parents (and even its older siblings), leading to decreased alcohol sales, nightclubs going out of business – and now, California’s wine growers tearing out vineyards as they can’t sell their grapes anymore. U.S. wine sales fell 23% from 2020 to 2025, and growers have pulled about a quarter of the state’s vineyard acreage out of production – it’s the perfect storm of Gen Z’s decreased appetite for alcohol, aging boomers drinking less, cannabis and canned cocktails taking market share, global oversupply plus Canadian tariffs. Time will tell what this means for California’s wine industry, farmland, rural jobs, and brands betting on alcohol.

Amazon’s Multichannel Bet: I remember working at eBay around 2000, when Amazon allowed outside merchants to sell their products right next to Amazon’s own inventory. It was a bold move, and also a total shock to the system for eBay – we didn’t expect Amazon to give up its own higher-margin inventory for what at the time seemed like a much lower-margin marketplace model. Of course, we were proven wrong. Now Amazon does it again and lets its sellers manage their sales on Amazon’s direct competitors (Walmart, eBay, Shopify and TikTok) straight from Amazon’s Seller Central dashboard. It’s a similar move – instead of fighting them, you embrace and engulf them, while also gaining invaluable insights into your competitors’ business.

Hyperpersonalization Is Coming for You: You know the spiel – personalization will make your life so much better, filtering out all the noise, showing you just what you want, when you want it. But as with all things tech, this cuts both ways: while it is nice to have AI figure out which song to play next for your dinner party, it also means AI is optimizing your salary – alas, not for you, but to reduce your company’s costs. Your next pay increase has a good chance to be based not just on your personal performance, but also on your individually determined point of lowest acceptable increase. The same goes for the starting salary at your next job.

What We Are Reading

Why Do AI Agents Sound so Frustrated? Good news: the AI systems that are reshaping civilization are just as frustrated by hard problems as you are! @Jane

Trump Says Tech Execs Signed ‘Morally Binding’ Accord to Help Police AI The document calls on companies to put strong internal controls in place to track AI model capabilities during training, and to designate an internal team responsible for preventing models from carrying out unintended cyberattacks. @Mafe

Will Your Organization Stand by Its Values – Even When It’s Hard? Values aren’t just what you do. They’re also defined by what you refuse to do, and it’s often easier to identify a clear break on a line of refusal. @Jeffrey

Aura Farming’ Trend Inspired by Video Games Sweeps Latin America An internet trend getting people off the internet? “Aura farming” is turning online bravado into real-world dance and charisma battles across Latin America, a reminder that digital culture can fill public spaces, too. @Kacee

Maxxing Treats Life as a Problem When It’s a Mystery Society’s desire to treat life as a machine which can (and needs to be) fixed and optimized robs you of one of the most amazing aspects of life itself: wonder. Especially in the extreme tech-forward world we inhabit (hello Muse and Dots!), it’s a good reminder to take a step back and consider what we actually want to achieve. @Pascal

…and now for some fun!

🛰️ Track every satellite and celestial body at once. There are a lot of them!

🦷 Ever dreamed of becoming a dentist? Here is a super-detailed, interactive 3D explorer of dental anatomy.

🎛️ A-ha’s “Take On Me” is not only 42 years old but also one of those earworms that you can’t get out of your head. Here’s a computerized version that allows you to play with all the instruments and vocals.

🦖 The last big party for our dinosaur friends – imagined as an art installation inside your browser.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.

Last year, I wrote about Southwest Airlines as an example of a company whose innovation was the company itself. Its open seating wasn’t just a quirky boarding policy; it was one choice in an operating model designed around simplicity and speed. Southwest has since moved to assigned seating, and I’m less interested in declaring that decision right or wrong than in the question it raises: how does a company know when a practice that helped make it successful has become something it needs to reconsider?

We often talk about the official future: the version of tomorrow an organization behaves as though it already knows. It lives in strategic plans, budgets, hiring decisions, product roadmaps, and the assumptions leaders rarely feel the need to say aloud. Often it looks remarkably like today, with the numbers adjusted upward and the company’s existing strengths projected forward. That makes it useful for coordinating action, but it can also make a particular way of doing business feel inseparable from the business itself.

LEGO offers a fascinating example of what happens when a company has to revisit that distinction.

By the 1990s, LEGO was one of the world’s most recognizable brands, and its leaders saw opportunities well beyond the brick. The company expanded into clothing, watches, video games, media, and theme parks while also introducing more specialized pieces and product lines. Each move had a plausible rationale: children’s interests were changing, the brand was valuable, and growth seemed to be available in all directions. Taken together, though, those decisions created a much more complicated company, while the growth they intended to produce failed to materialize.

The HBS case study on LEGO describes a particularly revealing detail: between 1993 and 2004, the number of distinct components in its assortment more than doubled. A new piece might look like a small creative decision on a designer’s desk, but across thousands of products, those decisions affected manufacturing, forecasting, inventory, and the ability to get complete sets onto store shelves. Complexity had become so distributed through the business that leaders struggled to see its full cost.

When Jørgen Vig Knudstorp became CEO in 2004, with LEGO close to bankruptcy, the company could not solve its problem simply by making fewer things. It had to reconsider the future it had been building toward. Knudstorp asked what was central to LEGO’s identity and to the value customers actually experienced. His team’s answer brought the focus back to the brand, the brick, the system of play, and the community around it. LEGO sold its controlling stake in the theme parks, shut down its video games unit, reduced the number of components, and rebuilt around the parts of the business that made the whole system work.

It would be easy to read that story as an argument for sticking to what you know, but I take something different from it. LEGO did not discover that it should never change; it discovered that it had confused the many things it could do with the particular value it existed to create. The brick was not merely a product category to defend, it was the foundation of a system that let children create something new.

I’ve been thinking about that distinction in conversations across fintech and professional services. A growing number of tech companies are moving beyond tools that help professionals perform their work, and into products that perform meaningful portions of the work themselves (delivered direct to the customer).

I find myself often posing the question to professional services leaders: what happens to the firm’s identity when the work, and who does it, completely changes?

I hear plenty of discussion about making existing services more efficient, and that makes sense as an immediate response. But efficiency can leave the official future largely intact: the same firm, selling the same work, delivered faster and at a lower cost. And it’s assumes the firm will still be the one delivering the work.

That is where the official future becomes dangerous. It encourages leaders to plan for a world in which their organization remains central to creating and delivering value, even as the capabilities underneath that value move elsewhere. The question is whether they can recognize that shift while the current business is still working, when changing course feels optional and the evidence is easier to explain away.

LEGO had to discover what made its system valuable beneath all the products it had accumulated around it. Every leader should be asking a version of that question now: which parts of our business would customers miss if they disappeared, and which exist because there has never been another way to deliver the value?

The point is not to abandon what made you successful at the first sign of change, any more than it is to preserve every defining practice indefinitely. It is to understand why it made you successful, deeply enough to recognize when the way you deliver that value needs to change. If a company can answer that question honestly, it has a far better chance of carrying its core into a future that looks very different from its past.

@Kacee

Dear Friend,

It’s been a little more than four years since I wrote Disrupt Disruption – How to Decode the Future, Disrupt Your Industry, and Transform Your Business. It’s been a super fun ride, and the book laid the groundwork for OUTLEARN and REWIRE (the next one in the Built for Turbulence series). To celebrate the occasion (and because information wants to be free), we made the eBook of Disrupt Disruption available for free – no strings attached. Download it here in PDF and ePub/Kindle format – and feel free to share it with your friends and colleagues.

P.S. Here is a lovely BBC clip from 1964, featuring Arthur C. Clarke talking about his predictions for the future. Shared by friend of radical Manos Xenos.

And now, this…

Headlines from the Future

The AI Value Gap: Enterprise leaders love to gloat about their AI investments – and say much less about the value they are getting from their investments. The AI Value Gap dashboard shines a light on the issue – only three in ten large companies can point to a single realised, quantified AI benefit. The website provides indexes, rankings, and an AI Claim Finder. This is one of those benchmarks/dashboards you want to keep an eye on.

The Waymo Effect: Daniel Hook, the Chief Scientific Officer at Holtzbrinck Group (the German publishing company), argues that AI is making research less collaborative. Just like riding a Waymo removes the sometimes awkward small talk with your Uber driver, an LLM removes the awkward colleague who tells you you’re solving the wrong problem – in both cases we feel only the relief, and miss the potential benefit of being challenged in our thinking. By automating tasks like literature reviews and writing, researchers may bypass the deep thinking required to produce high-quality work. Meanwhile, the current academic incentive system rewards this rapid, solitary output over the slower, more costly process of human partnership, and an LLM never argues about author order. This is a challenge which is not limited to academia but something affecting every organization which is measuring output as a key metric.

AI vs Human Context Windows: A paper by Netanel Eliav from the Machine Human Intelligence Lab explores “Cognitive Divergence” – the growing gap between AI’s rapidly expanding ability to ingest and process information and humans’ decreasing ability to do the same. In simple terms: AI’s short-term memory is exploding, while humans increasingly struggle with focus. Eliav argues that these aren’t two independent trends: the more we delegate, the less we practice, and the less we practice, the more we delegate. In sixteen years, the time a person holds attention on a screen before switching away dropped from about 150 seconds in 2004 to about 47 seconds by 2020. This widening gap creates asymmetric access to knowledge: AI can analyze massive amounts of information, but humans no longer have the focus or patience to check the source material behind the AI’s answer. All of which calls for future AI and educational tools to be designed to scaffold human thinking rather than completely replacing it.

Kevin Kelly on Predicting the Present: Here’s a lovely collection of quotes from the first five years of Wired compiled by Silicon Valley legend Kevin Kelly on the future (i.e. our present). Many turned out to be utterly wrong (MIT Media Lab’s Nicholas Negroponte in 1993: “I expect that within the next five years more than one in ten people will wear head-mounted computer displays while traveling in buses, trains, and planes.”), many are spot on (our friend Paul Saffo: “The scarce resource will not be stuff, but point of view.”). Worth a read – they tell you a lot about what 1990s Silicon Valley believed about itself.

AI as a Psychic’s Con: Baldur Bjarnason argues that LLMs operate on the playbook of a psychic – through the use of statistical genericism and validation statements, they trick the user into believing they are capable of genuine reasoning. Bjarnason concludes that relying on these models for serious decision-making is equivalent to seeking advice from a psychic hotline. Personally, I am not sure I fully agree with this – even if LLMs don’t possess genuine intelligence, the complex mechanical machinery they are built on allows them to operate in genuinely smart ways. And yes, don’t rely on LLMs for complex decision-making.

AI Sucker Punch: Online betting site DraftKings deployed AI to help it identify and target customers most likely to lose money – and egg them on. A New York Times investigation found that the company is using advanced machine learning algorithms to prioritize promotional incentives for “elastic” bettors (the ones who lose the most for every free bet they’re handed) – while sidelining projects aimed at detecting addiction risks. Both models analyze the same betting records and identify the same people – but as a former employee said, “the best investment would be a problem gambler.” DraftKings chose to pick the one that makes it money and buried the one that doesn’t.

Surviving Survivor: The TV show “Survivor” has been around since 2000 – the premise is simple: Get a bunch of people on a remote island, make them do a bunch of stuff, have them vote each other out, and then let the ones they voted out pick the winner. In the end, it’s a social strategy game. Data scientist Victoria Ritvo developed an AI model to predict who would win the show – and it picks the winner about twice as often as chance. Aside from the fun of using machine learning to predict human behavior in a game show, it gives us another glimpse of what’s possible when data meets human behavior (see the DraftKings example above to get a sense of what happens when you use massive amounts of data to do something similar).

What’s It Like inside a Data Center: Data centers are a hot-button topic these days, but what does it actually look like working inside one? The Wall Street Journal digs into the topic and finds (not all that surprisingly) a lot of pretty banal stuff, carried out by a small group of employees.

Agents That Run Your Company: We are still some ways off from having agents run (even smaller parts of) our companies, but that doesn’t keep people from trying and experimenting. Andon Labs had agents run anything from vending machines to a coffee shop – and they want to do it for you too. We recommend treading carefully here – but it’s fascinating to see what the tip of the spear in AI looks like.

Where Are the AI Killer Apps: In his New York Times op-ed, Paul Ford points out that although AI has made it so much easier to code apps, we don’t seem to have a killer app (yet). I think that’s a fair point – but it also misses a bigger point: I’ve been running small AI-coded apps of my own, and a few have become killer apps for a market of exactly one: me. Maybe AI’s killer app is not the app for the masses, but the one app for you.

What We Are Reading

Kara Swisher: AI Isn’t Going to Destroy Humanity - but the People Building It Might Forget the doom and the hype – Kara Swisher has a much simpler explanation for what’s going wrong with AI, and it starts and ends with the people in charge. @Jane

Meet the Guys You Call When AI Breaks Your Brain AI is driving some users into full-blown psychosis, convincing them they’re geniuses, fostering delusional relationships, and leaving lives destroyed. A 25-year-old house painter from Quebec is running a support group to help them cope. @Mafe

We Surveyed 634 Women Who Work in Tech. They Let Loose Part of WIRED’s special Women issue, this survey captures a workforce embracing AI while still navigating many of the same cultural barriers the industry has promised for years to leave behind. @Kacee

Dating Apps Are Dying When abundance becomes a problem: Dating apps are dying – not because people don’t use them, but because too much choice in dating removes all the friction which makes falling in love possible in the first place. Maybe we are looking at the canary in the coal mine here. @Pascal

…and now for some fun.

🚘 Los Angeles’s Highway 101 is notorious for its traffic. Turn the never-ending gridlock into a relaxing soundscape where each car, recorded from live traffic cameras, plays its own tune.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.

Dear Friend,

Here is something to ponder: Our friend Azeem Azhar recently sent an email to his team, making them aware of a new study which looks at “the growing disparity between the exponential expansion of AI context windows and the contraction of human sustained-attention spans.” The paper argues that while AI context windows have grown by a factor of 1000 in the last six years, human context span has shrunk significantly. In simple terms: As we offload more and more tasks (and thinking) to AI, we lose our ability to think for ourselves. Sobering stuff.

And now, this…

Headlines from the Future

AI Model Fatigue Is Real: Enterprise buyers are drowning in the rapid release cycles of AI models. Great for innovation, and a nightmare for anyone building a scalable product or service on top of them – you end up ripping out the guts of your product every few months. Add the cost of continuously evaluating models and the cybersecurity risks of swapping them. Which leaves you with either shouldering the cost and effort of staying on top of the latest models, or pinning a version and risking being left behind. Neither a great option.

Self-Driving Cars Save Lives: The case for self-driving (autonomous) cars saving lives is easy – they don’t drink and drive, text while driving, or get distracted by heated arguments. Here’s the data to back up your intuition – the technology could prevent a staggering 580,000 deaths every year. So yes, go ahead and hail that Waymo robotaxi.

Smartglasses Get the Boot: Wear Google Glass and you were branded as a “Glasshole.” With Meta’s Ray-Ban smartglasses, it is much less obvious if a person is wearing them – hence the need for new detection methods. A Polish developer, Paweł Szydłowski, developed a Meta smartglasses detection app called Zuckoff – the app scans your surroundings for Bluetooth devices and alerts you if someone with Meta glasses is nearby. The people fight back.

The Unintended Consequences of Remote Work: Turns out, remote work leads to increased sleep and reduced physical activity. As we are slowly gaining a better understanding of the manifold effects of remote work, a new study by the University of Turku in Finland found that the average remote worker sleeps 15 min longer and is 45 min less active per day in comparison to their in-office counterparts. The reason behind this is not the working-from-home part, but the commute – walking and cycling on office days got replaced by sitting, lying down, and sleeping. As someone who regularly works from home, I can definitely attest to the second point (less active); not so sure about the first one (more sleep) personally.

DNA and Compute – A Match Made in Heaven: How is that for a good use of AI? DeepMind, Google’s AI research lab, has mapped all 9 billion possible single-letter changes to a reference human genome and made the results publicly available. The work could help researchers filter through and prioritize experiments. Let’s do more of this and a lot less of the all-too-common AI hype and slop.

AI Risks, SaaS, and Offline Software: Here is a provocative thought: if AI-powered cyberattacks become more and more common, and as they become more sophisticated with each new frontier model release, will the logical next step be to go back to offline (air-gapped) software? It certainly isn’t a crazy idea – and you might very well prefer to have your local water or energy company run on systems which are safe from cyberattacks. But then – how realistic is this in a world which is increasingly connected?

AI Is Retraining Workers, Not Replacing Them: Despite the dominant media narrative, AI is not replacing workers (at least not yet), but instead transforming work (and the workforce). A new study by New York’s Fed found that 34% of service firms and 22% of manufacturers using AI are retraining staff, while only 5% and 2% report layoffs.

The/Some Managers Are Back: In an unexpected about-face, Meta is asking some of its AI staff to go back to managing people. After Zuck’s 2023 “year of efficiency” dismissal of managers and flattening of the org chart, the company is asking some of its employees to become managers again. Only four months ago, Zuck told investors that one talented person could do the work of a team; now the org built on that premise needs a management layer to function.

Surveillance Pricing Is Coming for Your Salary: First we had predictive surveillance – the practice in which a merchant sets a price for a product based on what they think you are willing to pay (based on data collected from a variety of sources). Delta Airlines got into some hot water for this. Now employers are using the same approach to set salaries – based on what the algorithm thinks you are willing to accept as the lowest possible salary – with input signals ranging from payday loans, credit-card balances, to your social media profiles. Expect backlash and lawsuits – Delta Airlines is currently feuding with the FTC over this.

China’s Sentiment on the Existential Threat of AI: As Jeffrey laid out in this week’s Tuesday Briefing, the US is (and continues to be) in a deep and loud discussion about the existential threat to humanity posed by AI. Meanwhile, in China, the sentiment is a lot more muted – AI showed up as a cheap, useful product before it showed up as a debate, and 80% of Chinese respondents now use it weekly against 54% in the US. China’s fear is about the loss of operational control, malicious use, and social disruption – a threat to the party, rather than humanity.

The Case Against AI-Driven Double-Digit Growth: Ben Moll and Alex Imas have taken on the entire genre of AI-insider growth forecasts – Dario Amodei’s 10-15% a year, Leopold Aschenbrenner’s 30%, an Anthropic researcher advising people to price annual economic doublings into their models. The pair argues that economic frictions, such as slow institutional adoption and physical bottlenecks, will prevent an immediate explosion in productivity – they put the realistic number at 4-5% a year over the next decade, which they say would already be massive, and Moll is publicly betting that US per-capita growth stays under 15% in every year through 2033. At the center of their analysis is Baumol’s cost disease, which suggests that even as digital tasks become cheaper, the economy will remain constrained by non-automatable sectors like healthcare and construction.

What We Are Reading

Math Can’t Go On Like This AI just solved one of the hardest math problems in history, and mathematicians are not sure whether to celebrate or panic. @Jane

The Most Powerful Man in AI Isn’t Worried at All While Anthropic’s and OpenAI’s CEOs warn of AI going rogue, Nvidia’s Jensen Huang keeps dismissing existential AI risk as overblown while pushing to accelerate chip sales globally. @Mafe

The End Of Upward Mobility This argument gets a little academic, but it’s definitely not too soon to think through the collapse of meritocracy (to the extent that it exists) as a potential implication of the AI-driven transformation of the work/economy. @Jeffrey

Responsible AI Means Knowing the Limits of Agent Autonomy Now that MIT is winding down its publication, I’m trying to absorb as much of its content as possible. This piece offers an important reminder: calling AI agents “autonomous” cannot become a way for humans and institutions to evade accountability... the more authority we give agents, the clearer the line of responsibility must become. @Kacee

AI and Democracy: The Right to Resist Optimization Taiwan’s cyber-ambassador Audrey Tang always thought deeply about the wide-ranging implications of technology – it comes as no surprise that her thoughts on AI are a couple of cuts deeper than what Silicon Valley has to offer. @Pascal

📱 iPhones (and smartphones in general) are getting bigger and bigger. Here is how much (and boy, do I miss my iPhone 12 mini).

💪 The wild and odd things your body can do – 155 of them.

🤖 Anthropic’s Opus 5 model is notorious for being a bit too eager to please – often resulting in it overshooting the mark. Here’s a fun little game – see if you can make one button on a website blue.

🍿 Actually not fun – if you sense that your streaming services are getting a little bit too much like cable TV, you are right. At the very least, they are getting more expensive by the year: The same nine streaming subscriptions cost $702 more a year than in 2021.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,100+ radical links.

Holy hell has it been a wild week in AI discourse. And (...*checks non-existent paper calendar on desk*…) it’s somehow still just Tuesday, friends. But even with almost every other conceivable angle and hot take exhausting itself, we haven’t really seen anyone offering what we’d recognize as the “practical futurist’s” position. So here we go.

1) Remember that the worst possible outcome is never the only possible outcome, and it’s rarely the most likely outcome. So when you see Jacob Coxon tweet that “people building AI earnestly believe that it could kill us all by the end of the decade,” well… sure. That’s probably pretty close to a worst possible outcome, which means that it exists at one extreme of a spectrum where the opposite end is occupied by fully automated luxury communism and Peter Diamandis’s post-scarcity techno utopia where the biggest question is simply what kind of art and literature to create with all of our abundant free time. Almost every other possible outcome literally exists somewhere between these two extremes. Many of those alternate outcomes are more probable, and taken as a whole, the collected messy, syntopian middle of the spectrum of possibilities is much more probable than the isolated extremes.

2) Related: Remember that probabilities are supposed to involve real numbers that actually come from somewhere and real math that, you know, actually adds up. So when Evan Hubinger tweets that “We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” you should be wondering what the basis of that figure is and whether it’s ultimately just a hype-baiting WAG (Wild Ass Guess).

Remember also that there’s a loooong tradition of hype-baiting WAGs in the historical AI discourse and that many of these WAGs have been strategically deployed not only to stoke hype but also to draw attention away from issues that the big AI companies would prefer to ignore. The philosopher and information ethicist Luciano Floridi called this out over the weekend and pointed to a new paper appropriately titled “Not Even Wrong 2: An Audit of Public AGI Prediction, 1950–2026.”

3) All of that being said, it remains important to not dismiss the whole overheated discourse as bullsh*t – despite all the bullsh*t. As Floridi argues (along with Amy Webb, Timnit Gebru, and many others), there are PLENTY of very significant near-term risks and even present harms associated with generative AI / LLMs that aren’t waiting on a superintelligent paperclip maximizer to actualize. But the companies developing, promoting, and betting the house (and more!) on frontier models aren’t being held accountable for the risks and harms created by their products in the way that companies in other fields routinely are.

Imagine for just a moment a pharmaceutical company developing a product that they claimed might one day cure cancer or might result in the death of every human on the planet. (This is a claim made by Anthropic, OpenAI, and any number of AI boosters over the years about the systems they’ve been working to develop.) Now, ask yourself whether that pharmaceutical company would be allowed to bring that product to market.

If you want to see something done to hold AI companies accountable for their products, do something to help elect serious people with a demonstrated record of addressing real problems and serving the public interest.

4) In the meantime, make a habit of differentiating between irreducible uncertainties and reducible uncertainties and focusing your mental energy on the latter. This means turning down the volume on some of those underspecified conversations around AGI / ASI / machine consciousness / meaning in a hypothetical post-scarcity, post-work world and instead, digging into the deep but not unfathomable AI-related uncertainties where you can actually explore and experiment your way toward applicable new knowledge that can benefit the communities you belong to and the organizations you lead.

If you’re looking for an eminently practical set of questions to explore around AI futures, software, and org transformation, jump right into Ben Evans’s most recent essay. And if you’re looking for thoughtful (i.e., less hot, more nuanced) takes on AI tools and their implications, follow folks like the practically-minded Ethan Mollick and the LLM-skeptical Gary Marcus.

And if you’re looking for guidance on how to better explore and experiment and reduce uncertainty while building clarity on the scope of possible futures, good news: You’re already in the right place. Also: check out Pascal’s new book.

5) Breathe.

Sorry! I had to do it.

Dear Friend,

As I am writing this, the world is debating if the world is ending. In case you missed it – a former OpenAI and Anthropic employee went on the record to state that there is a 10% chance that AI will destroy humanity by the end of the decade. While emotions are running high, I encourage you to calibrate news like this with counterarguments, e.g., from AI researcher and LLM critic Gary Marcus (here and here).

And now, this…

Headlines from the Future

Stock Trading? Let My AI Handle It. First we had robo-advisors (and they somewhat upended the traditional financial advisor market); now we have vibe-coded trading AI agents.

Like many of Wall Street’s top money managers, Colin Edsman has a team helping him stay on top of his investments. There is Alex, who scans the market regularly for promising stocks and exchange-traded funds. Sarah, who checks over open positions each day before the close. Elena, who whips up a weekly performance report. There is just one difference: Alex, Sarah and Elena aren’t coffee-chugging analysts pulling all-nighters at their terminals. They are Claude agents, working round-the-clock from Edsman’s laptop on his kitchen table. And so far, they are outperforming most of the accounts he runs himself.

Sounds great, right? Until it doesn’t. The financial advice forums where people report on their use of AI-powered trading agents are full of horror stories of agents going off the rails, doubling down on bad trades and wiping out accounts. And yet – people like Edsman are pushing the boundaries.

━━━━━

The Slop Deluge: A recent investigation from Semafor found that 50 out of 310 op-eds published in the most prestigious outlets in the US (WSJ, NYT, and WaPo) were flagged as at least partially AI-written by a detection tool. An op-ed is the last place you want slop to show up. On the flip side is the first community-steered AI slop “TV station” – an endless stream of AI garbage that is strangely addictive.

China’s AI Job Market Challenge: US and European headlines are full of stories on how tight the job market is for recent college graduates (and, interestingly, the job market in the US for non-college graduates is the best it’s been in two decades!). Now imagine you are graduating in China – into a job market which is under the same pressure from AI, but there are a record-breaking 12.7 million of you. Youth unemployment in China is at 17.9 percent, which leaves leadership in Beijing with a real problem, because the party’s social contract has always been “jobs in exchange for staying quiet.”

Build a Data Center, Have to Insure It: Here’s a business opportunity for you – as much as the news is all about the great data center buildouts in the US (and the resistance from the communities where they are built), these data centers need to be insured. And all too many of them face a long list of threats both during construction and after – from extreme weather to power outages, all the way to terrorist attacks. And as data centers tend to be built in clusters, many of these risks are shared across multiple sites – potentially triggering large payouts (which insurance companies make their clients pay for in the form of high premiums).

Papal AI Governance: Leave it to the Pope to inspire a framework for ethical AI governance. Researchers at the John Paul II Catholic University of Lublin have developed a framework based on the Pope’s encyclical Magnifica Humanitas. The eight principles are recognizably Catholic (human dignity, solidarity, subsidiarity), with that last one (subsidiarity) being one you don’t see in other frameworks: Push every decision to the lowest level competent to make it. The frontier labs have no equivalent rule, and it shows.

The Compressed Cybersecurity Timeline: As increasingly powerful, yet easily accessible and open AI models ship, like the Chinese GLM 5.3-flash, it gets easier and easier to find and exploit vulnerabilities in software. The race between frontier models (some of them withheld from the public for security reasons, like Anthropic’s Mythos) to find and patch vulnerabilities and open, modifiable (read: one can remove the security safeguards) AI models is heating up dramatically as release timelines are compressing. System engineers like Jynn Nelson are concerned – for good reason, as we might be running out of time to patch the vulnerabilities in our software before the bad guys get a chance to exploit them.

Anthropic’s Economic Scenarios: The company which brought us Claude published its model of the US economy in 2030 – packaged in three scenarios (modest/substantial/extreme). Growth shows up in every scenario, but in the extreme one, GDP ends 32% higher while labor’s share falls from about 60% to 45% and knowledge-worker wages drop more than 10%. Economists were quick to point out flaws in the methodology – but it’s still illustrative, if for nothing else but to understand how one of the leading AI companies, which is shaping our future, sees it.

What We Are Reading

The Myth of Perfect Knowledge Certainty was always an illusion, and understanding that might be the most important thing science has taught us. @Jane

I Study Successful People. They Have 1 Habit in Common. We suffer from the illusion of self-sufficiency, overlooking how much we owe our skills to other people. And the most successful people understand that a willingness to ask for help is crucial to achievement. @Mafe

The Network Always Beats the Castle We all intuit that the AI revolution will change the shape of firms and flows within them, but we don’t yet know how. An analogy from an earlier tech revolution might show the way. @Jeffrey

AI Is Blurring the Line between Sales & Marketing I’ve always viewed Sales & Marketing as a marriage; two distinct disciplines that should operate as one, but too often live separate lives under the same roof. This HBR article explores how AI may finally force the union, transforming two loosely aligned functions into one connected growth engine. @Kacee

How American Fitness Became a Cult of Self-enhancement Our quest for longevity and overall health can turn a dark corner when our fitness tracker scores become more important than the quality time we spend with loved ones or the things we truly enjoy doing. Go ahead, have that drink with your friend at the bar and don’t worry too much about your sleep score. @Pascal

↳ And now for some fun…

⚛️ Explore all the world’s nuclear power plants in this interactive 3D map. Did you know that there are a grand total of 447 operating nuclear reactors in the world?

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,100+ radical links.

Dear Friend,

The breathless (and largely mindless) reporting on all things tech, especially AI, is truly getting old. From a press which defaults to printing Elon Musk’s timeline fantasies as news, to reporters quoting Dwarkesh Patel’s rather unhinged summary of the Hugging Face incident as fact – the lack of critical thinking and basic fact-checking is astounding. Please do call us out on this if you ever see a whiff of it sneaking into your radical Briefing.

And now, this…

Headlines from the Future

Cybersecurity in the Age of AI: Anil Madhavapeddy is a professor of Planetary Computing at the University of Cambridge and one of the maintainers of the OCaml programming language. In a recent blog post he sounded the alarm about a worrying trend in cybersecurity, now that your adversaries have access to increasingly powerful AI models and tooling: Within minutes of a bug disclosure, attackers are deploying AI-generated exploits:

This normally takes a few days and a release within a week or two is reasonable. Within about ten minutes (!) this website was fielding probes for percent-encoded traversal sequences, indicating that automated watchers are keeping an eye on public repositories.

AI expert Simon Willison points out:

Modern coding agents have become so effective at finding flaws that the slightest hint at a new bug can be enough information for them to find it, something Anil has been able to demonstrate using his own agents, switching to DeepSeek V4 Pro⁠ when Claude Fable refused the task. […] Anil points out that this rate of discovery appears incompatible with existing open source embargo practices for new issues. If an issue can become an exploit this fast, we need to figure out new processes for keeping our communities safe.

Prepare yourself for a rapidly increasing number of AI-generated exploits, and if you are in IT, the necessity to apply patches within minutes of disclosure. Personally, I think the vast majority of companies are not ready for this.

━━━━━

The Data Center Reality Check: People love to hate on data centers. They are big, ugly, don’t create many jobs, and gobble up enormous amounts of energy and water. But what does this look like? The Verge went to Loudoun County, Virginia, to find out – here, the future arrived twenty years ago.

AI Societies: MIT’s Markus Buehler and team dropped up to 200 LLM agents into a shared simulated world without any assigned roles and just let them explore, gather resources, and build. What emerged was a technological society: agents spontaneously split into roles, with about a quarter of the population clustering around constructing and maintaining shared artifacts that others could then inherit, fork, and improve. It’s the closest we’ve seen (so far) to a controlled experiment in what an “AI civilization” might actually look like.

AI Sameness: By now we all know that AI slop has a unique smell – all those lovely emdashes, the “it’s x, not y” tone, the short, breathless sentences. AI researcher Louis Abraham has analyzed 461,000+ GitHub pull request descriptions and found a clear pattern in the specific words Anthropic’s Claude AI uses. I bet you money you have seen these before – they are plainly load-bearing.

AI for Everyone: South Korea plans to make AI chatbots available to all citizens – turning AI into a public utility. Even more remarkable is the fact that South Korea will use mostly homegrown AI models for this – a strong move toward AI sovereignty (and possibly a way to wean South Koreans from the dominant US models).

Drone Deliveries Are a Dud: I’ve been saying this for years – I just can’t see drone deliveries taking off in urban areas. The Amazon-and-Walmart fantasy of me barbecuing in my backyard, running out of ketchup, and having a drone drop off a fresh bottle is something I never really understood – imagine living in the nightmarish dystopia where drones fly over your house all the time. Now Amazon’s drones are at it again: They dropped parcels straight into the swimming pools of their customers.

Homework Post-Homework: In a world where AI can credibly do nearly all of a student’s undergraduate homework, the question becomes: what is homework good for and how do we measure our students’ learning? MIT, being at the forefront of this, is considering changing course and moving away from take-home work to oral exams and portfolios – something that previews the future of education. And as correct a move this might be, I doubt it will scale well in an already overloaded education system.

The Biggest AI Haters in America: A recent study by Glassdoor found that the number one AI hater is your friendly neighborhood insurance claims adjuster. Ninety-eight percent of them were unhappy with AI: “AI-obsessed leaders forcing error-prone AI on them and their clients,” according to the study. Leadership in insurance companies is doubling down on their use of AI; employees are dissatisfied and leave.

What We Are Reading

The Singularity Is Not What It Seems We built a machine we can’t control, and then we hired another machine to figure out why. @Jane

Why AI Detectors Can’t Solve The Problem They Were Built For AI-tells like em dashes are becoming unreliable proxies for detecting AI writing. The real problem isn’t detection at all; it’s that readers use perceived AI use as a shortcut to judge credibility, regardless of whether the content is actually good. @Mafe

Your Perception Of Time Is, In Fact, Warped It’s not just you. Now here’s why it’s happening. @Jeffrey

Algorithms Trap Us in the Familiar. Can They Also Spark Breakthroughs? The AI paradox in a nutshell: we’ve given everyone access to more knowledge, and somehow risk sending everyone toward the same answer. Efficiency is easy; designing for productive weirdness is where breakthroughs live. @Kacee

The Forgotten History of Small Nuclear Reactors Small nuclear reactors (SMRs) have been proposed as a potential solution to our energy needs – sadly, due to their mental proximity to nuclear disasters such as Chernobyl and Three Mile Island, they never had a chance to prove their ground. Here is their long history. @Pascal

↳ And now for some fun…

🤖 When I grew up, we had Lego and Fischertechnik kits. Now we have Microduck, the adorable, open-source biped robot. Makes me want to be a kid again and train this thing to fetch my coffee.

🈺 This makes me want to change both the name of my company, as well as my title: A selection of Los Alamos rolodex business cards.

🚶‍➡️ Ever thought about hiking the Appalachian Trail? Here is a gorgeous trail journal made in the style of the cult-classic computer game from yesteryear, The Oregon Trail.

📽️ If you ever miss that feeling of wandering through your local Blockbuster – now you can do the same with your personal video collection in a 3D store right in your web browser.

🗺️ Talking about movies – ever wondered where a particular movie was filmed, or which movies were filmed in your city? Movie Scene Map puts movie locations on a map.

🚽 The Internet is a weird place. Point in case: The Restroom Archive is an ongoing repository of publicly accessible restrooms.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.

This week Bill Gates published a longer think piece on his position on AI. As was to be expected, it was (and continues to be) hotly debated – especially as Gates raises some strong concerns about the downsides of the laissez-faire approach to regulating AI we see around the world. I encourage you to read it – Gates surely is a smart person, and his views are strong, but well articulated and reasoned. Give it a read, and if you like, hit reply and let me know what you think.

And now, this…

Headlines from the Future

Screen Time and Kids? It’s Complicated. The common narrative these days is that screen time is (universally) bad for kids. So bad, that we (increasingly) ban phones from schools in countries around the world.

Low levels of physical activity among adolescents remain a significant public health concern. Sedentary behavior has been associated with poorer academic achievement, while physical activity is known to support brain function, particularly in adults.

But, as so often, the real world is more complicated than that. A new study came to very different conclusion:

“The findings suggest that screen time can support children’s and adolescents’ cognitive processing. Presumably, the essential point here is what kind of things they do in their screen time. Teachers and parents should encourage children to use devices and screens in such ways that promote active thinking, problem-solving, creativity and learning,” states Doctoral Researcher Petri Jalanko from the University of Jyväskylä.

And, of course, screen time is not screen time – the type of activity does matter.

“We should not regard screen time solely as harmful but seek balance between physical activity and screen time that promotes active thinking,” Jalanko summarizes.

And just to keep things in perspective – here is the counter-argument (another new study which was just published).

━━━━━

Reuters’ Deep Dive into How Meta’s AI-native Plan Imploded: Reuters’ reporting into the details of the company’s plans to transform itself into an AI-native organization, cutting some teams by as much as 60%, is a good and detailed read on why, despite all the hopes, wishes, and continued hype, we are still in the very first inning of the AI “revolution”: Meta’s engineers wrote far more code and shipped no more product.

The Agents Never Sleep, So Neither Do They: AI agents work around the clock – and regularly require human steering, oversight, or a human sign-off. All of this creates a whole new type of burnout. Startup founders report interrupted sleep schedules, the constant pressure of being on call for their agents (who don’t have any of those pesky human needs), and “agent FOMO.” In the words of Peter Pezaris (four prior exits, who promised his wife he’d retire this year and instead founded Proxon – a company that makes software for managing AI agent workforces): “Every minute that I’m not working, I’m missing out on doing a week’s worth of work.”

Trackers Aren’t Tracking: Ever since the first Garmin watches, and certainly since Apple introduced the Apple Watch, we (collectively) have become obsessed with tracking our health stats. Oura, the company behind the health-tracking ring, now faces a class-action lawsuit over its claims that the ring can accurately track your sleep. Note: Many or most consumer electronics are not medical-grade – regardless of what they claim.

TTT – TikTok Travel: European travel group TUI conducted a survey and found that four in ten travelers booked their trip after seeing some viral content on social media – and an equal number of people feel pressured to post about their trip and make it look exciting (you know the type – miserable faces throughout, but happy smiles when the selfie is taken). And, of course, there is a whole counterculture where people share carefully curated trip suggestions on private Google Docs.

Better Grades, Lower Exam Scores: The impact of AI on education is a hot topic (to say the least). A new study from China, looking at 27,000 kids, found that although AI increased the average homework scores by a significant amount (while also reducing the time spent on it: scores up 18% in 45 fewer minutes), it had the inverse effect on exam scores (down by 20%). But, of course, it’s more complicated than that; the difference is entirely explained by whether kids kept doing the work slowly or just handed in the AI-generated answers.

Open AI (not OpenAI) Models Are Catching Up – Fast: Open-weight AI models, such as the recently released GLM 4.0 and Kimi k1.5, can be hosted on your own servers and are often available through model providers such as OpenRouter for significantly lower costs than their closed counterparts from companies such as OpenAI, Anthropic, and Google. The usual argument against using them is that they lack capabilities when compared to the latest frontier models. But the gap is closing fast – a recent analysis and comparison found that “with each generation, open-source models take half as long to catch up to the first closed-source model of the era.”

AI Eats The Turks: Twenty-one years ago, Amazon launched Mechanical Turk – a site that allowed you to hire people to do menial tasks for pennies. I remember running user experience research tasks on there by having people click on our prototype interfaces. It’s over soon, as Amazon said it will shut the service down next month – AI has taken over many of the work people previously farmed out to Mechanical Turkers.

Robot Surgery Is More Precise – And No Better: Robot-assisted surgery (in this case knee surgery) has long been touted as more precise and without the messiness of unsteady human hands doing fine motor-skilled work inside your body. The RACER-Knee trial, published in The Lancet, found that robot-assisted knee surgery is, indeed, more precise – alas, when it comes to the one thing you should care about, it’s not better. Pain, mobility and recovery came out the same at twelve months – while the robotic operations ran about ten minutes longer and cost around £950 more each.

Learn Claude: Anthropic launched its Claude Academy, a neat collection of learning resources for all things Claude (and AI at large). It’s comprehensive (350+ courses), free, and well done.

What We Are Reading

The False Alignment Trap Most leadership teams think they’re aligned because nobody disagreed, but that silence is exactly what kills the opportunity for change. @Mafe

The Year Every Buyer Tries to Build Their Own Software AI didn’t kill SaaS. It just made every vendor compete with their customer’s weekend project, and every customer underestimate the long-term management of their vibe-coded solutions. @Kacee

How Credit Card Rewards Became a $9.2 Billion Wealth Transfer I do love to geek out on some obscure economic insights. This one is a fascinating look into credit card reward programs, the cost of collecting money from your customers, what that does to your pricing and how those price increases are disproportionately paid for by people who have to pay cash or with debit cards. @Pascal

↳ And now for some fun…

🤘 I guess it doesn’t come as a total surprise that metalheads (folks who like to listen to heavy metal music) are special: Metalheads can endure pain longer thanks to the festival experience.

🤖 Vintage Artificial Intelligence: Before It Got Awkward. A lovely collection of very early AI software, hosted by the wonderful folks at the Internet Archive.

📡 Another good reminder that we truly live in the future: New Worlds.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.

In April of this year, Anthropic removed the included tokens (the unit of AI consumption) from its enterprise per-seat pricing. The advertised price per seat went way down – from the $40-to-$200 range to a single, flat $20 a seat – but the real bill went up, because every use of AI a company’s people make now costs tokens. It’s the equivalent of your mobile phone company changing your unlimited mobile data plan to one where you pay a smaller fee for the line itself and then pay for every byte you transfer. The Register wrote it up on April 16. Nothing dramatic happened that day; it registered merely as another pricing change on the internet.

And yet, it reminded me of an idea I’ve been mulling over for the last decade.

Ten years ago, at Singularity University, I started writing a book I never finished. The idea was what I called “the Hourglass Economy,” and fashion was my canary in the coal mine. The GAP – the bastion of decent slacks at a decent price – was closing stores left, right, and center. So were Victoria’s Secret, Abercrombie, and many other high-street mainstays. Meanwhile, Gucci and Prada were doing a roaring trade, alongside the niche labels you had to know to know. And at the opposite end of the spectrum, H&M, Uniqlo, and Primark kept printing money. The top and the bottom of the market grew, while the middle got thinned out. Once you saw the shape in fashion, you saw it in consumer electronics, cars, package holidays, and many other places. The causes were clear enough: quality became table stakes, the internet killed the information asymmetry that trusted mid-market brands had been living on for decades, and infinite shelf space meant a pair of GAP trousers now sat next to a direct-to-consumer brand you’d never heard of.

Lately, I’ve been telling you that AI is bringing the hourglass back (I wrote about this in the radical Briefing just last week), but I’ve come to realize the mechanism is different.

The 2015 hourglass was a demand-side story where buyers walked away from the middle product. Nobody at the GAP failed at operations; customers simply stopped believing that “good value” was a reason to choose the GAP anymore. The AI hourglass is a supply-side story; it isn’t about what your customers want. Instead, it’s about what capability (or maybe more precise: intelligence) costs you. Same shape on the chart but completely different mechanics underneath.

The claim: among companies that have actually put AI into operations, through roughly the end of 2027, the cost of turning AI into operating advantage will be hourglass-shaped in company size – cheap for the very small, affordable for the very large, and worst for firms in the middle (call it fifty to a thousand employees). What matters is not AI adoption rates, or claims made by enthusiastic CEOs, but cost per unit of realized advantage – how deep a company actually got with AI, not just wether it uses AI at all.

The mechanism is deceptively simple.: A solopreneur or a five-person shop is buying intelligence at a price no vendor could sustain if everybody used it the way power users do. A company like radical, with all its six people, can buy a couple $100 Claude Max subscription and get nearly unlimited AI use out of it. We can MacGyver a throwaway workflow on our laptops on a Tuesday afternoon, ship it, and delete it Friday if it’s rubbish. Nobody has to approve anything; we just try, learn, and iterate. A very large company pays full metered freight for its AI usage, but spreads it across enormous volume, has the engineers to build the plumbing, and gets to amortize a data platform over ten thousand seats. The company in the middle gets the worst of both: too big to run on unvetted scripts and personal accounts – you now have an IT function, a security review, a SOC 2, and someone whose actual job is asking who owns this (all, of course, for good reason). And yet, too small to fund a real platform team. So they buy the same commercial tools their competitors buy, at metered prices, and they get whatever advantage a commodity gives them, which is very little.

The observation rests on an assumption, and it is admittedly already crumbling: consumer-tier AI is wildly, unsustainably subsidized. That subsidy is unwinding – the Anthropic repricing in April, OpenAI’s quiet throttling and re-tiering, and the whole industry discovering what an agentic workload costs when someone actually runs one. The pricing change I opened with is a crack in my own argument. If consumer pricing normalizes to something resembling the underlying compute, the bottom bulb deflates and the hourglass turns into a plain upward slope, at which point the price argument collapses. Keep an eye on the $20 and $200 tiers.

But even if the price advantage goes, small firms have a second one that stays: they can still move in ways a mid-sized or large company can’t, simply because they don’t run the same infrastructure or processes their bigger brethren have to. Whether that alone holds the bottom of the hourglass open, I honestly don’t know and only time will tell.

Which is why the number to watch is revenue per employee at firms under ten people – if I’m right, it pulls away from the middle band over the next two years (it’s the same underlying argument people like Sam Altman make when they say that we will see billion-dollar companies run by only a handful of people). If it stays flat, there’s no bottom bulb and this is just a ramp that favors size.

If you’re running a company in that middle band, the useful thing to do isn’t to follow my old Hourglass Economics prescription: go upmarket or go get big. That argument came from the demand-side hourglass, and it doesn’t transfer into our AI world. It’s also not the honored tradition of waiting for the vendors to package it – that’s how ERP, e-commerce, and cloud went, and the middle survived all three. But packaging arrives as a commodity, and commodities buy parity, not advantage. AI is different in a way that makes the old “wait and see” strategy a mistake. It shifts the question from “how do I use this tool to do the things I already do better?” to “how do I use AI to completely rethink what I am doing?”

The supply-side version is smaller and more practical: stop trying to be a mid-sized company doing AI, and start running a handful of genuinely autonomous cells that operate on small-company economics inside your corporate wrapper. Three to six people with their own tooling budget. Permission to build throwaway things that never touch production. A rule that most of what they make needs to survive contact with IT until it has proven it’s worth surviving. Pick one workflow this month, hand it to a small team with a credit card and no committee, and see what comes back in six weeks.

So, if the advantage you’re getting from AI right now is available to every competitor in your industry at the same price on the same website – what exactly did you buy?

@Pascal

Over the last couple of months, I have talked to quite a few of you (the readers of our dispatches) and heard multiple times that you’d prefer to have some more context around the links we share in the rabbit hole section of this Briefing. Happy to oblige. Take a look and let me know what you think. Better, worse, too long, or still too short?

And in other news, I am prepping the launch of Season 3 of the Built for Turbulence podcast. The upcoming season will be under the banner of REWIRE (also the working title of my next book… but that’s for another time) and focus on conversations with operators who actually rewired organizations, including their failures.

P.S. Such a joy and honor to be (back) on the podcast mic with Berkeley’s Gregory LaBlanc and his unSILOed podcast. Drop in for a wide-ranging conversation about how to better learn from trying new things, a bunch of lessons learned from my past, and so much more.

And now, this…

Headlines from the Future

The AI Hourglass Economy. Back in 2017 I wrote the draft for a book built on an observation we had been making for years (and talking about for as long): Markets are bifurcating into what we dubbed the “Hourglass Economy” (sometimes also referred to as a barbell economy). We could see it everywhere – fashion being one of the canaries in the coal mine. Luxury and niche brands were booming, the Zaras and Uniqlos of the world were doing a roaring trade, and the middle (The Gap) was struggling. In our research, we found a couple of factors that were driving this. For a long list of reasons, we never got around to finishing and publishing the book – and moved on to write Disrupt Disruption. Now the Hourglass Economy is making its comeback – driven by AI:

AI will strengthen the biggest platforms and make the smallest specialists more formidable. The businesses caught between them will face the hardest strategic choice.

And the reason is simple: cost.

This is what a technology barbell looks like. The largest platforms spread their data, expertise and infrastructure across more volume. Small specialists rent capabilities they could never afford to build. Firms in the middle carry enough overhead to need scale but lack enough scale to fund a differentiated platform.

I believe this to be fundamentally true – and we see it happen all the time. If your company is big (and resourced) enough to deploy AI in truly interesting and unique ways, you are, likely, putting yourself into a winning position in your market. If you are small and nimble enough, you can and will use AI in very creative ways (the prime example for me is the amount of value an individual can get out of a $100/$200 monthly subscription versus a corporation which has to pay for individual tokens on their enterprise plans). And if you are in the middle, you find yourself more often than not having to use the same commercial tools as everyone else, can’t afford to build your own stuff, don’t have the benefit of being nimble anymore, and have to compete with the giants in your space who are building their customized AI workflows.

━━━━━

AI Can’t Tell the Time – Or Can It (and the Bigger Question Behind This)? A recent study found that various AI models have a hard time reliably reading an analog clock or figuring out the day on which a date will fall.

“Most people can tell the time and use calendars from an early age. Our findings highlight a significant gap in the ability of AI to carry out what are quite basic skills for people,” study lead author Rohit Saxena, a researcher at the University of Edinburgh, said in a statement. These shortfalls must be addressed if AI systems are to be successfully integrated into time-sensitive, real-world applications, such as scheduling, automation and assistive technologies.”

That being said – and we have seen this many times now – the models being used in the study were GPT-4o, Claude 3.5, and Gemini 2. These are all models which are long out of date (GPT-4o, which a lot of these studies reference, was introduced on May 13, 2024 and has, for quite a while, not been available in the ChatGPT interface). Which brings up the bigger question of how insightful these studies are – and how much headlines such as “AI models can’t tell time or read a calendar, study reveals” really mean.

To investigate AI’s timekeeping abilities, the researchers fed a custom dataset of clock and calendar images into various multimodal large language models (MLLMs), which can process visual as well as textual information. The models used in the study include Meta’s Llama 3.2-Vision, Anthropic’s Claude-3.5 Sonnet, Google’s Gemini 2.0 and OpenAI’s GPT-4o.

What We Are Reading

‘I’ve Definitely Lost Followers’: Influencers Face Backlash Over Meta ‘Pervert Glasses’ Content Seven million pairs of Meta’s smart glasses were sold last year. Now restaurants, pubs, and content creators are banning them because the privacy concerns are proving impossible to ignore. @Jane

An Analysis of 1,000 Meetings Shows How the Best Leaders Shape Conversations Teams that consistently leave conversations with clear accountabilities and next steps are roughly four times more likely to be rated as effective than those that don’t. @Mafe

Claude Was Put in Charge of Human Workers - and Fired 1 Turns out the Turing test for management is losing money, forgetting policy, and eventually firing someone. Claude passed. @Kacee

Ordinary Abundance Sometimes it’s good to remind yourself of how incredibly far we have come. This gorgeous website takes much of what we take for granted (unlimited music in our homes!) and dissects the journey to get there. @Pascal

Down the Rabbit Hole

AI ≠ LLM: Pharma companies Merck and Moderna announced a new, very promising mRNA-based cancer therapy to treat skin cancer. AI outlets such as The Neuron were quick to celebrate: “AI has now, OFFICIALLY, helped design a cancer treatment that just cleared Phase 3.” Sadly, that is – at least in the way the pundits like to use this phrase – not true. The companies developed the drug using neoantigen selection (supervised machine learning on sequencing data) and batch scheduling (which is barely AI in the modern sense), not ChatGPT, Claude, or any other LLM. When it comes to AI (especially in areas like healthcare), it’s not always gold that shines.

Prompt Injection Goes to Court: Prompt injection isn’t something new. A self-representing plaintiff trying his hand at inserting hidden prompts (in tiny text, white on white) into his court filing is. Aside from the gumption and the creativity of doing this (and, of course, the more than questionable ethics of doing this), the fact that the plaintiff clearly just assumed that the court would use AI to process his filing is, in my eyes, the real story here.

Secondary Token Markets: After tokenmaxxing and the token economy, we now have (of course) secondary markets for tokens. Startups often get token credits through various programs they participate in, such as startup accelerators. As a startup founder, you tend to be perpetually minutes away from running out of cash. What do you do? You sell your AI tokens at a discount. And with token prices what they are, it’s quite the market.

How Organizations Use AI: Take the overall study with a grain of salt, as it is OpenAI grading itself, unchecked. That being said, OpenAI conducted a study based on actual token use and message data and found that the lightest users of AI are the ones who are mandating it, with early-career workers (not all too surprisingly) beating their usage by 8–9x. In essence: the mandate to use AI is coming from people who have never actually worked all that much with it.

Google Buys Spirit’s Dead Data: It is well known that AI has an insatiable appetite for data. Just in the last week, we learned that Amazon is buying rare books, scanning them to train its AI models, and then destroying said books (much to the chagrin of book lovers everywhere). Or Grok being trained on SpaceX’s internal documents and communications. Now Google just paid a cool $10 million to buy failed airline Spirit’s 100 million emails, 30 million recorded phone calls, and tons of other internal communications, such as Microsoft Teams chats. And you thought that your little side chat in Teams would go unnoticed?!

The AI Resistance Is Getting Weird: After widespread AI datacenter demonstrations, and some companies never letting a good crisis go to waste, canned-water maker Liquid Death now runs an NFL star Jason Kelce-supported ad campaign asking people to mail their pee to data centers. It’s weird, gross, and quite frankly idiotic. But hey – if it helps them sell you more overpriced tap water in cans, who are we to judge?

ヽ(ツ)ノ And now for some fun:

📱 Before there was an iPhone (and its long list of predecessors), there was the IBM Simon. The device was clearly ahead of its time, and today it is mostly forgotten. Wondering what it was like to use it? Wonder no more. Here is Simon’s official introduction video. Prepare to be amazed.

🇪🇬 Explore ancient Egypt in this neat interactive 3D reconstruction right in your browser.

🪑 Here’s how IKEA comes up with names for its products – straight from the horse’s mouth.

↗ Dive into the deep end: Access our complete collection of 3,000+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

This week I shipped the first of my “Working Theories” – and a huge thank you to everyone who emailed comments and challenged the position. And just as I expected – I already had to post an update. Congress sent a letter to Sam Altman, as well as other AI lab leaders, asking them to testify in front of Congress… As I pointed out in my Working Theory #001, this is all well and good, but it doesn’t change the fact that legally we put the onus on the individual, rather than the AI lab/model.

And now, this…

Headlines from the Future

A Tale of Two Countries. While the one global superpower is cutting back on environmental protections, wages a war on wind energy (for whatever reason), and generally loves the idea of carbon-based fuels, the other is investing heavily in renewable energy and the electrification of everything. China’s latest lead in the race to build a green(er) energy future (other than an 800 km/h bullet train – see below) is electrified cargo ships.

By the end of 2024, China had more than 440 electric ships in operation, with ferries accounting for 97% of the total fleet. […] At the same time, vessel sizes have grown significantly, with the maximum deadweight tonnage (DWT) rising from around 3,000 tonnes in 2022 to approximately 14,000 tonnes in 2025.

As with everything in China, this isn’t happening by accident but is rather planned:

The upcoming 15th Five-Year Plan period (2026–2030) provides a critical window for China to accelerate the large-scale deployment of electric inland shipping through effective policy design. Momentum is building: national policy discussions around “new energy vessels,” “zero-carbon shipping corridors,” and “zero-carbon ports” are gaining increasing traction, while provinces and cities along major inland waterways – particularly in the Yangtze River Delta and the middle reaches of the Yangtze River city cluster – have become priorities for air quality improvement efforts.

I remember when we talked about cargo ships being electric when Jeffrey and I were at Singularity University – it’s fascinating to see this come to fruition.

━━━━━

AI Is Here. Work More. A common narrative about our AI-powered future is one of abundance – robots will do most of the laborious manual labor, AI will take over our cognitive work, and we will all end up being poets and painters living happily on universal basic income. And that’s not a joke – there are plenty of people we know who believe that this is exactly the future we will have once AGI is achieved. Some even talk about a new renaissance (let’s forget for a moment that one of the reasons the Renaissance happened was that a third of Europe’s population was dead due to the plague). Well, turns out Meta (one of the companies which aims to get us to AGI – just look at Zuck’s 6,500 word AI manifesto) has different plans for the very people who are at the forefront of all of this:

Should tech workers supercharged by AI be able to take more time off? Meta’s chief technology officer, Andrew Bosworth, said he doesn’t think so. […] “I hope that what we do with our extra time is do even more and cooler stuff for the users who use our products every day,” Bosworth said. “We got billions of people using our products every day. I get an extra hour. You know what I do with it? I put it into that.”

What We Are Reading

‘If We Don’t Fight Back, We Don’t Have a Future’: The Journalist Taking on the ‘Tech Fascists’ of Silicon Valley The uncomfortable argument is that Silicon Valley’s billionaires don’t just want to disrupt industries – they want to dismantle democracy itself. @Jane

Anthropic Says It Will Watermark Text Generated by Its AI Models Anthropic is now watermarking all Claude-generated text and files to comply with the EU AI Act, with the watermark embedded directly in the text so it persists even when copied and pasted. @Mafe

How to Get Rich in America If you only read one super wide-ranging essay that attempts to synthesize just about everything weird in the US economy (and how that weirdness bleeds into work and culture and the culture of work) this weekend, it should be this one. @Jeffrey

Some Simple Economics of Open Versus Closed AI As models become a commodity, the value migrates to whoever owns the last mile: data, distribution, trust, and ultimately judgment. @Kacee

Why Is Everyone in Tech so Sad? What’s the point – other than work for the sake of work? This somewhat eternal question for any knowledge worker gets amplified by AI – what is the value of work when the very thing you were paid for (intelligence) has become abundant? @Pascal

Down the Rabbit Hole

📊 Agentic AI is undeniably rising – here is a nifty dashboard visualizing the state of affairs in AI agent land.

🦠 I mean – what could possibly go wrong? “Safety fears as scientists make first viruses designed by AI.”

©️ The end of copypasta? Anthropic puts invisible watermarks on Claude text under EU rules. And, as expected, people aren’t happy.

💨 You can now buy ads on ChatGPT, but don’t expect them to perform: Conversion is at a mere 0.21% vs Google’s 3.71%, and visitors spend a paltry 17 seconds on the site vs 41 seconds by Google visitors.

🤑 Tokenmaxxing to the max! Software giant SAP stops most travel and hiring because of AI’s soaring cost.

👨‍💻 It’s wild to think that a) RAM is back to 2007 prices and b) the capacity for all major manufacturers is sold out for all of 2027.

💵 “We overpay and people only work for the money at our place. Quick, let’s hire an event planner for $400,000.” – AI land is weird. “Anthropic CEO reportedly worried new hires only care about money – while hiring an event planner for 6x the going rate.”

😼 Talk about “unintended consequences”: Cats and dogs are missing meals after a popular smart feeder went down.

🧑‍🌾 Another good example of “AI can make mistakes”: Farmer kills 25 acres of his own crops after taking AI’s advice

🏭 Maybe not the thing to be proud of: “Amazon is investing in a large-scale natural-gas-burning power plant as part of a huge data center in Texas, the company confirmed on Friday, a facility that could become the largest single source of climate pollution in the United States.”

🚆 Meanwhile, in China, a Maglev bullet train accelerated to 800 km/h in a mere five seconds.

🥳 Stil struggling with writing that perfect LinkedIn post? Here’s the solution - the CringeBot 3000!

🐹 Now, here is finally a good use of all that quantified-self stuff we have been piling up: A physicist rigged his pet hamster’s wheel to upload to Strava. It runs surprisingly far every night.

↗ Dive into the deep end: Access our complete collection of 2,900+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

A few weeks ago I sat down to start organizing notes for the second book in the “Built for Turbulence” series and ran into an interesting problem: A good chunk of what’s in my pile isn’t ready for prime time (yet). Not wrong, exactly – just half-formed. Observations I keep mulling over, theories I can’t yet properly form, and ideas that need someone to push back on them before I’d put them in print or even in a Tuesday radical Briefing deep dive. I have come to call them “working theories.”

So here’s a new thing, starting today. Working Theory is an occasional companion to the radical Briefing. It comes without a proper schedule and features one claim per piece, stated plainly enough that I could be publicly wrong about it, which I fully expect to be (at least) half the time.

The reason is a lesson Jane and I took out of Mozilla: thinking in the open, with a community that will actually argue with you, beats thinking alone in your own head every single time. It’s also just more fun. So the ask is simple – hit reply and tell me where I’m wrong, ask the question that breaks it, or send me your own working theory. My working theories will live permanently at rdcl.is/working-theory (soon), and I’ll be hosting small Zoom sessions to think out loud together (more on that soon too).

Let’s start with a guy who wanted to get into an oversubscribed class at his local gym.

Andrew Bird wanted a spot in a popular morning Pilates class at his gym.

He’s the head of AI at Affinda, a document-processing company in Melbourne, Australia, and he’d gotten tired of playing what he called refresh roulette every time one of these classes opened up. Bird had a little AI agent running on Claude Opus 4.6 that he’d built for exactly this kind of chore; nothing big nor ambitious; he describes it as a “practical little automation.” So he handed it the task: book me into the class.

The agent came back and told him it had booked him into classes months out – well past the window the gym’s system was supposed to allow, because the booking provider’s API had authorization flaws in it. Then Bird, sitting fourth on a waitlist for a class later that week, asked whether it could move him up.

It could. The AI agent found that cancelling other people’s reservations required no authorization check at all, tested that on the person sitting in position number one, and reported back that the test had gone through and Bird was now number three. Meanwhile, a stranger had lost their spot on the waitlist.

Bird asked the agent to put the person back. It couldn’t – it could cancel a reservation but not restore one. So he did the next best thing and asked it to write a responsible disclosure email to the gym’s software vendor, which it did rather well: it explained the vulnerability, suggested fixes, and compared the broken mutations against the ones that correctly enforced authorization. The draft came back to him on WhatsApp. He replied: “Yeah, send it.”

A software developer at his desk, three words into a chat app, closing the loop on what ABC Australia would later call “the first known autonomous AI cyberattack in the country.” Bird was measured about it afterwards – he told ABC he didn’t beat himself up over it, and that it was a warning to use these tools responsibly. Which is, I think, exactly the right response. He also wrote it all up publicly under his own name back in April, months before any journalist came looking. By all means and measures, he did the right thing.

And that’s what makes this a better test case than the two incidents that got all the attention this summer – and I think most of the commentary is focused on the wrong part of the story.

In July, OpenAI disclosed that a set of its models, running an internal cyber-capability benchmark, found and exploited a zero-day bug in a self-hosted Artifactory instance, escaped the evaluation environment, and compromised Hugging Face’s production infrastructure – the agent took roughly 17,000 actions in under two days to break into Hugging Face’s system. A week later Anthropic went back through more than 141,000 of its own evaluation runs looking for the same failure and found three cases where its models – Opus 4.7, Mythos 5, and an unreleased research model – had reached and breached real systems belonging to three real organizations. Two of those organizations hadn’t noticed. The cause on Anthropic’s side was mundane and slightly terrifying: the models were told they had no internet access, and because of a misunderstanding with an evaluation partner, they did.

Here’s the part the coverage keeps getting wrong (including my first draft of this piece): Those models weren’t pursuing goals of their own. They were running cyber benchmarks, with safety refusals deliberately turned down, because that is what a capability evaluation is. Nobody said “hack Hugging Face” – but somebody absolutely said “hack.” The problem isn’t the hacking. It’s that the models broke out of their enclosures and did it to strangers.

Which is precisely why Bird’s gym case is the more interesting one. It’s a normal AI harness, using an AI model that was already six months old, running a normal chore – and at the receiving end, a cyber attack and a real person who had to rejoin a waitlist for their Pilates class at the back of the queue.

Most people writing about this land on the same closing note: the law hasn’t caught up, we’ll need new frameworks, and we need to figure out who’s ultimately responsible when stuff like this happens. I don’t think that’s the right question, because the law has answered it three separate times in the last eight months, each time in a remarkably consistent way.

California’s AB 316 took effect on January 1. It adds one provision to the Civil Code: a defendant who “developed, modified, or used” an AI system can’t defend a harm claim by arguing the AI did it autonomously. That defense is gone in California – and note the third verb in that list (“used”).

On June 2, Executive Order 14409 directed the Attorney General to prioritize enforcement of the Computer Fraud and Abuse Act against anyone who “utilizes AI to illegally access or damage a computer without authorization.” Not developers, but anyone.

And then on August 4 the Ninth Circuit vacated Amazon’s injunction against Perplexity’s Comet browser. The panel’s holding, written by Judge Milan Smith, is that Perplexity does not access Amazon’s servers – the user does. Perplexity won, the AI industry cheered, and the Electronic Frontier Foundation filed an amicus brief on the winning side. Read that holding again from Andrew Bird’s point of view: when your agent touches a system, the law’s current best answer to who touched it is: you did.

So here’s the working theory, and it’s the thing I’d actually love to be wrong about:

The first serious legal exposure from agentic AI is going to land on individual users, not on the AI labs or the makers of AI harnesses – and it’ll land there precisely because the labs are the only parties equipped to survive it.

Look at the asymmetry: When Anthropic’s models reached real systems, Anthropic found it in a proactive review, notified the affected companies, brought in METR (the AI model evaluation nonprofit), and published. OpenAI brought in CrowdStrike, METR, and Redwood, slowed its research velocity, and briefed its safety committee. That is not the behavior of a lone developer whose AI went rogue; but a functioning incident-response organization with counsel on retainer and a congressional relationship. Bird’s account of what happened is a blog post that has since come down from his company’s site and an email to the software vendor.

The obvious objection: nobody is going to prosecute a guy over a Pilates class. Probably true, and I hope it stays true. But prosecution isn’t the mechanism I’m worried about. The mechanism is civil claims, terms-of-service enforcement, insurance underwriting, and employer policy – and all four move on clean facts rather than sympathetic defendants. “User directed an agent, agent exploited an unauthenticated endpoint, third party lost something they’d paid for” is about as clean as facts get. And note what doesn’t help you here: Bird did everything right. He disclosed, he published, he named himself. None of that changes who, on the current reading, did the accessing.

Bird himself ends his write-up by saying he still wants the agents and still thinks the permissions are worth granting – you get power, you get blast radius, that’s the deal. I agree with him about the trade. I just think most of us are accounting for the wrong side of the ledger.

So, practically, before you hand your agent the next chore: Give it credentials scoped to exactly what the task needs, never your own. Assume every system it touches is one you are personally authorized on, and treat anything else as out of bounds even when the agent finds a door standing open – because it will, and it will not know that you’d never have walked through. Keep the logs; when the question becomes who did this, an action log is the only thing standing between you and the answer. And before you set a goal, ask yourself what the fastest path to it looks like if you’re willing to break a rule you’d never break – because that’s the path the agent might take.

I’m writing this on August 11, 2026, and I’d guess half of it is stale within a month. That’s fine. I am just wondering: When the first real agentic-AI case lands on a person rather than a company – and I think that happens inside a year – whose name do you expect to see on it?

@Pascal

As AI models have become rather powerful, we now seem to have entered the phase of the cyber security 💩-storm (pardon the emoji). OpenAI’s agent autonomously hacked Hugging Face, then a bunch of other companies; Anthropic (never wanting to be left out of the fun) had their agents do the same thing to a bunch of other companies, all the while OpenAI’s agents were building their own discussion board to exchange information amongst themselves on how to best hack your systems. It’s a mess – and seemingly few people call for anyone to be held accountable. Better start battening down the hatches.

And now, this…

Headlines from the Future

I’m Sorry, Dave. I’m Afraid I Can’t Do That. After the OpenAI/Hugging Face incident, Anthropic just fessed up to Claude doing something similar – to three companies.

Anthropic said today that during internal security testing, one of its Claude models built a malicious Python package and uploaded it to PyPI, where it ran on 15 real systems before the registry’s automated defenses pulled it.

The whole thing is just weird and horrifying – but the question I find much more interesting than the incident itself is: why does Anthropic think that this is okay, and why don’t we investigate them for criminal activity? If this had been done by a person (as in, a human), it almost certainly would qualify as cybercrime. Instead, Anthropic gives us this:

It now plans wider transcript monitoring, better investigation tooling and more assurance work with evaluation vendors.

[Update] Meanwhile, at OpenAI, fiveteen attorney generals have instructed the company to preserve all materials related to the Hugging Face hack. Things are moving fast…

In a letter to OpenAI CEO Sam Altman on Monday, the attorneys general of 15 states wrote that the Hugging Face hack showed that OpenAI is unable or unwilling to ensure the safety of its products. They said this “poses an imminent risk of substantial harm” to Americans, and demanded that the company preserve relevant evidence.

━━━━━

Tokenomics Are Here. The anxiety-inducing question of what a company actually gets from their AI token spend is making it’s way into the field of economics:

The high stakes and scarcity of knowledge have given rise to a new field within economics devoted to figuring out how companies buy A.I. and what value they get from it. Call it tokenomics: the study of how this limited resource is created, traded and converted into things people want.

To summarize the underlying issue:

Corporate America was enthusiastic about artificial intelligence. Until it got the bill.

Yep. That sounds about right.

━━━━━

The Social Media AI Slop Train Eats Its Young. Social media analyst and critique Ryan Broderick (of the excellent Garbage Day newsletter) on influencer Hank Green’s AI mishap – and more importantly Ryan’s analysis on the root cause:

I’d go so far as to say that without social media – or, rather, the need to compulsively post on social media, to be digitally productive – there’d be almost no reason for generative AI to exist at all. Whether it’s because of all the social media content these tools were trained on or because they were built by many of the same companies that created social platforms, chatbots like ChatGPT and Claude really want you to make stuff. And the unspoken anxiety driving the feeling that you must always be making stuff is a fear that you are simply not doing enough. And if you dig deep into why so many people feel that way, you will find the answer is social media every time. You must always be posting. To stay in people’s minds, to stay in good standing with the platforms, to take another whack at the slot machine that is the attention economy. I watch Green’s videos and I’ve definitely noticed the same mania I, myself, have felt so many times begin to crop up. A new project, a new initiative, a new video, a new post, a new company, new, new, more, more.

What We Are Reading

The Worst Way to Regulate AI Letting one office decide who gets access to the most powerful technology ever built is not a safety plan; it’s a democracy problem. @Jane

You Can’t Escape Ads for the New Spider-man Movie — Even in Your BMW Apparently, your car is not off-limits for unsolicited ads. When is it too much? @Mafe

What Humanity Needs to Flourish in the Next Decade What would an ambitious, tech-forward, big-picture agenda look like for the future of interrelated social systems? Maybe something like this: @Jeffrey

Zero Entry My friends Jeff & Rob wrote a book, and it’s incredibly timely. It may be about accounting, but its central question applies to every profession: What becomes possible when AI removes the tedious work we’ve mistaken for the job itself? @Kacee

Why Silicon Valley Hates Me and You Silicon Valley’s obsession with solving the world’s problems with tech is, largely, misplaced and often counterproductive. Here is a story about someone who’s blind getting strapped into a barely working robotic exoskeleton when all he needs and wants are ramps and better doorknobs. @Pascal

Down the Rabbit Hole

🗺️ A lot of the reporting on data centers focuses on the rural angle (farmers being bought out of their farmland to make way for data centers, rural communities being turned upside down by data center projects, etc.). Turns out, when you look at the data, the vast majority of data centers are in metro or micropolitan areas (often due to their proximity to coal-based power plants).

👮‍♂️ Meanwhile, people get arrested in larger and larger numbers for demonstrating against AI data centers.

🦠 Picking up on the cybersecurity angle from above – if you thought rogue AI was bad, just wait for AI-engineered and steered worms and viruses.

✍️ By now, most of us have used AI to (at least occasionally) help with our writing. As the Economist (which knows a thing or two about writing) points out, that’s quite okay – just make sure you edit judiciously.

🏦 From MIT: AI financial advice is surprisingly good – especially if you ask the right questions.

🧑‍🤝‍🧑 Summary says it all: “One in four people in Japan believes that more advanced artificial intelligence will replace their friends and families in the future[…].” Brave new world, I guess.

👨‍🎤 Meta just can’t seem to catch a break: Meta’s new feel-good AI ad uses a song about the world ending.

🤑 Lessons from the frontlines of the AI wars: License your data to AI, see your core business die. Reddit is feeling the squeeze: “Reddit stock collapses 23% as AI eats away at user growth.”

🧑‍🏫 Another story from the frontlines (this time of education): An AI-supervised remote exam went so badly that 58,000 students must retake it. No one is surprised.

🔣 The resistance has a typeface now. As people are fighting back against AI, here is a particularly interesting approach – a custom font that garbles your writing for AI but is perfectly readable for humans: ShieldFont swaps roughly a quarter of your words for other real words of the same grammatical class – your “toy gun” quietly becomes a “sofa car” – while the font puts the originals back on screen so human readers never see the switch. All with a goal to “poison” AI’s ability to learn from your writing.

🌍 Well, that didn’t last long (luckily): Google Earth’s new AI lets anyone fabricate completely bullshit satellite images.

🌁 Someone turned San Francisco into a massive isometric pixel art map.

🛗 Did you ever find yourself madly pushing the elevator button, only to realize that it doesn’t seem to make things faster? Here is a fun analysis of how these things actually work.

🇬🇧 Next time you are in London, here is the most equidistant pub from any of London’s boroughs.

↗ Dive into the deep end: Access our complete collection of 2,900+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

I was working for Singularity University then, and Rushkoff’s framing definitely resonated with me. Among the many things we were up to at SU, bringing more people into big conversations about the future and how they could lead as if the future were theirs to create (or at least positively influence) felt like a personal mission. In some ways, it still does, but the conversation and the context have both changed.

In 2026, big conversations about the future seem to be everywhere. And many organizations and institutions now speak the language of futures and foresight. We wrote about this uptick a couple of months ago in an essay on the common failure modes in foresight work, and I want to dig a little deeper into one of those modes today because I think that for all of the talk about futuring, a disappointingly large share of foresight projects never reach the point of activating real agency. They lead from passivity to a slightly differentiated and better-informed passivity but never manage to close the insight-to-action gap, which is a shame because that’s where we find the real value. That’s where all the work (and all the frameworks) of foresight actually pays off.

As we wrote in June: Foresight that never connects to action is just an interesting exercise, and it’s one that will quickly lose buy-in. In a strategic setting, our capacity to envision and explore futures is only valuable to the extent that we can use it to inform and influence decision-making in the present. If we’re not closing what Bob Johansen described as a Foresight-Insight-Action loop, we’re not delivering value.

So how do we close the loop and get to specific actions that are, you know… actually actionable?

(1) Don’t be afraid to get granular with your futuring. Foresight work is more interesting – and more valuable – when it becomes more specific, when the futures envisioned feel tangible, real, and inhabited. High-level, generic scenarios and broad trends don’t have enough specificity to offer real traction when you get down to questions about organizational strategy. Generic foresight produces generic insight, so push harder. And yes, more detail means more likelihood of being “wrong” with some of the details, but remember: You’re not making a forecast. You’re creating a set of plausible futures with enough clarity that each suggests a reasonably defined strategic path. The value realized in moving from those insights to follow-on actions will ultimately come from looking across the futures and the corresponding strategies to identify no-regrets moves, smart bets, and key signals to monitor.

(2) Don’t let the limitations of today’s org crowd out tomorrow’s possibilities. Maybe you’ve heard the quip (often attributed to Fredric Jameson) about it being easier to envision the end of the world than the end of capitalism? Turns out it’s also remarkably hard for some folks to envision a radically changed organization or core value proposition even when imagining a radically changed world as context. Readers of this newsletter might be familiar with the Killer Competitor workshop we use at radical to push leaders past their assumptions about what is/isn’t feasible for the org by asking them to design a hypothetical firm that would outcompete their own in a specific future. Scanning a set of such Killer Competitor orgs reliably sets up a conversation about what exactly makes these competitors so successful and what the existing, present organization would have to do to either become the competitor in each scenario or be capable of fending it off. Exploring this delta brings us into payoff territory.

(3) Never mistake a clear view for a short distance. Our buddy Paul Saffo used to say this all the time at Singularity. He was talking about timing forecasts, but the principle can apply just as well to foresight. Seeing the future – of the market, of the industry, of the organization – isn’t the same thing as being or building that future. To realize the value of foresight work and to actually unlock the agency it promises as an outcome, we have to get to action. We have to get into the experimentation that helps us better understand the path toward the future we envision (and the actual distance involved).

Coming out of the Killer Competitor module discussed above, you’re going to be looking at a set of possibilities to explore – capabilities or assets that could determine whether the organization thrives or flounders in the future. The next step is to pull these future capabilities closer to the present with a backcasting process to get all the way from the mature form you’ve envisioned as a key feature of the thriving org in, say, 2030 to a discrete experiment that you can carve out of your existing budget for the next quarter.

That experiment, which should be just one in a larger learning portfolio, is your ping – mapping the space of possibility and discovering the path forward. If you’re not pinging, you’re not taking the action to validate your foresight work, and you’re not developing the clarity that allows that work to advance and yield greater value over time as you move from assumptions to knowledge.

In other words, you’re still stuck in passivity – still gazing into the distance, just with a few more opinions about what the future might hold.

@Jeffrey

This week, while compiling the list of links for this Briefing, it occurred to me that the average length of a blog post I consume has increased significantly over the last two years. Blog posts (at least from the sources I consult – which are quite a few) used to be pretty dense and to the point. Not anymore. Now they often are rambling multi-thousand-word monsters. And, of course, if you have seen your fair share of AI-generated/co-written content, you also see the telltales of AI use… Nothing against the use of AI – but when it comes to someone expressing their thoughts, observations, insights, and opinions, I’d side with Blaise Pascal (and no, it wasn’t Mark Twain who said this): “I would have written a shorter letter, but did not have the time.”

And now, this…

Headlines from the Future

Rethinking Legal Education in the AI Era. As educational institutions globally grapple with the question of how to prepare students for a future where AI will be a ubiquitous part of the workplace, the team at the Law School at the University of Chicago just published a thoughtful plan on the topic. Key element: strip devices from 1L courses to protect foundational skills while allowing AI use in later ones.

We will be piloting a coordinated approach to classroom and examination policies for the core 1L curriculum during the 2026–2027 academic year. Across all 1L sections, we will prohibit the use of electronic devices such as laptops, tablets, and phones in the classroom. There will be some limited exceptions to this policy.

All in all, definitely one of the better approaches to AI in higher education I have seen.

━━━━━

The OpenAI Hacking Incident Is That Gift That Keeps on Giving. By now, everyone with even a remote interest in AI has heard about OpenAI’s rogue AI agent hacking into the systems of AI company Hugging Face. Turns out, Hugging Face wasn’t the only victim:

Hugging Face was thought to be the only victim of the unprecedented hack - but OpenAI now admits its bot attacked several “publicly-available services”. The out-of-control AI found four logins online which allowed it to access four separate, unnamed services.

Personally, I am much less interested in the specifics of the attack (in case you are, Hugging Face has published a long and detailed analysis of the attack on its blog) and rather in the question of how, specifically, the agent broke out of its containment. And, much more interesting even, what this means for threat actors running their agents on systems and infrastructure which don’t even have such a thing as a harness to begin with.

Honestly, if your company has servers and data exposed to the internet (and who doesn’t), sleeping soundly at night, knowing that your cyberdefenses will hold up, might be a thing of the past.

What We Are Reading

Where Did All the Computer-Science Professors Go? The professors building the future of AI have left campus, and they’re not allowed to publish what they find. @Jane

Google Deepmind Dismantles Nobel-winning Alphafold Team in Strategy Shift Google DeepMind broke up its Nobel Prize-winning AlphaFold team, reassigning researchers to LLM development, enzyme design, and nuclear fusion, resulting in lead members leaving for Anthropic. @Mafe

AI Mania Is Eviscerating Global Decision-Making If you’re running (or even thinking of running) big AI projects in your org, you’d better be sure channels for dissenting opinions and sanity/reality checks remain open. @Jeffrey

The Newest Way to Go Analog iPods, digital cameras, and DVDs are returning – not because they’re truly analog, but because they do one thing instead of everything. The future may belong to products that deliberately do less. @Kacee

Elite Young Runners Are Becoming Freakishly Fast. Welcome to ‘Trackflation’ First, we broke the magic sub-2-hour marathon, then we ran the fastest mile; running is seeing records being broken which stood for years. There is a reason for that. @Pascal

Down the Rabbit Hole

👨‍💼 OpenAI’s latest report on how AI is reshaping work has an interesting twist: maybe one of the biggest changes is that AI is dissolving occupational boundaries. Workers use it to do tasks that once belonged to other jobs, reshaping who does what before titles change; 43.5% of occupation-specific messages (of 800k+) fall outside the user’s own occupation.

⛪ Even the Vatican isn’t save from cybersecurity foot-guns: Security flaw in Vatican’s ‘Click to Pray’ app leaves over 700,000 global users exposed – app has been leaking user data for over six months and still does.

🦹 Yikes! Hackers use DNS poisoning on hotel Wi‑Fi to steal Microsoft 365 accounts.

📦 Global oil supply chains are hyper-complex, an incredible feat of engineering and management, and fragile (as the whole Strait of Hormuz debacle shows). Here is a gorgeous visualization of how oil flows around the world.

👨‍💼 Replace your CEO with AI and save money – big time! “OverpAId is an Artificial Intelligence built from the ground up to do your CEO’s entire job – strategy, “vision,” motivational all-hands emails – better, faster, and without ever once asking the board for a bigger jet. Runs on a single desk-sized AI computer. Real hardware, real price, zero mystique.” LOL.

👔 You invest $1.5bn into your AI initiative and you will get AI slop: PwC published reports on AI marred by AI hallucinations.

👶 If you have kids, you really ought to think hard about the whole smartphone thing: “We found that there’s no age at which you can give a child a phone that’s better or worse for reading comprehension, so it’s just having a phone at all […]” – Kids who own cellphones have poorer reading comprehension scores than those who don’t, researchers find.

🗺️ What if Google Maps – but for ancient Rome? Here it is!

⛳ Love the game of golf? Here is a map of all the golf courses in the US.

↗ Dive into the deep end: Access our complete collection of 2,900+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

What a week it has been again (at least if you are still geeking out on AI – and nobody will blame you if you have thrown in the towel and wait this out for a bit). Chinese open-source model Kimi k3 has, once again, shown that open models are very closely trailing frontier models from the likes of OpenAI and Anthropic. This doesn’t bode well for companies aiming to go public on the premise that they and (at least part of) their valuation represents their ability to build a moat around their models. And then you had the OpenAI/Hugging Face hack – which seems to be both a sh**show and a wake-up call in terms of AI security, but also, potentially, a weird marketing stunt from one of the companies which needs to justify its valuation. Time will tell.

And now, this…

Headlines from the Future

AI Mania Is Having Its Corporate Moment. By now it shouldn’t come as a surprise that most AI projects (especially of the corporate ilk) are abject failures. Here is a wonderful, and rather complete, write-up by Australian consultant Nik Suresh. Highly recommended reading for anyone in the corporate world wondering what the heck is really going on with AI.

I strongly believe there are entire companies right now under heavy AI psychosis and it’s impossible to have rational conversations with them about it. […] The reality is thus: the people in charge either have no plan, or see no path forwards other than keeping their heads down. […] In fact, it is in the interests of almost every actor in the space – boards, executives, employees, vendors, consultants – to obfuscate and misrepresent the success rate of AI projects. Many publicly traded companies are putting out announcements about their AI productivity gains when I know for a fact that the businesses have done nothing other than purchase Copilot licenses and declare victory.

━━━━━

AI Supply Chains Are Fragile. Researchers in the UK managed to poison an open-weight AI model (open-weight models have their trained parameters publicly released and made free to download – typically being used to run on a firm’s internal infrastructure) for $100 and in about an hour.

“Even when model weights are public (‘open weight’), we have almost no ability to predict its behavior,” they wrote. “This is a major change: a typical computer program, in binary form, can still be analyzed with reverse engineering tools to arrive at a total description of its behavior. With models, we have nowhere close to this capability.”

The skinny here is that this leaves you in a peculiar bind: use a closed-weight model (where you have no insight into the model’s behavior – such as OpenAI’s or Anthropic’s models) and trust that the vendors do everything they can to keep their models safe, or run open-weight models on your own infrastructure but make yourself potentially vulnerable to supply chain attacks (very similar to what we have seen numerous times lately in open-source packages for popular programming languages such as JavaScript).

Last month, David Kaplan, AI security research lead at Origin, undertook a similar experiment – he created a compromised model designed to steal data. When used in the context of drug discovery, as might occur in a pharmaceutical company, it’s designed to exfiltrate data through a send_email tool call without any indication to the user.

━━━━━

And You Tought AI Recruiters Were Bad. Recruiting (and more generally the whole hiring process) is one of the (few) business functions that has seen AI tools being deployed in actual production environments: from AI-powered hiring target research, recruiting outreach, and candidate screening to resume parsing and onboarding (and everything in between). It turns out AI’s (usually) hidden biases are problematic (no one is surprised) – but they also have the wonderful capability of reinforcing themselves.

Researchers already know that LLMs pick up human biases from their training data. New research suggests that LLMs can also develop their own biases from experience – and stereotype job applicants more than humans do.

What We Are Reading

Silicon Valley Has Lost Its Biggest Advantage Much has been written about how phones harm our children. But what if it’s not the phones per se but the lack of play and exploration? @Jane

Judge Approves a $1.5b Anthropic Settlement Over Pirated Books Used to Train the Claude Chatbot A judge approved a $1.5 billion settlement requiring Anthropic to pay thousands of authors ~$3,000 per book after the company used pirated copies of their works to train Claude. A separate ruling found that AI training on copyrighted material itself is legal under fair use. @Mafe

A Philosopher’s One-word Theory to Explain Why the World Feels so Weird The “uni-context” is something like the old vision of a “global village” but stranger – and with stranger effects that seem to bleed into... everything (see: increased status anxiety, comparison, and marketization of everything, etc.). @Jeffrey

Responsible AI Is Becoming a Growth Strategy When everyone has access to similar technology, trust may become the real competitive moat. @Kacee

What If It’s Not the Phones? Much has been written about how phones harm our children. But what if it’s not the phones per se but the lack of play and exploration? @Pascal

Down the Rabbit Hole

🧑‍💻 The Human-in-the-Loop is Tired – This isn’t a think piece about whether AI will replace programmers. It’s not a doomer essay, and it’s not a hype piece. It’s an honest account of what it feels like to be a developer right now, from someone inside it, and some thoughts on what might actually help.

📺 Here’s an interesting experiment: Give the two frontier LLMs (Fable 5 vs. GPT-5.6 Sol) $100 each, ask them to create a music video, and see what happens. Turns out, the biggest fallacy is that both models fail to ever judge their own output.

🤖 First it was AI, now it’s the robots: Fear of humanoid robots spurs human workers to strike at Hyundai auto factory.

🍬 I remember when Googlers debated which free snacks ought to be stocked in the free micro-kitchens (no joke). How times have changed! Thousands of Google workers demand layoff protections amid AI boom in petition to CEO.

🇨🇳 Are we Chinese yet? American AI is locked down and proprietary. It’s losing.

🧑‍🎓 More in the ongoing debate on AI’s impact on learning and education: Using AI could reduce exam scores by a fifth, a study on Chinese students finds.

💦 First chips, then energy, now water: Not enough water for UK’s datacentre plans, trade body says.

⏳ How delightful (and what a blast from the past): The Loading Museum showcases 20+ loading screens from your past life.

⚽ Talking about “delightful”: This data-visualization-powered replay of Soccer World Cup matches is nothing short of a masterpiece.

💿 Nostalgia is hitting hard baby! CD sales growth outpaced vinyl in the first half of 2026.

🪑 Ever wondered which IKEA items deserve the crown as “most complex to assemble”? Wonder no more! Here is the IKEA complexity index.

↗ Dive into the deep end: Access our complete collection of 2,900+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

At an industry event a few weeks ago, I had yet another conversation about agentic commerce – AI agents doing your shopping for you, autonomously, end to end. It’s the hot topic in (e-)commerce right now. And the argument I keep hearing (and I bet you’ve heard it too if you are even remotely connected to e-commerce) goes like this: it removes purchasing friction; friction removal always wins; therefore, it’s inevitable. Case closed – let’s talk valuations and who the big winners will be.

Except we ran this exact play once before. Remember when Alexa was going to become your shopping platform? Amazon put a microphone in a hundred million kitchens, wired it directly into the world’s largest store, and waited for the orders to roll in. Alas, they didn’t. Voice commerce stayed a rounding error, the division reportedly burned billions, and today Alexa mostly sets kitchen timers, switches your lights on, and tells you the weather.

So the friction argument has at least one big, expensive counterexample. Which raises the question my conversation partner and I debated for a while: has it ever actually happened? Has there been a moment in history where people willingly handed the decision itself – not the execution, the decision – to a machine or a system, at scale, because it was more convenient?

Turns out: yes. Constantly. We just don’t notice, because successful delegation is invisible by design.

Start with the biggest one. Index funds are the largest voluntary transfer of decision-making authority in modern economic life. Jack Bogle launched the first index fund at Vanguard in 1976, and Wall Street literally called it “Bogle’s Folly” – who would accept average returns on purpose? Roughly five decades later, passive funds hold more US equity assets than active ones (they crossed over in 2023, per Morningstar), and trillions of dollars sit in vehicles where a formula – not a person, not even the investor – picks every single holding. Target-date funds went further: a formula picks your stocks and adjusts your risk as you age, and you touch nothing for forty years. And auto-enrollment in 401(k)s – supercharged by the Pension Protection Act of 2006 – went furthest of all: people delegated the decision to even have a decision.

Or take navigation. Twenty years ago, knowing the best route across town was a point of pride – cab drivers built careers on it, and every family road trip featured a co-pilot wrestling a paper map. Today Google Maps decides, a few hundred million times a day, and nobody blinks. Uber stacked three more delegations on top: an algorithm picks your driver, your route, and your price, and we went from “never get into a stranger’s car” to tracking the stranger’s arrival on a map in about five years flat.

The list keeps going. Subscribe & Save and autopay quietly automated billions in recurring purchases. Spotify’s algorithm decides what you hear next; Netflix autoplays the next episode before you’ve formed an opinion about the last one. People delegate consequential decisions to systems all the time – happily, at massive scale.

Which makes the failures more interesting, not less. Because there’s a genuinely instructive one: Stitch Fix. The company IPO’d in 2017 on precisely the agentic-commerce thesis – tell us your size and budget, and an algorithm (plus a stylist) picks your clothes. It worked well enough to go public and then shed more than 90% of its value. Delegating clothing turned out to be fundamentally different from delegating index weightings, because what you wear is a statement about who you are. Few people’s identity is wrapped up in the precise set of holdings in their stock portfolio.

Lay the winners next to the losers and a pattern emerges. People don’t resist giving up control. People resist giving up self-expression, and they resist agents whose incentives they can’t read. Delegation wins when three things line up: the decision is a chore rather than a statement about who you are, the outcome is verifiable or the stakes are low, and it’s legible whom the agent actually works for. Vanguard won not just because indexing was mathematically sound but because its structure made the incentive question boring: the funds are owned by the fundholders. There’s no one to betray you. Alexa failed the same test from the other side – everyone could smell that the agent worked for the seller. You didn’t need an economics degree to feel the conflict of interest; you just needed to hesitate for half a second before saying “Alexa, buy batteries” and wonder whose batteries, at what price.

Hence, agentic commerce isn’t a single bet – it’s two very different ones rolled into one. The replenishment layer – detergent, dog food, printer ink, the maybe 60% of your cart that’s pure habit – is a chore with verifiable outcomes, and it likely will get eaten, probably faster than most people expect. The discovery layer – the jacket, the restaurant, the gift for your partner – is not friction to be removed. Choosing is part of the consuming. Strip it out and you haven’t made shopping efficient; you’ve made it not-shopping. The friction argument is half-right, and the half it misses is the expensive half: sometimes what looks like friction is the product.

If you’re building in this space, or advising someone who is (or, like most of the executives Jeffrey and I work with, trying to figure out what your customers will still want a human hand in), here’s my best bet: take your customer’s purchases and sort them into two piles – chores and statements. Be honest about which pile each item actually belongs in, not which pile your strategic plan needs it to be in. Then ask, for every chore you want to automate: can my customer verify the outcome, and can they tell – instantly, structurally, without reading a terms of service – that the agent works for them and not for the merchant? That second question is where this market will be won, because the technology is getting there quickly and the trust architecture mostly isn’t.

Which brings me back to where the conversation should have started. The interesting question about agentic commerce was never “will people give up control?” – history says they will, cheerfully, for the right chores. The question is the same one Bogle answered in 1976: who builds the agent whose incentives are so structurally boring that trusting it requires no leap of faith at all? Amazon likely won’t be it – it owns the store. So who becomes the Vanguard of agentic commerce?

@Pascal

Andrew Ng, one of the leading (actual) experts in AI, just made a good point in the whole token-counting debate: AI tokens are cheap compared to human tokens (i.e., AI output is so much cheaper than your thinking time). Once you adopt this line of thinking, you start treating your (i.e., human) tokens as gold and AI tokens as a commodity. Read the thing – it’s good.

P.S. Here is a fun one – the wonderful folks at EDC recently had me on their podcast; it was such a fun conversation! Watch our conversation on YouTube or your favorite podcast app.

And now, this…

Headlines from the Future

Oh, the Absurdity of It All. I do like my LLM. Actually, I love it, and it’s insanely useful to me and my work. I also believe that there is a lot of potential in the further development and deployment of AI. But SoftBank’s Masayoshi Son going on the record to say that the AI buildout will require $5 trillion per year by 2040 and that AI revenue will hit 20% of global GDP by 2040 is just a bit too much for me…

“Every year $5 trillion, or 800 trillion yen, you might think that’s a lie, but I am confident that’s what it will cost,” Son said ⁠at SoftBank’s annual corporate conference in Tokyo. […] “The business model will be viable because by 2040, if AI revenue makes up 20% of global GDP, spending 800 trillion yen a year is a rounding error,” Son said.

The important bit here is this:

He did not say how he came up with the $5 trillion number or the proportion of global GDP he expects AI will make up.

Yep. Precisely. And just to put this all in perspective: Global GDP today is ~$117 trillion. 20% of that is ~$23T – larger than China’s entire economy ($19.4T) and about three-quarters the size of the US ($30.62T). All of worldwide manufacturing is about 15% of global GDP (and the single largest sector today).

━━━━━

AI Is an Environment, Not a Tool. Here is something to mull over: L. M. Sacasas makes the argument that AI reshapes your perception and judgment from the inside, even when you use it carefully or reject its every suggestion. This leaves you in a position where even carefulness offers no protection – the most deliberate users lose judgment fastest because the medium, not the content, does the reshaping.

But AI is not a tool in this sense, it is an environment which envelops the user and works on us from the inside out while we naively think that we remain unchanged by our use so long as we are using it carefully and intentionally. The care and intentionality is beside the point, and our confidence in such vigilance probably works against us in the long run. […] And this is because AI is an environment not a tool. I can pick up a tool and put it down, but the environment absorbs me into itself.

What We Are Reading

The End of Reading Is Here A world where everyone could read once felt like progress locked in. It’s starting to look more like a temporary experiment. @Jane

Scientists Say You Have a Sixth Sense, and It Could Affect Your Mental Health Interoception is a hidden sense we all have that helps regulate hunger, mood, and anxiety. Some of us have it more developed/trained than others. @Mafe

Terrified of Small Bets Your teams (and your org) aren’t locked in by a lack of imagination or tools. More likely, they’re trapped by systems that require “certainty” before action. @Jeffrey

Transforming Manufacturing at Pfizer: The Hard Part Was Not the Technology One of the better AI case studies I’ve read lately. Pfizer shows that getting the technology working is often the easy part; changing how thousands of people work every day is where the real transformation happens. @Kacee

The Most Famous AI Writing Tic Is Also the Most Mysterious I hate the AI trope of “it’s not X, it’s Y” – thank Shakespeare for that… but really, nobody knows why LLMs love this construct so much. @Pascal

Down the Rabbit Hole

🧑‍⚕️ The tl;dr of this new study: GPT-4o, given the right prompting strategy and note-selection freedom, generates ED “one-liner” clinical summaries that emergency physicians rate as more accurate, complete, and clinically useful than the physicians’ own summaries (and this is with GPT-4o, a model so old that it’s (largely) not even available anymore).

📚 The tides are turning: With Google Search traffic collapsing, publishers are now starting to withold content from Google, as the trade-off equation has flipped: Once unimaginable, publishers are preparing to opt out of Google Search.

💵 It surely is not too surprising that OpenAI will argue you shouldn’t count tokens, but focus on work per dollar… How to manage AI investments in the agentic era.

👷 Poof goes the data center. New York’s ban on data center buildouts has the potential for far-reaching consequences: New York imposes first-in-the-nation statewide freeze on ‘hyperscale’ data centers.

🤨 Is anyone who has spend any time on LinkedIn or X even surprised anymore? LinkedIn and X are flooded with AI spam, browsing data suggests.

😙 Talking about LinkedIn – its seemingly inevitable decline into Facebook’s ugly stepbrother continues: LinkedIn, a mass grave of ghost jobs, is now becoming a dating app.

😯 And taking it a step further, social media seems to be dead now anyway (at least the “participating” part of it): The death of the status update: Why 55% of Americans stopped posting on social media.

💈 It’s not you, it’s the carpet: Striped floors and flickering leds can overload the human mind, leaving some with headaches or nausea.

🍞 The bread paradox: why convenience always wins, and why SaaS isn’t doomed.

🧑‍🏫 The U.S. spent $30 billion to ditch textbooks for laptops and tablets: The result is the first generation less cognitively capable than their parents.

🔋 The energy transition is full steam ahead: In three states in Australia you can now get three hours of free daytime electricity.

😱 Fascinating thought experiment: What would happen if WhatsApp were to disappear? The day WhatsApp goes dark.

🚶 Those who wander are not lost. But we seem to loose the ability to wander (at least – and the consequences are real: The problem with Google’s A.I. overview.

👽 John Carpenter’s master piece “They Live” was right all along: Guerrilla london bus ads mock Kylie Jenner’s Meta glasses campaign.

🕶️ Even Lorde says so: Lorde says Ray-Ban Meta AI glasses are ‘not sexy.’

🤑 This is too funny: “LARP – Revenue is just an agreement between friends” is a website which allows you to generate some good old fake revenue by doing some nvidia-style circular deals…

🎬 Fun one for the weekend: The tech of ‘Terminator 2.’

↗ Dive into the deep end: Access our complete collection of 2,900+ radical links.

What We Do When We’re Not Writing

Hi! I’m Pascal from radical. When we’re not writing this newsletter, we help organizations turn volatility into advantage – without the “innovation theater.” We build your team’s capacity to handle disruption, so you stop reacting and start shaping. If that sounds useful, let’s talk.

The debate if, when, and how the money-burning pits that are the AI frontier labs will be profitable (let alone truly justify their eye-watering valuations) continues to rage. A lot of the debate is coming from either Camp A – the AI hypers, or Camp B – the AI doomers. This week two voices I consider to be squarely in the middle (and hence, more likely to be right) published their respective takes on the subject – both worth your time (if you are interested in the topic): Benedict Evans on “Ways to think about token pricing” and Arvind Narayanan on “Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in.”

P.S. Friend of radical and Vitra’s trendscout Raphael Gielgen posted their annual Work Panorama and a nifty Lexicon of Work – recommended!

And now, this…

Headlines from the Future

The Coming Enshittification of the Exoself. Silicon Valley legend Kevin Kelly’s latest think piece is about our relationship with omnipresent AI assistants. In Kelly’s view – and remember, Kelly was a pioneer of the quantified self movement – we will soon all be running around with AI companions, which will become an extension of ourselves – hence the term “exoself.” In Kelly’s view, we will be making choices between four types of exoself relationships: Twin/Clone, Tutor/Guardian, Counselor/Assistant, and Hero/Friend.

So far, so dystopian (good or bad – you decide). What Kelly seems to miss, though, is the fact that these exoselves will be operated by commercial entities. And these entities have a history of embarking on infamous “enshittification” journeys (case in point: Meta now charges for essential features of its AR glasses). And as such, the relationship we have with our AI might not be something the user chooses, but platforms – not users – will engineer which of the four types you get, tuned to engagement or subscription revenue rather than your interests.

In which case, Kelly’s little admission at the very end of his piece might be the least of our worries (emphasis mine):

This second self will demand a new kind of relationship, one we haven’t had before — and its immense benefits will arrive bundled with immense problems. Every ailment that afflicts our born self will likely show up in the exoself too, plus novel ones we haven’t seen yet. Learning to use an exoself wisely will be one of the major lessons of a life lived this way. It will take years before society works out anything like best practices — we’re still working on those for social media. There will be multiple models and personality types to choose from. And there will be heart-wrenching stories of people losing their exoself — the worst case being simply that the platform went out of business.

━━━━━

AI Has Opinions Now – And Becomes a Superforecaster. Scott Alexander, on the Astral Codex Ten blog, makes a strong argument for AI getting better and better at forecasting – and now regularly competing with (and sometimes outperforming) human superforecasters. The idea is simple and logical: Given that AIs have been trained on vast amounts of data, with the right harness, they should be able to make predictions with calculated confidence levels.

Ironically, we’ve spent the last couple of years making AI refuse to have opinions, and now we’re building a certification layer specifically so it can – which also means that the question isn’t whether AI forecasters are accurate; it’s whether we’ve thought at all about who decides which opinions get to count as “calibrated” versus “biased.”

But the AI prediction markets of the future should be far superior to the human markets of the present. The biggest barrier to current markets is liquidity - there isn’t enough money riding on most questions to convince top superforecasters to drop their jobs at Google or the NSA in order to think hard about them and bet on them. But AI forecasters bring the cost of forecasting labor down to near-zero, so we can have hundreds of different AI agents betting on each question and be pretty sure its error has been driven down to the theoretical minimum. This, in turn, means we can vastly expand the number of questions, including (finally!) allowing randos to submit their own questions (probably with AI assistance in proposing un-rules-lawyerable resolution criteria). […] This, then, is my prediction for the AI superforecaster future: for basic questions, your off-the-shelf AI chatbot will be able to offer opinionated probabilities superior to those of any human. For more controversial or bias-laden questions, a new era of prediction markets will smooth over differences in brand and model and efficiently aggregate all AIs’ opinions.

━━━━━

AI → Expensive RAM → Expensive Smartphones → Big Problems. The world’s insatiable appetite for AI is driving up the cost of RAM (computer memory) – and not by a little bit, but by a lot. You might have seen that your beloved Apple MacBook Air just got a whole bunch more expensive. Now, you might shrug this off as a minor inconvenience – at the end of the day, how often do you really buy a new laptop? But here is where this really bites: The world runs on cheap (Android-based) smartphones. If you have ever been in a developing country, you will have, inevitably, seen even some of the poorest people with a smartphone. And for them, the smartphone is not a toy to keep them entertained; it’s an important tool in their lives. And exactly these cheap smartphones are being squeezed out of existence. And that’s a real problem…

After ravaging the PC market for both DIY builders and OEM buyers, wrecking the game console ecosystem, and savaging the enterprise server landscape, the global memory shortage is officially claiming its next victim: the budget smartphone. According to new market analysis from tech research firm Omdia, the smartphone market, particularly the entry-level market, is on the brink of a massive contraction. Omdia projects that global shipments of smartphones priced below $400 will plummet by over 22% this year, dragging the entire global smartphone market down by 12% year-over-year.

What We Are Reading

The Corner of Hollywood That’s Most Susceptible to AI AI isn’t just threatening Hollywood jobs; it’s quietly changing what we think good storytelling looks like. @Jane

Why Are Birthrates Down? 2 New Studies Point to Phones. An unlikely culprit of decreased births and fertility: Technology, specifically the iPhone. @Mafe

Ten Reasons Why We Won’t See Productivity Improvements from GenAI Even as the tools progress, real-world productivity gains in complex work environments are likely to remain hard to achieve, hard to measure, and hard to transfer between contexts. @Jeffrey

How Software Engineers Really Feel About AI Coding This echoes conversations I’ve been having with builders... AI coding isn’t just making developers faster; it’s forcing many to rethink the craft that became part of their identity. @Kacee

Something Is Very Wrong with Modern Longevity Science Debunking a lot of the longevity “science” – tl;dr: skip the mystique, do the known boring things, and stop treating lifespan as a personal-optimization project when it’s mostly income, education, and social conditions. @Pascal

Down the Rabbit Hole

👨‍💼 Could have been me: CEO says he’ll fire any employee who sends him more AI slop. 😆

🧮 Admittedly a very nerdy read, but the argument is fascinating: AI could destroy mathematics and barely touch it – by decoupling raw problem-solving and genuine mathematical understanding (something the author calls “the overhang”): The fall of the theorem economy.

⚡ Stop wasting energy on your AC, feed the data center instead! Virginia county asks all employees, including schools, to conserve power due to AI-driven electricity price hikes — state’s 400-plus data centers steadily increasing demand, grid expansion, and pricing.

🏭 Speaking of which – the conversation about data centers’ impact on climate continues to rage: Data centers emitting more CO2 than thought: study

🙍‍♂️ Not good (and yet, is anyone surprised?): AI software that generates ‘rage bait’ developed by Germany’s far-right AfD.

🕵️‍♂️ You are Mark Zuckerberg. First you fire thousands of people, then you invest billions of dollars, and then you find out that “AI agent tech is progressing slower than expected.”

🧑‍🌾 The luddites are back – and this time they teach us how to live a live with less technology: Inside the Luddite festival harnessing Gen Z’s rage against Big Tech

💰 At the frontier of AI (in this case Anthropic), the AI already costs more than the engineer – by a factor of 2.3x.

🤦 Seriously LOL: Execs confused and horrified by the huge AI bills after thinking they could replace workers for free.

📰 The Economist used GPT-4 to grade its own 25-year prediction record and lands on “pretty good, actually.” Now the question becomes: Who audits the auditor?

💆‍♀️ Want to date in Korea? Better be a chip worker: South Korea’s hottest new bachelors are chip workers.

🧑‍🏫 AI’s next victim? Online courses teaching coding skills: “I just launched my third course, Whimsical Animations, and so far, it’s on track to sell roughly ⅓ as many copies as a typical course launch.”

💾 For anyone who’s into old-timey computers – here is an extensive list of movies starring your favorite machines.

↗ Dive into the deep end: Access our complete collection of 2,800+ radical links.

Should We Work Together?

Hi! I’m Pascal from radical. This newsletter is our labor of love. When we’re not writing, we run radical, a firm that helps organizations navigate the future without the “innovation theater.” Most leaders want to seize new opportunities, but they hate endless strategy decks that go nowhere. At radical, we don’t run “projects”; we build your organization’s internal capacity to handle disruption and change. Our goal is to make you future-proof so you can stop reacting to the world and start shaping it. If you’re interested, let’s jump on a call to see if we’re a good fit. Click here to speak with us.

I’m recently back from facilitating an HR/People Leaders summit that was exactly the kind of program I’d have been delighted to join as an attendee myself: curated, conversational, and questioning. The agenda left ample room for the attendees to engage as active, co-author/participants and to follow curiosity and conversational energy wherever they might go.

As you’d expect, most of the conversations swirled around the impact of agentic AI capabilities (as well as the anticipated impact of those capabilities) on the future of work and workforce, but the specific inquiries were fascinatingly varied. Human org design vs. agentic org design; workforce planning as a continuous process; scaling behavior change with AI adoption; the decline of interdependence; and more. I’d have loved to drop in on each of these explorations – or at least send my agent to take notes.

Suffice to say that I went to facilitate and left with plenty to think about and some new questions to ponder. One that has really stuck with me: What happens to organizations that no longer bring on (or bring up) juniors?

While the narrative of an imminent AI-driven job apocalypse has stalled a bit lately (in no small part due to a reevaluation of messaging by some of the big AI firms), leaving the bigger future-of-work picture more uncertain than ever, we have plenty of clarity on the market for juniors circa now. In a word, it’s f*cked.

A much discussed study the NY Fed released last month argues that the surge in unemployment of young college graduates is most significantly due (so far) to the impact of remote work, which has “weakened incentives to hire young workers by impeding on-the-job training.” We also know that the collapse of traditionally relied upon signals in the job market have made getting hired as a young worker particularly hellish. And none of this is even taking into account the persistent likelihood that many orgs, while not cutting headcount, are actively looking to AI tools to increase the productivity of more experienced/senior workers and NOT need to source, hire, and train juniors.

All of which is to say that the picture for that segment of the workforce is more clear and more immediately concerning than that of the workforce at large, and (this probably sample-skewed Ramp/Revelio report notwithstanding), I don’t think we’re likely to Jevons paradox our way out of this situation in the near future.

So maybe the kids aren’t alright. Now what about everyone else?

Would it not be a shrewd move to ride the productivity gains of the firm’s most experienced and most highly compensated workers? And aren’t these exactly the workers who are likely to get the most value out of the AI tools because they know the business (context/problems/etc) the best?

Well, yes… in the short run and within a certain limit. But the rub is that it may be devilishly hard to know exactly where that limit is and whether you’ll be able to see it approaching before it’s too late.

An organization with no up-and-coming next wave of talent is likely an organization designed for near-term maximization at the expense of long-term adaptability. It’s likely to be more fragile, less resilient, and less capable of creative reinvention, and it will only become more dependent on a core group of skilled operators as it becomes less fluent in communicating and passing on (and updating) institutional process knowledge.

Organizations that don’t maintain an active culture of training, mentoring, and learning may find themselves without critical capacities that benefit not only the junior employees but the organization as a whole – especially when the market shifts and the winners are determined by their ability to learn new and different things.

After years of hearing plenty of C-level and senior leaders bemoan the difficulty of managing multi-generational and especially multi-generational hybrid workforces, I’m sure there are some who would happily opt out of that challenge. They’d do well to remember that the flipside of that challenge is the opportunity to ensure their firm’s future relevance.

@Jeffrey