In the spring of 2010, a team working under Craig Venter took a bacterial genome they had written on a computer, assembled it out of bottles of chemicals, and slipped it into an emptied cell. The cell woke up and started to divide. Every descendant carried a watermark the team had encoded into the DNA itself — the names of the scientists who built it, an email address, and a line borrowed from Richard Feynman: “what I cannot build, I cannot understand.”

That organism was a new form of life. Made out of the same building blocks life has always used, but arranged into an order that has never existed.

Chances are good you didn’t notice. (That’s less about you and more about news editors — biology discoveries don’t typically make headline news.) President Obama asked his bioethics commission to take a look and they produced a report; journalists went back to arguing about whether Facebook was ruining democracy.

Venter died this past April at 79, and most of the obituaries led with the human genome race — the sailboat, the swagger, the standoff with Francis Collins. The synthetic cell landed near the bottom, in the paragraph before the survivors.

Hold onto that footnote for the next few minutes as you read on.

A few nights ago I was climbing a steep staircase in Lisbon when a text came in from a researcher friend at Stanford. I told my family to go ahead to dinner and said I'd catch up. A team in California had designed viruses that exist nowhere in nature, built them, and watched them work.

It’s the first proof that a generative model can compose an entire functional genome. I’m not talking about a gene, or a protein — but a complete instruction set for a self-replicating biological entity, written end to end by AI and viable when synthesized in a lab.

If you’ve been reading me for a while, or if you’ve seen me speak about the frontiers of AI and biology anytime in the past two years, this research and the people behind it will already be familiar. The work comes out of Brian Hie’s lab at Stanford and the Arc Institute — the group that built Evo and then Evo 2, a genome language model trained on roughly 9 trillion nucleotides of DNA spanning every domain of life. I’ve described it before as genBio — a ChatGPT for biology, which undersells it slightly. But it doesn’t just predict the next word. It predicts the next base pair.

Hie’s team aimed Evo at a single target: ΦX174, a phage that infects E. coli. Phages are viruses that eat bacteria, and doctors already use them as a kind of benevolent predator — mostly as a last resort, against severe drug-resistant infections that antibiotics can no longer touch.

They handed the model an opening fragment and asked it to write the rest, start to finish, in a single pass, without a human in the loop. Thousands of potential genomes came back. The team narrowed the field, chemically built about 300, and tested them in containment. Remarkably, 16 came out alive.

Several killed E. coli faster than the natural phage they were modeled on. Some were so different from anything on record they’d count as new species. One had assembled itself using a structural part borrowed from a distant relative — a solution nobody specified and nobody expected. And a mixture of the AI phages wiped out three strains of E. coli that had already learned to resist the original, which was interesting because a mixture of natural phages couldn’t do that.

It’s hard not to see this as synthetic biology’s “Attention Is All You Need” moment. That 2017 paper introduced the transformer architecture in AI, and it effectively catapulted AI from a sci-fi fantasy into scalable, commercial use. What Hie’s team has published in the journal Science is no less important. Whole-genome generative design is now an engineering problem rather than an open question, and engineering problems, I think, will eventually get solved on a schedule — rather than waiting and hoping for a research breakthrough. In fact, the authors say as much. They describe this as a foundation for designing larger and more complex genomes. (ΦX174 was chosen because it is small.)

Now that we can both edit and write life… what might we design?

Your head might immediately go to novel viruses — the dangerous kind. To their credit, the Stanford and Arc researchers excluded viruses that infect humans, animals, plants and fungi from their training data. They worked only with harmless laboratory strains, consulting biosafety experts throughout and red-teaming the model’s outputs. Writing in the same issue of Science, biosafety researchers from Johns Hopkins said that this team focused on safety more deliberately than most developers of powerful biological AI.

That’s both a compliment and a concern. Voluntary care by conscientious scientists is, at the moment, our entire safety architecture.

Downstream, things get more complicated. For example, once you design something you need to get it biologically printed, a process known as synthesis. Order genetic material and the supplier compares your sequence against databases of known pathogens and controlled agents. That works as long as you’re looking for a known pathogen — what, though, of a novel sequence that doesn’t look like anything on earth?

The way we regulate genBio in the US is a bit of a mess. There is no single American law covering the convergence of genetic engineering, synthetic biology and AI. What exists instead is agency guidance, and guidance is not authority. There is a 2024 screening framework that basically applies only to federally funded work. An executive order in May 2025 gave the White House science office 90 days to revise it; as of last month the Congressional Research Service couldn’t establish whether that had been done. Five bills are now circulating in Congress, each with a different theory of the problem.

I wrote about the lack of alignment and foresight in The Genesis Machine, my book on the convergence of AI and biology. Over the past four years I've been in rooms with members of Congress, the Defense and State Departments, and the White House Office of Science and Technology Policy, showing them scenarios for genBio and its potential impact on everything from national security to the future of our food supply. To date, the most coherent standards in existence belong to an industry consortium, and joining is optional.

Some future scenarios sound pretty great: imagine designing organisms that eat forever chemicals, crops that make their own fertilizer, phage cocktails designed overnight for just one patient’s specific infection. The catastrophic futures include self-replicating things we cannot recall, sequences so novel that we lose the ability to prove an outbreak was engineered — or prove it wasn’t. Attribution is what deterrence rests on, and nothing has to be released for its loss to destabilize the world.

Venter’s team encoded Feynman’s line into a living genome because they believed it. But Feynman's actual line was what I cannot create, I do not understand. Venter's team encoded a slightly garbled version into the genome… so the first watermark on synthetic life contained a human-made transcription error.

Sixteen years on, a model that understands nothing wrote a genome that works — including a structural trick its designers never anticipated and had to go looking for after the fact. Which means Feynman’s premise has been inverted.

We can now build what we do not understand.

I don’t think that’s a reason to stop. But I do think it's a reason to decide, deliberately and soon, what we want written. Because the writing has already started, and the only question left is how seriously we’re all contemplating the newest drafts of life.

In 2017, while I was drafting the future scenarios that became The Big Nine, the best generative music system in the world could produce about 10 seconds of plausible piano, and that was Google’s Magenta which made tinny loops that sounded like a music box with a head injury. WaveNet was only a year old. Nobody I knew in machine learning would have told you, on the record, that a computer would write a convincing rock song in your lifetime.

So I wrote this:

The original members of the Rolling Stones died years ago, but thanks to replicating algorithms, they’re still making new music. That sensation you felt after hearing the first 30 seconds of “Paint It, Black” for the first time—the melancholy guitar melody, followed by eight loud bangs on the drum and a repetitive hook that culminates in Mick Jagger singing “I see a red door and I want it painted black...”—was a singular moment of excitement and satisfaction. It didn’t seem possible you might get to feel that way again with a new Stones song, and yet their latest track is just as loud, hard and fulfilling.

Nine years later, an AI-generated country act called Breaking Rust topped a Billboard sales chart. An AI persona named Xania Monet became the first synthetic act to land on a radio airplay chart and then signed a multimillion-dollar record deal. As of today, LA-based rapper Fenix Flexin is denying that his hit new single was AI-generated after extensive allegations were made.

But what of the Rolling Stones?

Well, drummer Charlie Watts died in 2021… and appears on a Rolling Stones album, Foreign Tongues, released three weeks ago. And, John Lennon, who died in 1980, sings lead on a new Beatles single that won a Grammy.

My Scorecard

🧭 5 · ⏱️ 4 · 🎲 5 · 📐 5 · 🌊 2 = 21/25, 🟢 Called It

Analysis

Directionally, I was correct. I was mostly on time, I named a specific band and I was definitely not part of the consensus.

Charlie Watts is on Foreign Tongues because he recorded those takes before he died. One song on that album, “In the Stars,” is a deepfake of four living human people in a real room, whose faces have been swapped with de-aged band members. The Beatles track exists because a neural network pulled Lennon’s actual voice off a degraded 1979 cassette he recorded in his apartment. (Ringo Starr has said flatly that they would never fake John’s vocals.)

Both cases involved AI, but in replication rather than pure generation. The distinction sounds academic until you notice that it has become the entire legal architecture of the industry: California now requires explicit estate consent before an AI replica of a deceased performer can be used commercially. Notice, the rule is about “estate” consent, which means that a decision that my have run counter to someone’s personal wishes can still be made posthumously. That’s why there’s a new Prince album out now, which I have to think he’d have hated.

I was right about the technology and hilariously wrong about the band. I literally have no idea how or why, given their history… but Mick Jagger (82), Keith Richards (82) and Ronnie Wood (79) are still very much alive and still performing on stage.

Methodology

Each scored 1–5. Total out of 25.

Share Convergence by Amy Webb

Two weeks ago I sat in a board meeting where an 8-figure AI commitment was approved in under 20 minutes. The CFO asked one question, about the payment schedule. No one really asked for details about what the money was for.

Afterward, the chief strategy officer walked me to the elevator and said the thing people often say to me in hallways, away from their peers: “I know, I know… but we had to do something.”

I’ve heard a version of that sentence many, many, many(!) times.

For most of the past decade, AI decisions have been made using two guiding principles: fear and FOMO, the fear of missing out. Fear resulted in paralysis masquerading as “good governance” — lots of meetings, with no decision made. Meanwhile, FOMO was stupidly fast action, without any deep analysis or long-term strategy contemplated in advance.

Lately, I’m observing a third guiding principle, and it’s definitely worse.

I call it AGITA: AGI-Triggered Anxiety.

It seems like every partner in the Valley has it. From my perspective, it’s the only thing that explains a $300 million seed round for a company with nothing more than a landing page and a manifesto.

AGITA isn’t entirely irrational

The easy version of this argument is that everyone has lost their minds when it comes to AI, so the spending is stupid. That’s not what I’m saying. Some of this capital will be spectacularly well placed.

But, but, but… the problem is structural. Fear and FOMO were at least tethered to something observable. Fear pointed at a specific downside, like a regulator action, a potential data breach, or the probability of a lawsuit. FOMO was about a competitor’s launch, or a customer who might defect, or a deal from OpenAI or Anthropic that seemed so good it felt irresponsible to turn it down. In any of those cases, right or wrong, you’d see the impact of your (in)decisions.

AGITA, however, is connected to a timeline no one can defend, against a definition that changes depending on which company or lab you ask.

Unrelenting heartburn

As of now, a thesis for AGI — investment, product, process, etc. — isn’t clearly defined, so decisions can’t really be evaluated. Without a product, let alone a profit, there’s no true failure condition. For f’s sake, we don’t even have an agreed-upon true definition of AGI nor an alignment on specific benchmarks proving definitively it exists.

When a decision is made without strategic foresight — a structural forces check, a value migration map, value network diligence, partnership roadmaps, trade-off ledgers, future capability/ use case blueprints, capital alignment frameworks, scenarios and the like — the outcome tends not to go as planned. Which is why, 18-24 months later, there’s hell to pay in the boardroom. The defense is usually something like “we’re building capacity” or “we’re learning.” You can always claim to have learned something. But this often means the spending can’t be evaluated easily, which means it can’t be stopped, which means the only available direction is more.

The problem with AGITA is that there’s no Tums for frontier models.

The AI boom has resulted in a boomerang effect, with a clear beneficiary. The vendors and labs causing your AGITA have no incentive to help you specify the future you’re hedging against, because the moment it gets specified, it becomes a thing you can measure them against.

This has happened before

In 1982, Japan’s Ministry of International Trade and Industry launched the Fifth Generation Computer Systems project, which was a national, decade-long push to build machines that could reason. The response in the West was panic. This correlated with a sudden auto market dominated by fuel-efficient Japanese cars and a growing fear of Japanese business culture, so a number of big government initiatives were launched.

As the promise of AI seemed more imminent, the US launched the Microelectronics and Computer Technology Corporation (MCC) in Austin as the US’s first major for-profit tech research consortium. The UK and Europe launched similar defensive efforts, Alvey Programme and ESPRIT. Enormous amounts of capital moved, quickly, driven almost entirely by the fear of being structurally left behind in a technology (AI) whose end state nobody could describe.

The Fifth Generation project ended in 1992 having produced interesting research and basically none of the machines it promised. The counter-programs it triggered mostly dissolved. What almost nobody did during that decade was write down, in advance, what would have to be true for the panic to have been justified. They didn’t bother with strategic foresight.

Meds for this moment

I’m not going to tell you to spend less. I don’t know your business, and under soul-crushing deep uncertainty, aggressive allocation is sometimes the right move.

But I would want you to do a simple strategic foresight exercise first. (Obviously, this alone isn’t sufficient, but it’s better than allowing AGITA to dictate your investment.)

Answer these questions in detail:

1. What must be true in 12 months/ 3 years/ 5 years/ 10 years for this to work?

2. What would have to happen for us to conclude it isn’t working?

3. When and how will we evaluate?

You’ll may need to challenge your cherished beliefs, which may be uncomfortable. Better one tough day, than a really bad year.

Scenario: Paris, 1960, at a massive university of 180,000 students. An annual awards ceremony is underway, and every prize is for science or technical achievement — the humanities have all but disappeared. Our hero, Michel Dufrénoy, wins a prize for Latin verse… but the entire crowd laughs at him. Literacy is universal in this future Paris, but nobody reads anything except technical manuals.

When Michel goes looking for a copy of Hugo, or Balzac, or Lamartine, and discovers they've been erased from society’s memory entirely, his family is humiliated. He drifts from job to job, loses the one person he loves, and ends up collapsed in a cemetery during a snowstorm, crying out "O Paris!"

It's a bleak ending. This future scenario has remarkable inventions, like giant calculating machines described as huge pianos operated by a keyboard, connected to something like a fax machine, which banks use to coordinate their most advanced operations. It’s a world of total computation and communication infrastructure, where humans still have to do all the thinking, and simply choose not to.

The culture kills itself by deciding literature isn’t worth anyone's time once the machines can handle everything measurable. Anyone refusing to conform — to venture beyond the bounds of technology — is an outcast.

In this story, the future doesn’t punish the slow. It punishes the independent thinker.

This is from a novel written 160 years ago by futurist Jules Verne. It was such an alarming idea that his editor refused to publish it. It sat in a drawer for more than 130 years.

I’ve been thinking about what might be Verne’s most important story lately — not because I think we’re heading for Michel’s Paris exactly, but because Verne’s dystopia was never really about the future. It was about groupthink — the machinery by which an opinion becomes a fact simply by being repeated enough times, by enough confident people, in the same room.

There’s a lot of AI groupthink now. Someone says something offhand on a Lex Friedman or a16z podcast…it's a good line, so it gets clipped. The clip gets quoted without the three hours of context around it. By the third repost it's not a soundbite anymore, it's a consensus. That consensus becomes an assumption baked into every strategy deck in Silicon Valley, a talking point among executives, and eventually a catchphrase repeated on Capitol Hill. Somewhere in that chain, the part where anyone verified it with actual logic or actual data just sort of disappears.

This happens in private, too. There are closed-door, invitation only forums where tech leaders sit with senators, DoD officials, and others defining what AI risk and opportunity mean for American safety and competitiveness. I know, because I’ve been in many of those meetings. In the span of a year, I’ve heard that AGI is a decade away, five years away, and 12 months away, sometimes by the same person. I heard the 12 month timeline while I was at a dinner at the Aspen Ideas Festival a few weeks ago. When I pressed for more details, the exec sniffed, “that’s just general knowledge now.”

This is how a new norm in AI gets established — not through data, not through argument, just through repetition at scale. Call it a societal consensus algorithm: nobody wrote it, nobody voted on it, but everyone's executing it anyway.

Here’s what the consensus algorithm looks like right now: you are more likely today to be unemployed with a computer science degree (7%) than a philosophy degree (5.1%). Ten years ago this would have sounded absurd. Here’s what happened:

• Mark Cuban said studying philosophy could soon be worth more than a degree in computer science.

• Reed Hastings said young people should study philosophy.

• Goldman Sachs’ CIO Marco Argenti said that he told his daughter to study philosophy along with engineering.

In the span of a decade, computer science went from the safest bet in higher education to a major with departmental cutbacks. Rather than poaching entire academic robotics departments, the hottest tech companies are now scooping up philosophers.

Of course, none of this is really about which major you pick. It’s about how fast an entire culture can flip its convictions without anyone stopping to ask why and about who’s setting the terms of the flip.

Right now in America, it isn’t musicians, or athletes, or novelists who hold that kind of sway. It’s tech executives. A small number of people, mostly saying the same small number of things to each other, and the rest of us treating it as evidence rather than opinion.

A society allowing itself to be governed by consensus algorithm worries me, and it should worry you, too. Not which side of the philosophy-vs-code debate wins, but the fact that so few people in a room are willing to say something no one else in the room has already said.

In Verne’s book, Michel wasn’t destroyed because he couldn’t keep up. He was destroyed because he was the last person in Paris still willing to ask what any of it was for.

Somebody in the room has to be the one still asking what any of it is for. It might as well be you.

Consider a modern bike’s cockpit. It’s both insanely high-tech and a bit of a mess. Power-meter pedals count the watts in each leg. A speed sensor records k/ mph. A heart-rate strap reveals how hard you’re working. A bike computer fastened to the bars, pulls all those data together and layers in GPS, a map, the climbs ahead, and on a training day, the workout you’re supposed to be suffering through enjoying. It’s a marvel!

It’s also, right now, a mess. Different sensors from different manufacturers, all supposed to talk to each other. I can tell you from experience they frequently don’t. Connections drop, syncs fail. My favorite: just as your group rolls out, the pedals suddenly won’t record your watts because of a finicky battery.

That friction — that’s part of a tell. And if you consider what’s happening outside of cycling — in physical AI, edge computing, biotech — you can spot the early signs of a convergence.

Canyon is, I think, betting on a physical AI bike. Its new Predict system builds all of that scattered technology into the bike itself: edge AI, radar, cameras, a 360-degree sensor array, and a display set cleanly into the handlebar. So that’s hardware. It also fuses the sensor feeds with your own riding dynamics, speed and steering and stability, into a live model of the road, then predicts what’s about to happen. It scores the risk of a corner, a car, or a bad patch of surface before you’ve registered it, and warns you through lights, haptics, and the screen. That’s on-bike compute. Pair it with the company’s Stingr helmet — which has a drop-down visor display — and in a real emergency the bike can act on its own, dropping your seatpost to lower your center of gravity and steady you before a crash it saw coming and you didn’t. It runs on the bike, not the cloud, which cuts the lag and keeps the data local.

For now, Canyon’s Predict is just a prototype. The peloton tomorrow will undoubtedly have silly-looking aero helmets on the first team time trial stage, but they won’t have a slick heads up display.

But, but, but… let’s think about what that bike represents. It senses the living world, interprets it, and responds in real time. That’s the exact shape of what I’ve been calling Living Intelligence, arriving in one of the oldest machines we own, and it’s the same pattern already moving toward glasses, wristbands, helmets, and cars. Physical AI doesn’t show up as a robot. It shows up as familiar objects that suddenly perceive.

This is a clear sign of a potentially catastrophic storm for everyone in the industry.

Cycling today is a hodgepodge, and every piece belongs to a different owner. Garmin makes the computer. Shimano, SRAM, and Favero make the power meters. Whoop or Apple or Garmin track your body. Strava and TrainingPeaks hold your rides. I could keep listing companies and components. There are dozens, each with a slice, none of them owning the whole picture.

Integration collapses that. Once one brand combines the sensors, the AI, the safety system, and your data into a single stack that just works every time, the fragmented ecosystem rebundles into a few walled gardens, and the value migrates with it. It stops accruing to whoever makes the best power meter and starts accruing to whoever owns the integrated model of you on the road. Convenience for the rider on one side of the ledger. Lock-in, and a situational model of your body and your movements living inside one company, on the other.

So you’re not into cycling… why should you care? Well, what's happening on that handlebar — a swarm of dumb sensors resolving into one system that perceives, predicts, and acts — is the same thing about to happen to the objects you do care about. Your glasses, your watch, your doorbell, your kitchen. Therefore, anyone in telecommunications, in media, in health, in OEMs (see what I’m doing here?) — you ought to pay close attention. The bicycle is simply early, because it's a small, contained machine with a rider who wants every possible edge. Watch what consolidates here and you're watching a rehearsal for the rest of your physical life

Since we’re under an excessive heat warning here in the northeast, I’ll be riding on my trainer indoors, watching the TDF’s first stage, happy for Campy’s vlog and sad that Wout can’t compete this year. But in the long term, the more consequential race is the one to own the handlebar. Whoever wins it won’t just be selling bikes.

Share

I thought about her recently, because a thief in San Francisco just made a cleaner getaway than Dillinger ever managed, and no one heard a thing.

Someone robbed a yoga studio in San Francisco, walked out with an armful of merchandise, and climbed into a driverless Waymo idling at the curb. When the thief was inside, presumably with the seatbelt on, the car pulled away. There was no driver, no chase, no witnesses. And… six months later, no suspect.

What we know to be true: a Waymo is basically a rolling surveillance system. The cars are riddled with cameras and sensors. To summon one, you need an account registered to an app on your phone, and you need a valid credit card payment on file. So this should have been an easy solve for the local police.

Instead, by the time the police got their warrant, the interior video was gone, the exterior faces had already been blurred for privacy, and the company reserved the right to push back on law enforcement to protect its riders.

The most surveilled getaway in the city’s history produced no one to arrest.

I suppose this is a win for privacy as we’re all increasingly surveilled. I’m still wondering whether this was a new form of physical AI penetration testing – someone testing the limits of security in an AI-mediated real world.

That said, we ought to think about how privacy might shield bad actors in the future. My grandmother was one of the witnesses in 1934. Today the witness is a corporation, and the record of what happened lives on its servers, under its retention schedule, released through its lawyers. Dillinger used a human shield. This thief used a privacy policy.

We assumed a city full of cameras would end anonymity. It may do the opposite: anonymity for whoever the platform decides to protect, and only for as long as the footage survives.

We're one step closer to generating human eggs without a womb, a body, or a donor. Emphasis here is on step. A biotech company just announced a milestone that’s big (if true)… but also somewhat oversold.

Conception, a biotech startup based in the US, just announced that it made human egg cells from blood. Take a donor’s blood, reprogram the cells into stem cells, coax them inside a lab-grown “mini-ovary,” and out come early human eggs. If that sounds like the opening of a story where reproduction detaches from the body entirely, it is. That’s why I’m flagging it.

But read the claim carefully, because the gap between what happened and what was announced is the whole lesson in reading a signal. Conception didn’t actually make eggs. It made primary oocytes — the earliest egg-cell precursors, nowhere near mature enough to fertilize. The news arrived by company blog and a CEO’s social post, not peer review. There’s no published data yet on efficiency, genetic normality, or safety. Clinical use is likely years and many animal studies away.

(If the miracle of “just a few X of blood” has you thinking Theranos, you’re not the only one.)

To be sure, there is a massive body of work already in progress in induced pluripotent stem cells (iPSC). Basically: researchers take a fully committed adult cell — a skin cell, a blood cell, a cell that long ago decided what it was going to be — flip a small set of genes, and reverse it into a blank slate that can become anything.

Most of this is happening in Japan, where I lived and worked for many years. I still visit regularly to meet with researchers working in biotech, robotics and AI. Mitinori Saitou’s team at Kyoto University has been pushing human stem cells toward the egg lineage inside lab-built ovaries since 2018. Recently in 2023, Katsuhiko Hayashi, who is a pioneer in generative reproduction, took skin cells from a male mouse, reprogrammed them into stem cells, got the cells to shedding their Y chromosome and to double the X, and grew them into functional eggs. Fertilized with sperm and carried by surrogates, those eggs produced healthy pups with two biological fathers. The success rate was punishing — only 7 live pups out of 630 embryos — but the team said that doing something similar in humans was likely on a decade away.

Immature cells in a dish are noise, if you’re reading one headline. They’re a signal, if you’re watching the field. What’s on our horizon: a world where an egg can come from anyone’s cells: two fathers, a sixty-year-old, a single person, a nearly unlimited supply.

When the embargo came in 1973 and prices quadrupled, nearly every major company was blindsided... except for Shell. Shell had already lived through it, on paper, and so it moved faster and more decisively than any of its rivals.

That is what a scenario is, at its core. Not an exact prediction, and not a guess dressed up in probabilities. A scenario is a disciplined, evidence-based story about a future that hasn’t happened yet, built well enough that the people who read it change how they think and what they do. The point was never to be right about a single outcome. The point was to make the unthinkable thinkable early enough to act on it.

Wack didn’t invent scenarios of nothing, and neither did I. When I wrote The Signals Are Talking, I traced foresight back through a very long lineage of people who each figured out a new way of seeing.

I’ll skip ahead to modern(ish) day and H.G. Wells, who did deliberately what no one had done before: he took the faint signals of his own moment (like, electricity becoming a general purpose technology, the machinery of modern war taking shape), asked “what if?” He used that reporting to write stories vivid enough to make readers feel a future that wasn’t yet here. I call this the Imagining Era of foresight.

After WWII, Olaf Helmer and Nicholas Rescher at RAND turned foresight into something more systematic. They built probabilistic models and made the practice rigorous. But they stopped at the edge of recommendation. Their job, as they understood it, was to show you what might happen, not to help you do anything about it. I call this the Calculating Era of foresight.

Then came Wack, and after him Peter Schwartz, who carried the method out of Shell and into the wider world, most famously in The Art of the Long View. This was I call the Exploring Era, and its insight was that the real value of a scenario isn’t the scenario itself... it’s what happens inside the mind of the person reading it. A good scenario doesn’t hand you an answer. It asks you to challenge your cherished beliefs and to updated your assumptions. That approach has governed the field for the better part of fifty years.

Now, here is the uncomfortable part, the one my own field doesn’t much like to say out loud. Exploratory scenarios are brilliant at changing minds but terrible at changing behavior. I know, because I’ve sat in the rooms where people leave with their assumptions productively shaken… and then nothing changes. Those scenarios goes into a binder, the binder goes onto a shelf, and nothing happens. Everyone may feel something, but no one does anything about it.

That gap is the problem I care about most right now, and it’s the reason this section exists on my Substack. The current era of foresight must do what the last one couldn’t: convert imagination into action you that can be measured. A scenario that expands your thinking and leaves you with nothing to do on Monday morning has done only half its job. You need to map that thinking to strategy: as in, what to accelerate, what to pause, what to build, and how you’ll know whether it worked.

So this is where I’ll do that, in the open. In this section I’ll build scenarios — when for our purposes I’m calling forecasts — that you can rehearse, and then I’ll go further than the tradition usually allows. I’ll stake claims specific enough that you and I can both check them later. Some of it will be wrong. All of it will be built to act on.

Wack didn’t just imagine the oil shock. He changed what Shell did about it, before it arrived. That was always the whole point.

Let’s rehearse the futures.

Go backward in time and you can watch a whole industry die and regenerate. Horse-drawn carriages gave way to railroads. Railroads gave way to the automobile. The personal automobile gave way to Uber. And right now, Uber is giving way to Waymo. Each wave of innovation didn’t incrementally improve upon the last, it quite literally created something net new.

There’s a name for this. In the 1940s, the economist Joseph Schumpeter called it “creative destruction,” and his idea was pretty simple: capitalism is a perpetual storm. It doesn’t drift toward equilibrium. It churns, destroying old industries and establishing new ones in the same motion. On one side of the ledger sits creation: new products, new technologies, new business models, new markets, new organizations. On the other sits destruction: incumbent firms, old jobs, outdated ways of making things, dead ways of selling them. The storm isn’t a flaw in the system as much as it is the system.

If this sounds familiar, it's probably because you've read Clayton Christensen. The late Harvard Business School professor described something adjacent, which he called "disruptive innovation”: the way a smaller, scrappier competitor slips in from below and topples an established leader. The most famous example is digital cameras causing the downfall of Kodak.

But Christensen and Schumpeter were basically studying the same weather from different distances. Christensen zoomed in: here is exactly how one competitor takes you down. Schumpeter zoomed out: here is how market forces create destruction that simply happens to you, whether or not anyone means it to happen.

Both were describing the uncomfortable truth. The storm doesn’t care about you. It isn’t personal, and it isn’t waiting for your next board meeting, or your next injection of capital, or your next policy meeting, or your product’s roadmap.

Which means you have to care about the storm before it arrives.

The thing is, if you can see the storm coming, then the destruction, and the creation, become a choice. You get to decide whether to build the new thing or defend the old one. You get agency. But only if you see it in time.

So I built a storm tracker.

Think about how weather actually works. Temperature, humidity, barometric pressure, wind speed: those are the readings. They’re useful, and you need them. But no meteorologist looks at the temperature alone and calls it a forecast. Like… nobody is going to evacuate a city because the barometer reads 29.8 inHg. (To further the point: is that number good? Bad? Numbers out of context don’t really help.)

Trends are the readings. A convergence is the storm system, the thing that emerges when all of those conditions start interacting and produce something none of them could have produced alone.

In my field — strategic foresight — we know what to do with a trend. One appears, we track it, we fold it into scenarios, we stress-test it. For us, trends are raw material. But a trend in the hands of someone who isn't running that strategic foresight process is something else: a single reading mistaken for a forecast. It tells you the temperature and lets you believe you understand the weather. That's the trap — not the trend itself, but the false confidence it hands a leader who tracks it and stops there. Which is why, if you're the one making the call rather than the one building the scenarios, the more useful thing to track isn't the trend but the convergence. You don't react to a convergence. You position ahead of it.

A trend tells you what’s changing. A convergence tells you what’s about to become inevitable, before it looks inevitable to anyone else. The window to act opens earlier, and the cost of missing it runs far higher, because by the time a convergence is obvious, the people who saw it early have already taken the ground.

Which is why a convergence is something you put to actual use. You hold it up against whatever you’re building and decide whether to accelerate, to pause, or to reframe the thing entirely.

Here is the working definition I created:

A convergence is what happens when multiple trends, forces, uncertainties, and catalysts intersect and interact to create a combined impact that is greater—and often different in kind—than the sum of their individual effects.

In practice, a real convergence has four characteristics.

1. First, it’s system-level. Convergences don’t just pile trends on top of each other.

   They operate across different domains, making them hard to see unless you are doing a specific type of cross-domain analysis.

2. Second, it creates new realities suddenly. What seemed inconceivable becomes inevitable, not gradually but suddenly, even though all the pieces were visible beforehand.

3. Third, it redistributes power and value. Convergences influence who wins, what’s valuable, and where there is leverage. They rewrite competitive dynamics, not just within industries but across them.

4. Fourth, it’s hard to reverse. Because multiple systems reinforce each other, convergences establish new market realities faster than traditional changes, making early detection increasingly critical.

Here’s a convergence I’ve been tracking. Last year I identified a convergence I call Living Intelligence: the collision of artificial intelligence, advanced sensors, and biotechnology into systems that can sense, interpret, and respond to the living world in real time. The data moves in both directions: the system reads the living world, and it acts back on it. On its own, each of those three technologies is a trend you could put on a roadmap. Together, they are a storm system, and the implications reach into nearly every industry I can name.

Here's the good news buried in all of this: convergences are only invisible until you start looking for them. The storm doesn't announce itself. It assembles from what’s already sitting in plain sight. You don't need to predict it. You need to start tracking it — early, and on purpose.

Each year my team at FTSG publishes the Convergence Outlook, a deep analysis of the convergences we're tracking most closely — what's driving them, who they reshape, and how fast. You can read highlights or download the whole Outlook for free here.

I get asked this question all the time, and my answer rarely satisfies people. If you have the right data and enough of it, and you’ve built a solid model, you can absolutely predict outcomes. These days, with ever-powerful computer systems, you can ingest a vast amount of data, run tons of simulations, and get probabilities on the other end.

The questions I get hired to answer sound impossible until you see the data behind them.

• Which US companies will see the biggest productivity gains from physical AI over the next decade, and why?

• What’s likely to make your current business model obsolete by 2031?

• Which policy choices will decide whether US AI, semiconductor, and biotechnology stay globally competitive?

• Which industries are about to become far more — or far less — insurable as climate, cyber, and geopolitical risk compound?

Every one of those questions is answerable with a good methodology, data and a model. Not a hunch.

I can do that. But given the volatility and rate of change, you’ll need to get in a habit of updating your priors. Which is to say: one-time predictions are useless without continual monitoring.

Convergence is where I do my continual monitoring in public. The weak signals, the system collisions, and the forecasts where technology, capital, geopolitics, climate, science, and human behavior collide and rewrite who holds power and what holds value.

So this is where I keep them. It’s more provisional than my books and more personal than our reports at FTSG. What you’ll find here is what I’d say to you over a meal, but definitely not in a polished deliverable.

Some of what I flag here will turn out to be noise. I write forecasts and will need to update them as needed. Occasionally (ok, often) I get irritated at business and government leaders. And, given that I keep one toe in Hollywood, I sometimes take side quests from reality into sci-fi (but you’ll know that when it’s happening).

Everything is free. If the work is valuable to you and you want to support it, just let your colleagues know. This work won’t sit behind a paywall.

Subscribe, and I’ll send you the future while it’s still early enough to act on.

Subscribe now