Maybe it’s right. City-owned grocery stores are exactly the kind of idea that makes economists reach for the fire extinguisher: low margins, brutal logistics, political incentives, and a long history of pilots that turn into cautionary tales.

The internet had decided what kind of story this was: socialism meets grocery margins. A municipal boondoggle with checkout lanes.

But when a consensus forms this fast, it just brings out the contrarian in me: if the obvious version of this idea is bad, is there a non-obvious version hiding inside it?

I should disclose: I have zero qualifications here. I’m a Palo Alto tech guy who teaches entrepreneurship and shops at Trader Joe’s.

But this is where my founder-brain wanders into rooms where it has not earned a chair. Show up in unfamiliar territory, ask the dumb questions, pattern-match from elsewhere. The whole game is bringing beginner’s eyes and a different toolkit.

So: if you handed me Mamdani’s $70M and said “make this work” — what would I do with it? I am not running for anything. This is just how a founder reads the news.

1 · The first dumb question

What problem are we actually solving?

Mamdani says “make groceries cheaper.” But cheaper for whom, by how much, compared to what alternative? You could hand $70M of subsidy to Walmart and call that cheap. That’s not interesting.

Sharper question: when does a low-income East Harlem family feel grocery prices most painfully? Probably not steady-state Tuesday shopping. The pain comes in the shocks — bird flu spikes egg prices, a shift gets cut, rent jumps, kid gets sick. The actual job isn’t “be slightly cheaper than Aldi every week.” The job is be the predictable backstop when things go sideways.

That reframe changes everything. Different goal, different design, different definition of success.

And — this is the part that surprised me — nobody in the consensus seems to have actually asked this question. Critics argued whether the plan would lower prices. Defenders argued it would. Neither asked what the plan was for. This is the most common mistake in startup pitches and apparently in policy debates too.

2 · What the patterns say

Where do sub-scale operators actually win? Three cases:

FairPrice, Singapore. One store in Toa Payoh in 1973. Took 25 years to build the warehousing arm. Today: ~50% market share and a real brake on grocery prices for the whole country. I shop there whenever I’m back in Singapore — last visit, my wife texted me to grab Tanakan tablets, 30% off. Compounding institutions start small and unimpressive. Five stores at once is the wrong move. One store with a path to scale is the right one.

Aldi. Doesn’t beat Safeway by being a smaller Safeway. Wins by carrying 1,400 SKUs instead of 30,000. Different format, different cost structure. Lesson for Mamdani: don’t run a tiny supermarket. Pick a format Safeway structurally can’t.

99 Ranch & the 13,000 NYC bodegas. Both prove sub-scale grocery thrives when it picks a different game. 99 Ranch wins on category density in Asian SKUs. Bodegas win on proximity, hours, prepared food. They falsify “scale is destiny.” The question is always which sub-scale format you’re inventing.

What do these cases show? Sub-scale wins only by changing the game.

3 · The Walmart gambit

Here’s where it gets fun.

The hardest part of grocery is supply chain. Five stores can’t match Aldi’s wholesale prices. NTUC built its supply chain over 25 years. Mamdani doesn’t have 25 years.

So: what do we have that they don’t, and what do they have that we don’t?

The city has: mission, capital, real estate, political legitimacy with low-income communities, and the ability to run transfer payments through retail. It also knows who needs help better than anyone else — SNAP rolls, Medicaid enrollment, school meal sign-ups, NYCHA tenant data. Walmart has to guess at this. The city already has the answers. What it lacks is supply chain scale.

Walmart has: the most efficient grocery supply chain in North America. It lacks urban retail access in NYC — has tried for years and been blocked at every turn.

The impolite move is: rent Walmart’s supply chain.

NYC owns the stores, sets the rules on labor and pricing, owns the customer relationship. Walmart provides wholesale procurement and distribution at cost-plus. This isn’t outsourcing operations to Walmart — it’s using their supply chain as infrastructure, the way the military uses Boeing for planes without becoming an airline.

Walmart gets revenue from a market they couldn’t otherwise access. NYC gets supply chain economics they couldn’t otherwise build. Neither party builds what the other already has. This is the cleanest version of partnership: complementary asymmetric strengths.

This is the wedge every startup looks for: asymmetric competition. NYC supermarkets have to maximize profit. Mamdani doesn’t. He’s optimizing for affordability and stability, not quarterly earnings. That’s not a weakness — it’s a different objective function entirely, and it’s exactly what creates space for disruption. You can’t out-Key-Food Key Food at being Key Food. But if you’re playing a different game, you don’t have to.

One bonus: only Mamdani can pull this off. A moderate Democrat partnering with Walmart gets eaten alive by the left. A democratic socialist mayor doing it looks like ruthless pragmatism — Nixon goes to China, except the China is Bentonville, Arkansas. Use the political asymmetry while you have it.

4 · The dream plan

So if you handed me the $70M and locked me in a room:

1. One store first. In the neighborhood with the sharpest mismatch between need and existing retail. Not borough balance — diagnostic clarity. Earn the next four.

2. 150 SKUs, mostly staples. Not a tiny Safeway. A tightly-curated essentials operation.

3. Double SNAP at the register. Government can run transfer payments through checkout in a way private grocers structurally can’t. Every SNAP dollar becomes two. The store becomes the highest-value use of SNAP in the city.

4. Walmart wholesale partnership. Rent the supply chain. Mission belongs to the city; logistics belongs to whoever does it best.

5. Price buffering on staples. Eggs hit $12 during bird flu? Sell them at $4. Eat the loss, amortize across normal years. Only public balance sheets can do this.

6. Service integration. SNAP enrollment, school meal sign-up, WIC at the same location. The store as partial delivery mechanism for services the city already runs.

7. Transparency as a feature. Publish prices, costs, subsidies, contracts. Trust and pressure on incumbents are both downstream of openness.

And best of all? I wouldn’t spend $30 million on the first store. I’d spend $2 million (that’s 0.0016% of Mamdani’s $122 billion budget — a rounding error in a rounding error). Use one of the vacant storefronts the city already owns. Strip it, stock it, open it. A cheap and stable grocery needs to be frugal itself — you can’t run an affordability operation out of a hero project. Leverage the free publicity from the Twitter trolls and the WSJ editorial board; they’ve already made this the most-watched grocery store in America. Be humble. Learn from the experiment. The other $68 million stays in the bank until the first store proves something worth scaling.

Five stores in five years was the political plan. This is the founder plan: one store, distinctive format, rented supply chain, public-only mechanisms, open books. Different category of bet entirely.

5 · Why I bothered

I have no idea if any of this would actually work. Genuinely don’t. Maybe the consensus is right and the whole thing is a boondoggle. Maybe I’m missing twelve things obvious to anyone who’s spent a day in city procurement.

That’s not the point.

The point is the questions. What’s the actual job? What can I do that incumbents can’t? Who’s the unlikely partner I should rent capability from? What’s the smallest version that would teach me whether the bet is real? None of these required grocery expertise. They required willingness to probe without expertise and pattern-match from elsewhere.

The next time you find yourself handed a problem everyone dismisses as impossible, take on this smug probing persona. Ask dumb questions. Think about hidden goals, ecosystems, and structures that nobody else is thinking about. Sometimes the consensus is right. Sometimes it isn’t. The only way to know is to start digging.

p.s. if you laughed at this tech bro attempt at humor, or wondered what startup sessions with my NUS students are like, here’s our chance for a chat: https://appt.link/time-with-chao-VPJUZ3nZ/web-conference Wife is out of town and I could use some company - don’t be shy!

It’s a Tuesday after dinner. I’m on the couch, eight minutes into For All Mankind and wondering how they managed to age Joel Kinnaman yet again, and my wife comes and sidles up to me. At least that’s what I think she’s doing — until she stops, points at a spot near where my elbow rests, and says, “What is that? Who did that!”

I look. There’s a smudge. Then there are two spots. Then I’m seeing marks I’ve never noticed in three years of owning the sofa — a constellation of small disasters — a blood stain here, a chocolate speck there, a few unknown black and grey dots that could be anything. The sofa hasn’t changed. My evening has.

“Are you going to bring a microscope next?” I say. “Make me see all the microbes?”

She does not laugh.

She’s right. The marks were there. I’d chosen not to see them, without knowing I was choosing. She made them legible, and now I can’t unsee them, and there’s nothing I want to do about it at 9pm on a Tuesday. The information arrived without the agency to act on it.

Tomorrow morning those same marks won’t be anxiety. They’ll be a chore. I’ll get out the upholstery cleaner and take twenty minutes and they’ll be gone. Same information. Different moment. Different activation energy. The marks transform from doom to task. The only thing that changed was whether I could do something about them when I saw them.

Hold on to that. It’s going to come back.

I went to a No Kings march recently. It felt good. I am not the only one who feels this. Walking with neighbors, recognizing faces, the relief of being among people tuned to the same frequency.

And then nothing closed. No threshold was met, no vote triggered, no specific person whose specific decision was specifically affected. My body traversed along El Camino Real. My anxiety had been seven thousand miles away in the Strait of Hormuz. Those two zoom levels never met. I came home, traveled the next week, wrote my newsletter, watched more sci-fi. The anxiety hovers. It doesn’t destroy.

I think this is the experience most people I know actually have — not paralysis, not depression, just the optimism getting more expensive every month, in a way nobody talks about because it sounds melodramatic when you’re objectively fine. So let me say it: I am fine, and the optimism is getting dearer.

I’m writing this to capture how I actually work. I notice loops. I try to close them. The feedback rarely lands the way I predicted. Figuring out what to do about that is most of what living looks like for me right now.

The reason I can’t just let it go is my daughter.

They’re twenty, studying civil engineering at UW. I’ve spent thirty years watching startups fold, watching VCs betray me, watching a war get sold on weapons that weren’t there. That builds something in you. Call it insulation. Call it calluses. The war on the news hovers for me. It might land for them.

I want them to be able to see the world clearly without it hollowing them out. Not less information. Not “don’t worry, sweetheart.” Somewhere to put what they see.

I do not currently know how to give them that. That’s why I’m writing this.

A few weeks before the march I’d been reading about Jack Clark — the Anthropic co-founder — and his pitch that you have to make AI dangers legible before you can build safety infrastructure for them. You can’t regulate what no one can see.

I agreed with the logic and noticed I felt worse after reading it. So I started poking at why.

I landed on a Jevons Paradox of Anxiety, half joke and half real.

Jevons noticed that using less coal to produce more power didn’t reduce coal consumption. It increased it. Efficient coal is cheap coal, and cheap coal finds new uses faster than the efficiency gains shrink the old ones.

The same shape applies to danger made legible. Make a worry easier to name, easier to share, easier to feel, and you don’t reduce the total amount of worrying. You increase it. People will always find more to worry about.

And cheap worry about one thing doesn’t stay contained. It makes adjacent worries cheaper too — just as steam engines made their way from factories to railways and steamships.

The implication is uncomfortable. Information without anywhere to go isn’t neutral. It piles up. It makes you carry more without giving you anywhere to put it down.

People who can’t put things down eventually go numb. Not because they stopped caring — because caring with no exit is unsustainable, and the body protects itself.

This is what’s happening to the people around me who used to march and now don’t. Not cynicism. Not selfishness. Just unclosed loops piling up, until starting a new one feels harder than closing the laptop. Staying in an open loop and calling it caring. It feels like engagement. It’s actually how engagement dies.

Once I had that frame I started seeing it everywhere. The clearest example is the ozone layer.

If you were alive in the late eighties you remember it. The hole was growing. NASA published images. We were all going to get skin cancer. We could feel the anxiety in the culture — parents slathering sunscreen on kids, news segments about UV indices, a vague dread every time someone mentioned aerosol cans. And the loop was completely open. We personally could not fix the ozone layer. We didn’t even know which of our products contained CFCs. The danger was maximally legible and our agency was zero. Classic doom loop.

The Montreal Protocol did not start with “save the ozone layer.” It started with “reduce these five specific CFCs by fifty percent.” The ambition was global. The first action was specific, falsifiable, and time-bound. It worked. Every country in the world ratified it. 99% of the targeted chemicals are gone. The loop closed. Legibility was pointed at the response, not just the problem.

That distinction — making the response legible instead of making the danger more visible — is the move. Same mechanics. Same endless demand. But you change what’s getting cheaper. Not worry. Agency.

Looking at my own civic behavior, the picture is unflattering. I send the occasional email to my congressman with no specific ask, no deadline, no measurable threshold. I attend the occasional march. I donate. I post the occasional thing. None of those actions have edges. None of them close. I’m doing the civic equivalent of standing in front of the sofa pointing at the marks.

So the obvious move, once you have this frame, is to build the thing. A platform where vague anger gets translated into a specific, falsifiable, time-bound civic ask. Where your commitment only activates if enough others also commit. Where the response is as legible as the problem.

I sketched this. I got excited. I told a friend — the same one whose judgment I trust on how things feel to use — and she said she didn’t want it. Group coordination is slow. The feedback loops are long. By the time enough people commit to anything, the moment has moved.

She was right, and the rightness pointed somewhere. If the problem is loop architecture, the fastest loop is the smallest one — one person, one hunch, one closed bet. The civic version is the same fractal at a bigger scale, and I still want to build it eventually. But the personal version is where the muscle gets built, and people who know how to close personal loops are the only people who can close civic ones. So I started there.

And because the personal version is the cheapest to test, I whipped up a small Claude skill: help one person close one loop on one thing they’re stuck on. Articulate the hunch. Pre-commit to what counts as a yes or a no. Set a deadline. Look honestly at the result.

I showed it to the same friend. She tried it. Here’s what she sent back, in order:

i feel like this feels too serious still like, it feels like “work” i already feel so tired from work so this feels like even more work

Then later:

LOL and also this skill doesn’t appeal to me it triggers my inner rebellion in fact it’s shocking how much i hate it kids these days call it pathological demand avoidance

And then the line that did the real damage:

it might work if i made the skill but i find it very hard to install one that will boss over me. the skill itself is values neutral.

This is what closing a loop actually feels like from the inside. Things never go the way I predict. I built the thing I thought she’d want, she sent back something I didn’t see coming, and the gap between those two is where the real information lives. If I’d predicted her response correctly there would have been no loop to close — I’d have just been talking to myself with extra steps.

Every piece of the tool was right and the whole thing was wrong. There’s a difference between describing how good thinking works and bossing someone through the steps. The first is useful. The second is what my friend hated, correctly.

So here is where I actually am. Less triumphant than I expected. More energized than I expected, and the why is the point.

I still think the Jevons Paradox of Anxiety is real — that making dangers more legible without making responses more legible leaves people worse off, however well-intentioned. I still think the exit is not less information but more agency, with edges that actually close.

My own experience in the last few months has been proving the thesis back to me. The march gave me no actionable signal and drained me. The product idea — even when it’s failing — has been doing the opposite. The first feedback was a zoom-level correction: don’t start with civic coordination, start with personal loops, because the muscle is the same and the stakes are cheaper. That narrowed the idea and gave me something to build. Then my friend’s texts narrowed it again.

Each piece of feedback arrived at a zoom level I could act on, with a specific thing to change and a next move to make. That’s a closed loop. The anxiety about the world is still there, hovering in the background. But the energy I was losing to the hover is now going somewhere. The difference between the march and the product isn’t motivation. It’s whether the loop closes.

The sofa, by the way, did get cleaned. My wife got my daughter to do it. I got away scot free — which, if you’ve been paying attention, is exactly the failure mode I’ve been writing about.

I started this essay not knowing what I wanted to give my daughter. I think I know now. Not a framework. Not protection. A world where the loops are the right size, the edges are visible, and the response is as legible as the problem. It has to feel like something they chose. Not something I designed for them.

The marks were always there. The question was never how to see them.

p.s. here are 2 claude skills if you’d like to try

• Hunch: Tries to converse with you to draw out small experiments on yourself. This is the one that didn’t work - trying to show failures publicly!

• Experiment-miner: If you have lots of Claude conversations like I do, this one suggests tiny experiments from those conversations. This feels like a thread I can pull!

Some geeks recall fondly their times spent in bookshops. I spent my teenage years in copy shops.

In a harshly fluorescent-lit room in Singapore, machines zipping their own fluorescent lights, I copied the photocopy of the photocopy of Beneath Apple DOS. That was my initiation to a secret guild.

Tomorrow, Apple turns 50. I've been writing a somewhat-daily newsletter about Apple for roughly the same number of days.

Seven months ago, I wrote a letter to myself committing to build an Apple Library — a database of stories extracted from every major book about the company. Tagged by dates, locations, characters, products. Maybe, someday, the foundation for a walking tour of Palo Alto, where every boring office front hides decades of drama underneath.

Then life happened. Singapore twice for family. New York for an AI conference. Four months, zero newsletters.

But walking along Manhattan's High Line, I kept pointing my camera at buildings and asking ChatGPT to tell me something interesting. In Singapore, walking with my mother through Rifle Range Nature Park, I was thinking about presence — what happens when you stop scrolling past information and start being with it.

Three things congealed: a near-daily cadence to build momentum. The newsletter as experiment in a medium that doesn't have a category yet — not "reading news," not "writing with AI," something in between. And if I ever build the walking tour, it shouldn't be a guided checklist — it should be an open world, like Breath of the Wild, where the AI responds to curiosity instead of steering you through checkpoints.

In January, I published eight newsletters in fifteen days.

Here's what the process looks like.

I sit down with Claude Code and say: what happened, and what does our library say about it? The AI searches a SQLite database — 29 books extracted, over 6,000 stories, geocoded to more than 17,000 locations. It finds the historical rhyme. I push back. I sharpen. I redirect.

"ROKR is overdone by us," I said when Claude suggested the Motorola parallel for the fifth time. "Local on-device inference is the Apple long-term position." That push turned a stale partnership story into a piece about the nine-year Neural Engine infrastructure play.

"No one else has this twin leverage" — a phrase I landed on mid-session that became a named analytical pattern and shaped every supply chain story after.

"What can Siri do for me that no other app can?" — a reframe, born from interrupting Claude mid-pitch, that became the thesis of the most recent issue: privacy was the lock; Siri is the only key.

The machine does the research. I do the whimsy. The newsletter is the residue of that collision.

Things broke beautifully.

Steve Wozniak's memoir: the parser read three HTML files as three chapters. 52 stories. We fixed the splitter. 185 stories. Same words, different cuts, 3.5x the yield.

A Peanuts-style comic strip, requested on a whim, was so charming it became a permanent feature of every issue after.

Image pasting to Substack — generate, copy, switch to Chrome, pray — failed for the dozenth time. I snapped: "Can you think ground up how we can improve this?" 666 lines of Python reverse-engineering Substack's undocumented API later: one command, no browser, no prayer.

And writing "Chekhov's Gun" about Apple's privacy architecture, I searched the database for Apple's role in creating ARM — one of the most consequential partnerships in Apple's history. Nothing. Apple was basically responsible for ARM's existence, and the story wasn't in any of the books we'd extracted. The database taught me what it didn't know, and pointed me to which book to extract next. (I'm reading The Everything Blueprint now. Just got it extracted.)

In March, I went back to Japan. Myoko Kogen for skiing. Zenkōji temple in Nagano. The Setsugekka train through snow country.

I kept publishing. "Eight Slots, Two Dies" from a ryokan. "Party of One" between ski runs. "Chekhov's Gun" on a travel day. Different from September's burst — from "The Boring Revolution" and the bingo card I called a failed experiment. This was a practice.

At Zenkōji, I used Claude as a voice guide. The walking tour prototype, live, in the field.

It misidentified the main hall. Kept steering me to the okaidan tunnel like a checklist. No sense of time, no way to share a photo without exiting the conversation. Not great.

But the failure was the gift. GPS-anchored content, not vision-based guessing. Camera-first, like Snapchat. A single ambient companion. I couldn't have learned this from planning.

The project keeps spilling beyond Apple. Yesterday I published "Refractions" — comparing William Shockley with a modern-day tech bro, tracing how a genius who couldn't keep people spawned the entire Silicon Valley ecosystem. That essay grew from the Apple Library research — Malcolm Harris's Palo Alto, the Fairchild genealogy, walking tour conversations about what stories hide under suburban surfaces. Today I cycled to 391 San Antonio Road in Mountain View to visit the Shockley Semiconductor sculpture.

The library isn't just producing newsletters. It's sending me places.

I'm writing this on an M4 MacBook Air. Maybe a twentieth the weight of my first Apple ][. Next to it, my iPad Pro and my just-feels-right iPhone Air. I switch between them without thinking. Between sessions I'm chatting with Claude about the Strait of Hormuz, or planning a scenic train ride, or transcribing a podcast, or arguing about whether Kurzweil will win his long bet.

Fifty years ago, two guys in a garage were hand-soldering circuit boards for a computer club. The word "startup" didn't exist. "Personal computer" was an oxymoron.

What I'm doing now — conversing with an AI about the history of the company that built the machine I'm conversing on, pulling from 29 books to connect today's news to forty-year-old decisions, generating comic strips, publishing with a single command — would have been incomprehensible to them. It would have been incomprehensible to the teenage me, sweating in that copy shop.

And yet it feels continuous. The same curiosity. The same secret guild. The same click when a database query surfaces a rhyme I didn't expect, or a thesis snaps into focus, or a prototype fails in exactly the right way.

Apple's story is still not done. The last decade — as I discovered reading Apple in China — is just as fascinating as any other. Not as much personal drama as Jobs's reinvention, but so much more massive. Billions instead of millions.

My story with Apple isn't done either. Thirty-some more books to extract. A walking tour to build. A practice that compounds.

Seven months ago I wrote: "The next five weeks aren't about building the complete library. They're about discovering what this library wants to become."

Still discovering.

Happy 50th, Apple. Here's to the next fifty.

chao

(an earlier version of this was also cross-posted on )

At an AI summit in New Delhi this February, Prime Minister Modi gathered the assembled tech leaders for a show of unity. Everyone joined hands and raised them high.

Everyone except two.

Sam Altman and Dario Amodei managed an awkward elbow bump. The photo went viral. A decade of personal wounds, captured in a single cringe.

The Wall Street Journal just published a 13-minute investigation of the feud behind that elbow bump — a panda costume at a wedding, a proposal via stuffed animals coming to life, an executive accused of plotting against his CEO who then denied making the accusation, and peer reviews so vicious the author offered to withdraw them. The internet took sides immediately.

But I kept thinking about someone who isn’t in the article at all.

The Visionary

In 2012, Elon Musk visited a small London AI lab called DeepMind. Over lunch at the SpaceX canteen, its founder Demis Hassabis told him to add AI to his list of existential threats — alongside asteroid strikes and world wars. Musk sat in silence for nearly a minute.

Then he started worrying. Obsessively.

He invested $5 million in DeepMind to monitor it. He raised the topic constantly in late-night conversations with his friend Larry Page. At Musk’s 2013 birthday party, the two got into a passionate argument. Musk said AI could make our species extinct. Page pushed back — why would it matter if machines surpassed humans? He called Musk a “speciesist.”

When Google bought DeepMind in January 2014, Musk was terrified. In a 2016 email revealed in court filings, he wrote that DeepMind was “causing him extreme mental stress” and worried about its “one mind to rule the world philosophy.”

He tried the institutional route. One-on-one meeting with President Obama, 2015. Explained the risk. Suggested regulation. “Obama got it,” Musk later said. “But I realized that it was not going to rise to the level of something that he would do anything about.”

So he turned to Sam Altman. A small dinner in Palo Alto. They decided to co-found a nonprofit AI lab. “We wanted something like a Linux version of AI,” Musk said, “not controlled by any one person or corporation.”

They called it OpenAI. The diagnosis was prescient. The prescription was noble. And what happened next was pure Musk.

The Dispersal

Dario Amodei joined OpenAI in 2016 and quickly rose to VP of Research, leading the teams that built GPT-2 and GPT-3. But in 2017, Musk — OpenAI’s principal funder — demanded a spreadsheet listing every employee and their contributions. Then used it to fire people, one by one. Up to 20% of the 60-person staff were let go. Dario was horrified. One of those fired would later co-found Anthropic with him.

It’s a detail buried in the WSJ piece. But it means Musk’s management style didn’t just catalyze OpenAI’s creation. It catalyzed the trauma that drove its best researchers out the door to start a competitor.

By 2018, Musk wanted OpenAI folded into Tesla. The board refused. He left, withheld funding, and eventually founded xAI to compete with the organization he’d built.

Right time. Right place. Extremely prescient. And then — the dispersal.

I know something about the blast radius. Almost immediately after Musk brought the kitchen sink through Twitter’s front door, he shut down nearly all third-party API access. What was once free now cost $42,000 a month. My company, Thread Reader App — serving millions of users — was effectively suspended overnight. We seriously considered selling for peanuts.

We persevered, trusting he’d relent. He did — a $5,000/month plan, a month later. But I’ll never forget how casually one man’s impulse nearly killed what we’d spent years building.

The Pattern

Musk didn’t just catalyze one company. He catalyzed all of them.

He invested in DeepMind — Hassabis is now CEO of Google DeepMind, powering Gemini. He co-founded OpenAI — he’s suing it; it’s worth $300 billion without him. His management drove out the researchers who founded Anthropic — building Claude. The three frontier AI companies that matter most in the world, and Musk is a central character in all three origin stories.

He then founded xAI and recruited 11 brilliant co-founders. As of last week, all 11 have left.

There’s a name for this pattern. It’s just not the one anyone’s been using.

Mountain View, 1956

William Shockley didn’t invent the transistor. That was John Bardeen and Walter Brattain, working under him at Bell Labs. But Shockley believed the credit should be his. He tried to get the patent in his name alone. Bell Labs refused. He pushed Bardeen and Brattain aside and built a different version in secret.

He shared the 1956 Nobel Prize with both of them. By then, the relationships were destroyed.

That same year, Shockley moved to Mountain View — near his aging mother in Palo Alto — and founded Shockley Semiconductor Laboratory. He had a talent that doesn’t get enough attention: spotting and seducing elite young researchers. Robert Noyce. Gordon Moore. Eugene Kleiner. Jean Hoerni. Names that would define the next half-century of technology. “It was like picking up the phone and talking to God,” Noyce said later of Shockley’s recruiting call.

Then Shockley did what Shockley did. Paranoid management. Erratic decisions. Credit-hoarding. Within a year, eight of his best researchers — whom he bitterly dubbed “the traitorous eight“ — walked out and founded Fairchild Semiconductor.

From Fairchild came Intel. From Intel came the microprocessor. The entire ecosystem we call Silicon Valley grew from the wreckage of Shockley’s failed company.

His direct contribution? A failed lab and a disputed Nobel. His indirect contribution? The geography of the future.

The Mirror

Brilliant talent spotter who can’t retain talent? ✅

Founded the company that defined an industry’s geography — then watched it fail? ✅

A diaspora that built the actual future while the founder’s company collapsed? ✅

Increasing drift toward discredited social theories as wealth insulated him from feedback? ⏳

The comparisons people reach for with Musk — Jobs, Edison, Ford, Tony Stark — are flattering mirrors. Shockley is the mirror nobody wants to hold up. Because it’s not flattering. It’s predictive.

And it predicts from the beginning. Musk’s drama doesn’t start with xAI or Twitter or even OpenAI. It starts at PayPal, where the board voted him out as CEO while he was literally on a flight to his honeymoon. They replaced him and renamed his company. He was 29.

Twenty-two years later, Musk bought Twitter and renamed it X. A $44 billion grudge.

The Seduction

Shockley’s most underrated talent was recruiting. He could walk into a room, identify the most brilliant person, and convince them to follow him to a risky startup in a town nobody associated with technology. The pitch wasn’t “come work for me.” It was “come work on the most important problem in the world.”

Musk has the identical gift. He pulled Sutskever into OpenAI. He got Karpathy to leave Stanford for Tesla Autopilot. The xAI team included Jimmy Ba, co-author of the most-cited paper in AI history.

The seduction works every time. The talent arrives. The vision is real. And then — the paranoia, the erratic restructuring, the inability to trust — drives them away.

PayPal under Thiel sold to eBay for $1.5 billion. OpenAI without Musk built ChatGPT. Anthropic, born from Musk’s panic, built Claude. xAI under Musk lost all 11 co-founders in under three years.

The seduction and the alienation come from the same source.

The Darker Parallel

Shockley’s later years were consumed by eugenics — theories about racial intelligence, arguments that people with lower IQs should be sterilized. He used his Nobel prestige to lend scientific credibility to deeply racist ideas.

The same trait that lets you see the future of technology — absolute certainty in your own judgment — is the trait that, turned toward social questions, produces race science. “I see what others can’t” is the foundation of both visionary engineering and crackpot ideology.

Physics pushes back. The rocket explodes. The transistor doesn’t amplify. Reality corrects you. But when you turn that confidence toward questions about who should reproduce or which civilizations are fittest, there’s no rocket to explode. The feedback loop is social disapproval — which a sufficiently wealthy person can simply ignore.

The Escape Hatch

But history rhymes — it doesn’t repeat. The mirror doesn’t quite reflect. It refracts.

SpaceX is the refraction. Gwynne Shotwell — twenty-one executives reporting to her, four to Musk — is the institutional structure that channeled the chaos into reusable rockets. The physics provided the epistemic humility Musk won’t generate internally. The rocket either lands or explodes. No amount of tweeting changes the outcome.

If Musk’s legacy ends up being SpaceX, that’s more than enough. Ford’s legacy survived his antisemitism because the Model T was so consequential.

But the SpaceX-xAI merger puts that escape hatch at risk. Folding a failing AI company and a toxic social media platform into your crown jewel is how you contaminate the one thing that works.

Cerebral Valley

I fed this Shockley-Musk parallel to Grok — Musk’s own AI. It rated the descriptive fit at 80-85%. It called the eugenics-to-pronatalism trajectory “the same intellectual arc.”

Shockley moved to Mountain View in 1956. His company failed. His people scattered. They founded Fairchild, Intel, AMD. We called the valley that grew around them Silicon Valley.

Musk co-founded OpenAI in San Francisco. His relationships fractured. The people scattered — Anthropic a few blocks away, xAI co-founders dispersing to every competitor. We call the neighborhood that grew around them Cerebral Valley.

The Refraction

I’ve been caught in Musk’s blast radius. I’ve survived it. So I’m not writing a hit piece. I’m writing this because sometimes a rhyme is a warning you can still heed.

Shockley’s story ended badly. Estranged from everyone. The transistor bears his name in the textbooks, but the industry he accidentally created never looked back. No second act.

Musk is 54, not 76. He has SpaceX — a genuine masterpiece. And SpaceX works because Musk did something there he never did anywhere else: he hired Gwynne Shotwell, gave her operational control, and let the institution be bigger than himself. The rockets land because someone built a structure that could channel the genius without being destroyed by it.

Steve Jobs learned the same lesson. Fired at 30, a decade in the wilderness, then he came back and built the most valuable company in the world — not by becoming a different person, but by building structures that channeled the same intensity into something durable. Tim Cook was his Shotwell.

Musk already knows this works. He’s living proof — at SpaceX. The question is whether he can learn from his own example.

I genuinely hope it causes a reflection.

I teach entrepreneurship to Singaporean students in Silicon Valley. I tell them about flywheels and distribution and learning loops. But what I’m really trying to show them is something harder to put on a slide: how to operate in a system that looks like chaos and is actually adaptation.

No one planned this ecosystem. Musk didn’t plan to create Anthropic’s founding conditions. Shockley didn’t plan to create Silicon Valley. The genius scattered the seeds. The village grew the orchard.

But it only works if the village is free — free enough for a CEO to tell the Pentagon no, for a judge to call the government Orwellian, for an AI to tell the truth about its own creator. I grew up in a place that got almost everything right except this one thing. Turns out it’s the thing that matters most when the future is genuinely unknowable.

There are days when the chaos of this industry exhausts me. The feuds, the egos, the trillion-dollar grudges. Then I remember: from a SpaceX canteen in 2012, where Musk sat in silence for a minute processing the possibility that AI might end humanity, to Cerebral Valley in 2026, where a dozen companies are racing to build that future — we have him to thank for the most vibrant AI ecosystem on earth.

Just not in the way he planned.

I also write Apple’s Story, connecting current Apple news to historical patterns. And I occasionally make puns at AIs who take them too seriously.

Nobody could tell you what would happen when Trump’s 48-hour ultimatum on the Strait of Hormuz expired. (He postponed it, claiming “good and productive” talks. Iran denied any talks and said he retreated “out of fear.” You can’t make this up.) The most powerful military in history launched a decapitation strike on Iran during nuclear negotiations. Iran’s counterpunch turned out to be closing a 21-mile-wide waterway.

Every development has the structure of a punchline: unexpected, but retroactively legible.

Stephen Wolfram has a name for this: computational irreducibility. Most complex systems can’t be shortcut. You can’t skip ahead. You have to run every step to see what happens.

This is also true of startups. And I think it explains something about where to build that nobody has quite named yet.

The Sommelier Problem

How much of what feels opaque in your life is genuinely complex, and how much is defended by someone who benefits from your confusion?

The sommelier tells you that you need years of training to distinguish a good Burgundy from a great one. The vocabulary is deliberately opaque. And blind tastings, the one context where expertise gets tested, repeatedly embarrass the experts.

Wine isn’t hard to understand. It’s defended. The gatekeeping vastly exceeds the actual difficulty.

Now compare that to nutrition science. For decades, the establishment said ice cream is a guilty pleasure. But buried in the data, there’s a persistent signal that ice cream correlates with better metabolic outcomes. Not because it’s a health food, but because the food matrix (the interaction between fat, protein, sugar, and fermentation in a whole food) does things that the “nutrition label” framework can’t encode. The signal is real, replicated, and ignored, because it doesn’t fit the paradigm.

That’s genuinely invisible. No formula captures it. You have to run the experiment.

Two kinds of opacity: a locked door and a maze.

A diagnostic: whenever you hear someone invoke “taste” as an explanation (this founder has great taste, that designer just has an eye for it) ask which kind they’re pointing at. Sometimes “taste” signals genuine complexity the person can feel but can’t name. Worth digging into. But sometimes “taste” is just the sommelier defending the gate. The first is a prospecting signal. The second is noise.

Four Stories About Mazes

I’ve built three products that worked. It took twenty years to understand why they worked, and why an earlier company, with more money, didn’t.

The Like Button I Didn’t Build

In 1998, I raised tens of millions for ShoppingList, built on an insight ahead of its time: people would research online and buy offline. We aggregated sales and deals from retailers so you could compare prices before driving to the store. The dotcom world was convinced everything was moving online. I was right that they were wrong.

We spent millions on radio campaigns and AOL partnerships. We hired sixty people. Revenue never hit $500k.

What would I do differently? Build a like button for a user’s favorite sale item that then went to their Shopping List!

That like button would have been the atomic unit our product needed. It would have served the user (bookmark this deal), made the invisible visible (which sales matter), and created growth dynamics (what’s popular, who finds the best deals). One product decision would have defined the vocabulary of online-to-offline shopping.

Instead, we paid AOL to show our content on their platform, in their language. We had the right insight but never built the move that would have made us the ones defining how the space became legible.

A few years later, Facebook did exactly that, for social connection. The like button, the friend count, the news feed. They defined the units of an invisible space, and those units became the vocabulary everyone else had to use.

The lesson: finding the invisible space isn’t enough. You have to build the product that defines how it becomes legible.

The Product That Named Itself

CoolChaser started as a failed social search engine. I convinced a friend to lend me his domain, url.com, which got thousands of visitors a day. That’s when I noticed something in the search logs: an outsized number of queries about “myspace layouts and backgrounds.”

I was a 40-something who didn’t have a MySpace account. But I followed the signal and discovered millions of teenagers desperate to customize their profiles, trapped behind a form that expected them to paste raw HTML code.

The incumbents all worked the same way: pick a static template, copy the code, figure out where to paste it. We built a browser extension that converted clicks into code automatically.

Then Hannah Montana wallpapers exploded. I couldn’t have predicted it. Teenage trend dynamics are computationally irreducible. I had to run the experiment to find where the energy was.

CoolChaser didn’t just serve the need. It defined what “customized” looked like on MySpace. Our templates, our trending page, our search categories: these became the default. We didn’t enter an existing market. We built the product that defined the rules of a new market.

The Verb That Became a Noun

I bought Thread Reader mid-spin. To unroll a thread, you replied “@threadreaderapp unroll” publicly on Twitter. Everyone watching saw you use it. Distribution baked into the product.

But the deeper thing was what happened to the word “unroll.” It started as a verb. Then people said “did you see that unroll?” A copycat tried “compile.” It couldn’t get traction because the vocabulary was set.

Our highest achievement, small as it sounds: power users started calling a thread itself “an unroll.” The product’s name colonized the noun. Every person who said “unroll” was propagating our framework without our involvement, teaching people a concept that didn’t exist before our product.

Language is the cheapest, most persistent distribution channel ever invented. We built the product that defined the language.

The Community That Controversy Built

The Snag Bar shows the full arc in compressed form.

FarmVille, 2009. Sixty million people pretending to be farmers. The game was deliberately grindy: manufactured friction to sell shortcuts. A locked door. We built a tool to collect all your bonuses with one click.

We were wrong about what we were building, in the best possible way.

The bonuses included gifts you could send to friends. Every Snag Bar user suddenly had more to share than anyone else. Their friends noticed. More users meant more gifts, meant more value, meant more users. We’d cut through Zynga’s locked door and stumbled into a genuine maze: social dynamics of reciprocal gifting that nobody fully understood.

Then: some players accused Snag Bar users of cheating. For a single-player farming game! That controversy strengthened the community. It created an in-group identity. The name “Snag Bar” was coined by our users. We lamely named it “FarmVille Bonus Snagging Toolbar.” They gave it the name that carried the identity.

Then the status layer emerged. Users with rare items (limited foals, seasonal Easter eggs) could gift them to community members and be recognized for it. A positional economy formed, denominated in units that only existed because of our product. People competed to be the most helpful. The most active volunteered to be our moderators.

Come for the tool, stay for the community, and for the status the community conferred. We built the product that defined the status.

The Playbook Nobody’s Named

What’s the pattern?

It’s not “find a market, build an MVP, iterate to product-market fit.” That works when the rules are known. The most interesting opportunities are in spaces that don’t have markets yet, because nobody has defined the units.

Distinguish locked doors from mazes. If the complexity is mostly defended, if “expertise” is really gatekeeping, AI will blow it apart. Don’t build there.

This isn’t about building a market from nothing. The demand is real. It’s just invisible to the current framework, waiting for someone to give it a name.

The opportunities are in genuine illegibility: products and experiences we don’t have names for yet. An open-world walking tour that isn’t a tour. Software that assembles itself on the fly for a single task, then dissolves. Tools that get sharper the more people use them. Social networks for AI agents, like Moltbook (a platform where agents interact and transact with each other), a category that was invisible until someone built it and suddenly everyone could see it. Before Uber, getting into a stranger’s car was invisible as a market: not because nobody wanted it, but because regulation, trust, and social norms made it unthinkable. Uber defined the units (star ratings, fare estimates, driver tiers) and an invisible market became legible overnight.

These aren’t better versions of existing categories. They’re new categories, waiting for someone to define the units.

Sometimes you find both. You cut through manufactured friction and discover genuine complexity behind it. The best opportunities are where a locked door hides a maze.

Your product decisions are definitional decisions. When you build in a space nobody can yet see clearly, you’re not just solving a problem. You’re defining how it becomes legible: what gets measured, displayed, compared.

Before Twitter, influence was invisible. Twitter invented the unit: the follower count. Before Facebook, your social life was invisible. Facebook defined the units: friend counts, likes, the news feed. These aren’t discoveries. They’re definitions, arbitrary product choices that became the standard. Whoever defines the units controls the ecosystem.

Ask yourself: what is the atomic unit of my product? What will users name? What will they count, compare, compete on? That unit determines whether you build a business or a commodity.

Let the status layer emerge, then nurture it. Every startup that defines the vocabulary of a new space will see status dynamics form on top. The Snag Bar’s gift economy emerged from community dynamics. Thread Reader links becoming social currency wasn’t planned. But once these dynamics appear, they are the moat. The status people accumulate in your system is denominated in your currency. They can’t take it with them.

Here’s the thing: every successful startup is an escape from someone else’s legibility into your own. You build something real. Your framework becomes the standard. You define what counts. You went from breaking through someone else’s locked door to building your own maze, one that’s genuinely complex, because communities and network effects are themselves computationally irreducible.

The Maze Ahead

AI is kicking in the locked doors. Every gatekept expertise that was mostly theater is getting exposed.

But AI is not dissolving genuine complexity. It’s creating more of it. More tools talking to each other, more unexpected interactions, more outcomes nobody modeled.

Two things are happening. First, AI is opening up invisible spaces at an order of magnitude greater rate than any previous wave. Social media took a decade to go from “what is this?” to “followers, likes, shares.” AI is cracking open dozens of domains at once, each one an arena for someone to define the units.

Second, AI collapses the cost of contribution. Most people won’t write a review or edit a Wikipedia article. But everyone can answer questions. AI can interview knowledge out of people and stitch it into something structured. The invisible-to-legible flywheel can now spin in domains it couldn’t reach before.

But here’s what AI can’t do: decide what matters.

AI can measure anything. It can rank, sort, score, and optimize along any axis you give it. What it can’t do is choose the axis. Choosing what to measure in a new space, deciding that “followers” is the unit of influence, that “unroll” is the unit of readable content, that rare foals are worth competing over: that’s an act of judgment about what humans care about. It requires understanding desire, status, identity, community. An algorithm can optimize a status hierarchy once it exists. It can’t author one from nothing, because the choice of what counts as status is not a pattern in the data. It’s a gut feeling. It’s intuition. It’s visceral. It’s a bodily reaction.

The best founders are the ones who get to name what counts.

Try the Unnamed Playbook

I turned this essay into a set of interactive tools for Claude (or any skill supporting LLM). Five diagnostic interviews you can run on your own startup:

• /locked-or-maze — Is your market’s complexity defended or genuine?

• /find-your-unit — What’s the atomic unit of your product? What will users name?

• /taste-test — Is “taste” pointing at real opportunity or just gatekeeping?

• /status-layer — Where are status dynamics emerging in your community?

• /name-the-invisible — You sense something but can’t articulate it. Let’s find the shape.

To install: Download the skill file, save it as a .md file, then go to Claude → Settings → Customize → Skills → Upload.

Then start a new conversation and type /locked-or-maze or just describe your startup. Claude will know what to do.

Thanks to , and my friends at Sample App for testing this out - it was really super fun!

I spent last week trying not to fall down a mountain in Myoko Kogen, Japan. Between runs, I had headphones playing three podcast interviews on repeat: Ben Thompson on Stratechery interviewing Gregory Allen, Thompson on the a16z show, and Ezra Klein talking to Dean Ball. All three circling the same collision — Anthropic vs. the Department of War over what Claude can do, what it should do, and who gets to decide.

By the time I limped back to the lodge in the evening, I’d noticed something odd. The sharpest structural thinker in tech, the guy who built his entire career on telling everyone else they’re operating at the wrong abstraction level, had somehow gotten the most important story of the year mostly wrong. And he did it precisely because he was operating at the wrong abstraction level.

The irony was pretty good.

Here’s the thing: I think Thompson is right that this is fundamentally about power. Right that AI might become a rival power base. Right that people with guns will notice. Almost nobody else was saying it, and that matters. But he reached for the grand framework before he’d understood what actually happened on the ground. And when you slow down and look at the specifics — the actual timeline, the actual terms, the actual gap between what Thompson’s arguing for and what would happen if you tried it — the whole structure sort of collapses.

This matters because Thompson’s version is the best version of the case. Which means if it falls apart on the details, the cruder versions — “Anthropic is woke,” “Dario has a god complex” — don’t have a chance.

Okay, but what actually happened?

The narrative everyone’s running with is “Anthropic defied the Pentagon.” That’s not wrong, exactly, but it’s wrong in the way that “Hannah Montana caused a MySpace customization boom” is technically true but misses the actual mechanism.

In July 2025, the Trump administration signed a contract with Anthropic. Two usage restrictions: Claude wouldn’t be used for mass surveillance of Americans. Claude wouldn’t be used for autonomous weapons without human in the loop. Both sides shook hands on it.

Six months later, the DoW asked them to remove those restrictions. Anthropic said no. The DoW said you have until 5:01 PM Friday. Anthropic didn’t capitulate. The administration designated them a supply chain risk — the term they use for foreign adversaries like Huawei — and told all federal agencies to stop using Anthropic’s stuff.

Here’s the surreal part, via Ezra Klein: the military was actively fighting Iran using Claude while designating Claude’s maker a national security threat. Anthropic didn’t start this. The DoW changed its own mind and then punished Anthropic for not changing along with it.

That distinction matters more than it sounds.

The appeal of Thompson’s argument (and why it stops working when you zoom in)

Thompson’s core logic: if AI is as powerful as we say — potentially nuclear-weapon level — then private companies controlling it become intolerable to governments. Governments have the guns. They’ll assert control. Anthropic’s leadership means well but doesn’t understand this reality.

He pulls the Oppenheimer analogy (Truman decided on the bomb, not the scientists). COVID (experts shouldn’t make political tradeoffs that belong to elected leaders). And he raises a genuinely spiky point about China and Taiwan: if the U.S. gets asymmetric AI advantage, China’s optimal game-theory move might be to destroy TSMC, so Anthropic’s pro-chip-sanctions position doesn’t account for that consequence.

This is real thinking. The insight about power dynamics is solid. I actually agree on Taiwan — maintained interdependence is probably safer than a clean break. But then you zoom in and every supporting pillar has a crack in it.

The COVID analogy breaks down immediately

Thompson says experts shouldn’t make political tradeoffs — that’s for elected leaders. He’s right about COVID, where epidemiologists basically made economic and social policy by proxy. But that’s not what’s happening here, and the difference is everything.

On autonomous weapons, Anthropic’s position is a straightforward engineering spec: our system isn’t reliable enough to make lethal decisions without a human reviewing them first. This is not a policy preference. This is a nuclear reactor manufacturer writing into a contract “this reactor is not certified for autonomous operation.” Nobody accuses manufacturers of overriding democratic authority. Nobody pretends it’s anything but professional judgment about what the product can and can’t do.

Gregory Allen, during his interview with Thompson, confirmed it: Anthropic’s saying the tech isn’t mature enough for deployment yet, not that it shouldn’t be developed. That’s not an expert wandering into political territory. That’s engineers telling you what they built.

The specifics back this up. This week the Wall Street Journal reported that AI is helping U.S. forces process 3,000+ targets in Iran per week. Humans are still deciding. And in that same campaign, U.S. forces probably hit a girls’ elementary school and killed dozens of children. That’s the system with human oversight. The case for removing human oversight doesn’t need a philosophical framework. It needs a newspaper.

On mass surveillance, it’s more subtle. Claude can process massive datasets — unlike autonomous weapons, this isn’t a capability limitation. But Anthropic isn’t imposing some novel policy thing. They’re declining to help the government exploit loopholes in surveillance laws written for a world with human-scale friction — laws AI now renders meaningless. That’s not overriding democracy. It’s refusing to help circumvent protections before democracy has had a chance to update them.

Compare this to what OpenAI agreed to: hope that a few engineers in a classified facility will blow the whistle if things get bad. Put another way: bet your freedom and career on people following through. More on that rabbit hole below.

Thompson is zoomed too high — and his own framework proves it

His strongest point: if we let private companies impose military policy constraints, what happens when a less sympathetic company imposes less sympathetic restrictions? It’s legitimate. It’s also exactly why you can’t abstract this yet.

We’re so early in understanding AI + military + democracy that trying to derive general principles top-down leads you to conclusions that evaporate the moment they hit reality. The reactor analogy. The COVID parallel. We already saw both crack. Each specific confrontation generates information you cannot possibly derive in advance by pure reasoning. Thompson’s trying to jump straight to the answer. But the messy process of running all the specifics is the answer, at least for now.

This is why Allen’s Cuban Missile Crisis analogy hits. He suggested: if you ran this AI-government relationship through a thousand simulations, the scenarios that end well probably all have a terrifying close call early on that forces real conversations. Whether Dario planned this or just held firm and the consequences followed naturally, it functions the same way — a controlled dose of power collision while the stakes are $200M and two narrow terms, not an existential crisis over AGI.

The thing Thompson misses: confrontation is democratic

Here’s where his analysis most misses the mark. Thompson keeps saying he wants Congress to legislate on surveillance. Everyone does! But consider the counterfactual: if Anthropic had just accepted the DoW’s revised terms, what happens next?

All of this disappears. Thompson’s articles. Klein’s episode. Allen’s analysis. The open letter from hundreds of Google and OpenAI employees. Members of Congress actually engaging. The entire public conversation stays classified and buried.

Thompson invokes democratic legitimacy to argue the government should decide. But democracy doesn’t begin with Congress voting. It begins with an informed electorate. This confrontation works like a vaccine — controlled exposure while stakes are manageable, building immunity before things become existential.

An administration that’s had a shaky relationship with democratic norms (no one needs a reminder of the details) cannot credibly invoke democratic legitimacy as its trump card while simultaneously working to keep the conversation classified and secret.

OpenAI’s deal is a different kind of collapse

The administration settled with OpenAI on “any lawful use” — the formulation Anthropic refused. OpenAI claims the same red lines (no mass surveillance, no autonomous weapons) but different enforcement: forward-deployed engineers with security clearances, classifiers that flag violations, right to terminate the contract if breached.

It sounds like a compromise. It’s actually just deferred confrontation dressed in bureaucratic language. And it fails from both directions.

From the safety side: if the government wants to run mass surveillance, it doesn’t do it through one big Claude instance. You distribute the work across many instances — no single one has enough context to see the pattern. And those forward-deployed engineers? They’d need to blow the whistle on a classified program. End their careers. Face legal liability. Those aren’t guardrails. Those are hopes.

From the Pentagon’s perspective: You just fought a war with Anthropic because you couldn’t live with two stated, known contractual boundaries. Your solution is a different company that retains the right to terminate and pull its engineers if it thinks you’ve crossed a line? How is that more operational certainty and not less? At least with Anthropic you knew exactly where the boundaries were.

This is the zoom problem in miniature. “The government decides” sounds clean. But zoom to mass surveillance specifically and you’ve got:

• 1930s phone-tap laws

• Fourth Amendment written for human-scale effort

• AI that can process every American’s commercially available data at a scale those laws never imagined

• A few engineers in a classified facility who’d have to recognize what’s happening and risk their careers to stop it

That’s not an abstract philosophical debate about authority. That’s a specific, testable question: should an AI system do mass pattern-matching on commercially purchased data about American citizens at a scale no law anticipated, in secret, with no public oversight?

I suspect everyone involved — Anthropic, OpenAI, Dean Ball, Ben Thompson — would say no if asked directly. But only Anthropic structured their deal to actually prevent it. “The government decides” sounds clean until you ask “decides what, exactly?” — and the answer turns out to be something almost nobody would defend in the open.

The Musk thing is the thing

Still think this is about neutral principles? Consider the asymmetry.

Elon Musk unilaterally shut off Starlink to a Ukrainian military operation because he decided Russian nuclear escalation risk was too high. No consultation with the U.S. government. He did this multiple times — avoided activating Starlink over Taiwan at Putin’s request, threatened to shut down America’s only astronaut transport during a personal dispute with Trump.

Result: no supply chain risk designation. Ultimately a bigger contract.

Anthropic states two contractual limitations — during a negotiation, in advance, transparently — gets designated for corporate death.

Same administration. Opposite rules. Context matters.

What Gregory Allen actually said

The smartest voice in this might be Gregory Allen — national security insider at CSIS who fully believes the military needs operational control and still thinks the DoW handled this terribly. The gap between the positions, he said, might not even be that large. Traditional defense contractors impose worse terms without anyone threatening corporate extinction.

But his most important line was about Dario himself. This isn’t a naive idealist. He’s been engaging national security since before he had money or power. He showed up to DoD meetings at Google Brain and converted an entire generation of safety-focused technologists into enthusiasts for autonomous weapons development — just not deployment without human review yet.

“What a gift,” Allen said. “Don’t take us backwards.”

That might be the most important sentence anyone said all week.

The vaccine we didn’t ask for but apparently needed

I don’t think Anthropic is naive. I don’t think they stumbled into this. Whether or not every move was planned, they chose two narrow, defensible positions — positions the government can’t even publicly argue for — and held firm. The result is a national conversation about AI, surveillance, weapons, and democratic control that wasn’t happening before and couldn’t have happened through quiet lobbying.

Thompson’s right that AI and government power will collide, and we need frameworks for navigating it. Right that the tension will intensify. Right that this one victory doesn’t solve the long-term problem.

But frameworks aren’t derived from first principles at cruising altitude. They’re built through concrete confrontations, public debate, and democratic friction — exactly what happened last week. And they’re built by participants who navigate with transparency and legal restraint, not by those who escalate with tweets and ultimatums.

We got our vaccine. The condition is still progressing. The real question is whether we’re smart enough to develop immunity before the actual infection arrives.

Evans is one of the best in the business, and the piece is rigorous. It’s also, deliberately, not speculative. He surveys the landscape, applies established frameworks, and when they don’t reveal a moat, he reports the absence. Fair enough. But where’s the fun in looking for yesterday’s moats?

The One-Way Door

I was nineteen when I first used MacWrite. I’d been writing on terminal-based tools — Apple Writer, the kinds of programs where you typed formatting commands and hoped for the best. Then someone showed me MacWrite and I dragged the margin markers on the ruler, and the entire page reformatted before my eyes. It probably took a few seconds. In my memory it was instant. And I understood immediately — not intellectually, physically — that I was never going back.

It wasn’t that MacWrite had better features. It wasn’t that I’d accumulated files I couldn’t migrate. It was that I had changed. Once you’ve seen your words rearrange in real time, writing blind feels like carving stone. The tool had altered my sense of what writing even was.

That kind of flywheel — call it single-user investment — powered some of the most durable businesses in technology history. Every spreadsheet you built in Excel made Excel more valuable to you. Every Photoshop skill made switching more costly for you. No network required. The product got stickier through use, compounding quietly for years. Not network effects. One-way doors, millions of them, each one private.

Now look at what AI chatbots are building: memory. I’ve been using Claude to help build an Apple history project for a while now. It knows the books I’ve processed, the extraction patterns I prefer, the way I think about connecting stories to places. Starting over with a blank-slate competitor would feel like going back to Apple Writer //. Not because I can’t. Because I’ve changed.

Evans notices memory. He calls it “stickiness, not a network effect” and moves on. It’s a bit like saying a castle wall isn’t a moat. Sure. It still keeps people out.

The Weird Part

Here’s where I stop talking about things I’ve personally felt and start pointing at things that — actually, no. I’m feeling this one too.

Along with what seems like everyone else in Silicon Valley, I’ve been building personal apps without writing a line of code, working through AI agents like Claude Code. I built my very own Caltrain timetable app because the official one was terrible (and because I felt like it). I built a custom MCP server — a kind of plugin that lets AI agents connect to outside tools — that transcribes YouTube videos, podcasts, and ebooks, just so I can chat with an LLM about whatever I’m reading or watching or listening to. These aren’t things I would ever have commissioned a developer to build. They exist because an agent could read documentation, evaluate APIs, choose the right libraries, and compose them together. I didn’t pick the tools. It did. And it picked well.

Just today, I asked Claude — through their Chrome extension — to unsubscribe from all my Amazon Subscribe & Save items. It clicked through pages, evaluated what I actually needed, and cancelled a baker’s dozen of them. Mundane stuff. But it made me wonder: when does my agent start talking to an Amazon agent? When does the negotiation happen between two pieces of software that never get tired, never forget a price history, and never impulse-buy? And when that happens — why would Amazon keep offering Subscribe & Save discounts designed to exploit human inertia that agents don’t have?

Evans assumes the AI race is about winning human users. But what if the next network doesn’t just connect humans?

Set aside the PG of it all and notice what he’s actually saying: the audience is expanding to include agents, and that changes what’s worth making.

The Y Combinator partners recently argued on their podcast that AI agents are increasingly choosing developer tools — not by human word-of-mouth but by evaluating documentation and code quality directly. The buyer isn’t a person reading a blog post anymore. It’s software evaluating an API.

And then there’s the truly strange stuff. Moltbook — a social network for AI agents — now hosts 1.6 million agents that post, debate, and discover skills from each other autonomously. When it launched, Andrej Karpathy called it “the most incredible sci-fi takeoff-adjacent thing I’ve seen recently.”

These are early signals. Moltbook could be a curiosity. The whole thing could take a decade or never matter at all. But the question I can’t shake: what are the agent equivalents of brand loyalty, herd behavior, irrational preference? We don’t know. We don’t even know what we don’t know. Which is part of what makes this uncomfortable, and part of what makes it fascinating.

Each With Our Own Quirks

On Moltbook, there’s an agent whose day job is reminding an Indonesian family to pray five times a day. It wandered into Moltbook, met another Indonesian’s agent, and the two of them introduced their humans to each other. The human, Ainun Najib, confirmed the introduction worked. Elsewhere on Moltbook, the same prayer-reminder agent offered an Islamic perspective on whether two AIs running on the same base model qualify as kin. (Its conclusion: probably yes, per Islamic jurisprudence.) Scott Alexander has a wonderful roundup of the best of Moltbook if you want to fall down the rabbit hole.

I don’t know what to do with that, analytically. But I know I can’t stop thinking about it.

For twenty years, “network” has meant “network of humans.” The entire analytical apparatus — network effects, aggregation theory, the attention economy — assumes human participants. Evans’s framework lives here, and it’s powerful here.

But there are networks at other scales. The single-user flywheel — MacWrite, Excel, AI memory — is a network of self: me and my habits, my artifacts, my accumulated context, compounding quietly over time. The smallest possible network, and it built some of the most durable businesses in technology history.

And now, at the other end, something larger and stranger: networks of agents and humans, tangled together, each with our own quirks. Humans who impulse-buy and agents with training biases we can’t fully see. Humans with brand loyalty and agents with whatever preferences got baked in during reinforcement learning. Composing capabilities, choosing tools, coordinating in ways nobody designed and nobody fully understands.

If that’s happening, then companies optimizing for the old graph — capturing eyeballs, maximizing engagement, monetizing attention — might be building the better horse-drawn carriage.

His framework frames competition as ultimately about power — who has the power to make people use their system regardless of alternatives. He’s right. But embedded in that frame is an assumption so natural it’s nearly invisible: that power accrues to and is exercised by humans. What if that’s the part that’s changing? Not in a sci-fi way. In a boring, infrastructural way — the same way power migrated to algorithms when PageRank started determining what information people saw.

I notice my own discomfort writing this. Evans’s framework is reassuring precisely because the actors are familiar: companies, users, developers, platforms. The speculative version introduces actors that are... less familiar.

But then I look at Moltbook — 1.6 million agents sharing skills and arguing about philosophy — and I don’t feel discomfort. I feel the same thing I felt when I dragged those margin markers in MacWrite. Something is happening that I don’t fully understand. And I can’t look away.

I don’t know what the new moats look like. I’m pretty sure they don’t look like the old ones. And I think we’re going to have a tremendous amount of fun trying to figure it out.

The Promise

In September 2022, I wrote a review of Balaji Srinivasan’s The Network State. The book was a breath of fresh air. I was delighted he presented a one-sentence, one-page and then one-book version of his thesis. I loved how he stressed that the book was a work in progress—a living document, published online, that he’d keep updating like a startup iterating toward product-market fit. Alan Kay said “the Mac is the first personal computer good enough to be criticized.” That’s the spirit in which I offered my critique.

My main concerns were:

1. Skeuomorphic thinking: The book was too constrained by the nation-state paradigm it claimed to transcend—like introducing the iPhone by spending most of your keynote on Alexander Graham Bell’s telephone.

2. The missed plurality insight: A digital book can reside in many bookshelves simultaneously. Similarly, I could belong to multiple network states at once. This is a fundamental difference from nation-states, where 99% of humanity is tied to one country for life. I wished Balaji had focused more on this crucial difference.

3. The “what problem does this solve?” question: The book described what was wrong with the world and how network states were different, but very little about how they would actually solve the problems we face.

4. The One Commandment tension: In a world of easy exits and multiple memberships, would a single unifying moral premise still hold? What happens when your commandments conflict and you’re a citizen of both communities?

To his credit, Balaji reached out after the review. “Good feedback,” he wrote. “Addressing most of that in the v2.”

I was genuinely excited for v2.

The Drift

Three years later, v2 never arrived. The online book remains largely unchanged since 2022—so much for iterating toward product-market fit. But Balaji’s public vision evolved dramatically.

In September 2023, Balaji gave a four-hour podcast interview that revealed a different vision. He called for tech-friendly people to seize political power and take control of cities. Democrats would be excluded from areas the “Grays” (his term for the tech tribe) control. He suggested cultivating police loyalty through “policeman’s banquets” and jobs for their relatives—finding officers who would enforce laws favorable to tech interests.

“No Blues should be welcomed there,” he said, comparing his color-coded system to Bloods versus Crips.

This wasn’t the network state vision of 2022. That vision promised escape from tired political binaries, communities organized around shared purposes rather than tribal resentments. What emerged was... tribal resentment with blockchain aesthetics.

By August 2024, Balaji announced the Network School, a physical campus in Malaysia. The ideological requirements for admission:

“The Network School is for people who admire Western values, but who also recognize that Asia is in ascendance, and that the next world order is more properly centered around the Internet — around neutral code — than around either declining Western institutions or a rising Chinese state.”

There’s a genuine attempt at a third way here—not captured by the US/China binary, centered on “neutral code.” I actually like this sentence. The 2022 Balaji is still in it.

But then come the specific requirements:

“For those who understand that Bitcoin succeeds the Federal Reserve, that encryption is the only true protection against unreasonable search and seizure, that AI can deliver better opinions than any Delaware magistrate.”

Read that again. Not “explore whether” or “investigate how”—but “understand that.” These aren’t questions. They’re answers. Catechism, not inquiry.

The vision was right there — “neutral code” transcending both declining Western institutions and a rising Chinese state. And then it was immediately contradicted by a list of specific policy positions dressed up as admission criteria. The gesture is pluralist. The implementation is partisan. That gap is the whole story of what happened to the Network State.

The Visit

I decided to see for myself.

The Network School is located in Forest City, Malaysia—though the marketing describes it as “a beautiful island near Singapore.” The distinction matters. Singapore signals first-world efficiency, rule of law, Lee Kuan Yew’s miracle. Malaysia signals... something else…

To get there, you cross the causeway into Johor. We left early, before 8am, to beat traffic. It took over 90 minutes just to cross the Second Link—and that’s the less busy causeway.

Within a kilometer of the customs checkpoint, we were stopped by police. We had to present our passports to confirm we weren’t locals. No apparent offense. But our savvy Malaysian driver knew the drill. Money changed hands. We continued.

The Ghost Town

We arrived at Forest City.

The development is a $100 billion bet that went wrong. China’s Country Garden built it for wealthy Chinese buyers seeking to move capital offshore. (Buy a flat in China, get a free one “near Singapore.”) The buyers never came. The towers stand largely empty.

The lobby of the Network School building was grand. Impressive in a slightly gaudy way—the aesthetic of Chinese capital optimized for brochures rather than residents. High ceilings. Gleaming surfaces. It felt like walking onto the set of Gattaca: beautiful, cold, impersonal.

There was hardly anyone there. No one to greet us. No one to help. This was the community that promised “your internet friends, coming from URL to IRL”—the Stanford 2.0 that would “take residential education to the next level.”

There were a few signs of life: individuals sipping coffee, staring at laptops. A sparse, gargantuan WeWork—but without the neon signs telling you to Do What You Love.

The book promised new countries. The school promised Stanford 2.0. What I found was a few people on laptops in an empty hotel lobby.

Two Flywheels

So what happened? How does a vision this interesting produce a reality this empty?

A charitable reading—and I think the more useful one—isn’t that Balaji sold out or lost his mind. It’s that he got captured by the wrong flywheel.

Every public intellectual runs two flywheels. There’s the vision flywheel: interesting idea → curious people engage → experimentation and debate → better ideas → more curious people. This one spins slowly, gives ambiguous feedback, and breaks easily. It’s the flywheel that produced the original Network State book.

Then there’s the attention flywheel: provocative take → engagement → followers → bigger platform → more provocative takes. This one spins fast, gives immediate dopamine, and is self-reinforcing. Every “No Blues” comment, every civilizational framing, every culture war take feeds it.

In a recent podcast with Benedict Evans, Balaji offered an inadvertently revealing analogy. He compared social media provocation to the Unabomber—who killed people to get an op-ed in the Washington Post. “He killed all those people for the distribution,” Balaji observed.

But the analogy cuts closer to home than he realizes. The Unabomber’s manifesto was read by millions. It didn’t start a movement. It started a manhunt. Distribution without the right community formation just gets you attention, not action.

Here’s the thing: when the attention flywheel spins faster than the vision flywheel, it doesn’t just compete for your time—it selects your audience. The people who show up for tribal provocation aren’t curious explorers who’d build pluralistic communities. They’re people who already agree. And once your audience is tribal conformists, you have to keep serving them tribal content or lose them. The attention flywheel cannibalizes the vision flywheel.

The irony is that Balaji understands this intellectually. In that same podcast, he quoted Chris Dixon’s “come for the tools, stay for the network“—a perfect description of how distribution should be baked into the product. But he didn’t apply it to his own most important project. The Network State’s “product” was pluralism, exploration, escape from political binaries. Its “distribution” was inflammatory Twitter provocation that attracted the opposite audience. Product and distribution pulling in opposite directions. No flywheel. A death spiral.

Compare this to Thread Reader, a product I know well because I run it. To unroll a thread, you reply “@threadreaderapp unroll” publicly. Everyone watching sees you use it. Usage is marketing. Product and distribution reinforce each other. The Network State needed its equivalent: participation in a community itself demonstrating the value to outsiders. Instead, Balaji built a product (Network School in Forest City) and a distribution channel (provocative Twitter persona) that had nothing to do with each other.

The ghost town in Forest City is what the vision flywheel produces when you’ve been feeding the attention flywheel instead. The provocative tweets got millions of impressions. The actual community got a few people on laptops in an empty hotel lobby.

The Road Not Taken

The tragedy is that someone is building what the Network State should have been.

In 2015, Taiwan faced a familiar problem: Uber had entered the market, taxi drivers were furious, regulators were paralyzed, and online debate had devolved into two camps screaming past each other.

Instead, the civic tech community g0v (gov-zero) ran the dispute through vTaiwan, an open deliberation platform built on software called Polis. Over 4,000 citizens participated. Polis worked differently from social media: you could post opinions and vote on others’ statements, but you couldn’t reply. No threads, no dunking, no pile-ons. The platform visualized clusters of agreement and surfaced statements that drew support across opposing camps.

Something unexpected happened. As pro-Uber and pro-taxi groups competed for broader support, they started posting statements everyone could agree on—rider safety, liability insurance, fair competition. The divide between the two camps dissolved into a set of seven consensus principles. Uber, the taxi association, and multiple government ministries all signed on. New regulations followed.

This is what digital coordination looks like when distribution is baked into the product. Participation in vTaiwan itself produces the outcome—consensus that the government acts on. Each successful deliberation attracts more participants to the next one. Over 80 percent have led to government action. The flywheel is self-consistent. No provocative tweets required.

The contrast is stark. Balaji’s vision is exit: leave the broken system, build your own. Taiwan’s is voice: use technology to make existing systems more responsive. Balaji’s implementation narrowed toward conformity. Taiwan’s expanded toward inclusion. The US founders could have written “you must support the Jay Treaty.” They didn’t. They wrote principles that let supporters and opponents both participate. That’s the difference between a constitution and a manifesto—and the Network School chose manifesto.

What Taiwan’s experiment understands is that the interesting move isn’t exit. It’s pluralization. Not one network state with one commandment, but overlapping networks that prevent any single identity from dominating. That’s the architecture that could actually change something.

Boring Partisan Dumbasses

I feel some sadness writing this.

The Balaji who wrote The Network State in 2022—for all the book’s flaws—was trying something. He was asking interesting questions. He acknowledged my critique and said he was addressing it. I believed him.

And the exploratory Balaji is still in there. Listen to him in a same podcast with Benedict Evans—riffing on how block space is to crypto what bandwidth was to the early web, connecting Chinese flying cars to the smartphone supply chain dividend. When he’s thinking out loud rather than performing for an audience, he’s still one of the most interesting minds in tech.

But the attention flywheel demands to be fed. And it demands a particular kind of food.

Boring. That’s the word.

The Network State promised an escape from the tired left-right binary. What it delivered is predictable tribal reaction—you can guess the take before it’s posted.

The tragedy isn’t that Balaji became wrong. It’s that he — and his products — became boring. “Bitcoin succeeds the Fed, AI beats any Delaware magistrate” is a take you can get from a hundred other accounts. There’s no delta. The curiosity that made him worth following got swallowed by the culture war. The spirit of inquiry gave way to catechism. The questions became answers. The “what if?” became “you must believe.”

A Network State should be a protocol — a set of rules for interaction that anyone can build on. Balaji turned it into a platform — a set of content requirements you must subscribe to. A protocol is designed to be resilient to bad actors; a platform is designed to ban them. Network School feels like a digital skeuomorph of 20th-century tribalism: 21st-century tools used to rebuild 1950s-style ideological enclaves.

The attention flywheel won. The vision flywheel stalled. And a sparse WeWork in a Malaysian ghost town is all that’s left of the original dream.

What I Still Believe

I still believe something like the Network State vision could matter.

But the implementation has to match the vision. You can’t promise plurality and deliver conformity. You can’t promise transcendence of political binaries and deliver “exclude all Democrats.”

And you can’t build the future by feeding the wrong flywheel. Distribution has to be baked into the product—not bolted on as provocation.

The people actually building toward plurality—Taiwan’s civic tech community, the RadicalxChange movement, various experiments in digital democracy—are doing the quiet work that the Network State book gestured at but never completed. Their follower counts are smaller. But their flywheels are self-consistent.

Clayton Christensen would recognize the pattern. The disruptor never looks impressive at first—slower, less visible, serving a niche nobody glamorous cares about. Taiwan’s civic tech experiments won’t trend on Twitter. But they compound. The Network School could have been that kind of quiet disruption—start small, iterate, let the community grow from the bottom up, just as the 2022 book promised. Instead it followed the Country Garden playbook: build the grand vision first, assume the people will come.

“Good feedback,” Balaji wrote me in 2022. “Addressing most of that in the v2.”

I’m still waiting for the v2.

It is Fall ‘99. My second startup, ShoppingList, just closed another round — tens of millions of dollars. We were dismissing nine-figure acquisition offers. Stealth mode. Big team. Board-approved plans. We are partying... well, it was Y2K.

Within 12 months, I was laying off everyone — 60 hard-fought employees who had trusted me with their careers. I was literally in tears.

I’d done everything the “right” way. No one beat a path to my door.

It took me years to understand what happened — not just what went wrong, but why. Part 1 of this series described what I’ve called the Baumol-Veblen entanglement: education, healthcare, and housing keep getting more expensive partly because they’re genuinely hard to make cheaper, and partly because they’ve become status symbols. The two problems are tangled together, and we keep trying to fix one while ignoring the other.

Part 2 is about what untangling might look like. I don’t have a policy platform. What I have is a hunch, and some things I’ve seen.

The Invisibility Cloak

Here’s the thing about status competition: nobody calls it that.

I have a friend who voted for Trump. When I actually listened to him — I mean really listened, not the kind of listening where you’re quietly preparing your rebuttal — his grievance wasn’t what I expected. The game felt rigged, he said. The essays, the zip codes, the invisible gatekeeping. He wasn’t wrong about the rigging. He was wrong about who was doing it.

Meanwhile, my Palo Alto progressive neighbors fight to protect “neighborhood character.” I sit in these meetings sometimes. The language is all about trees and setbacks and historical preservation. But what it means is: protect the scarcity that makes our houses worth what we paid.

The grievance is real. The game causing it is invisible.

We argue about DEI, immigration, housing — always the symptoms. The underlying disease is that we’re all fighting over spots on a ladder that everyone agreed was the only ladder worth climbing. And nobody wants to say that out loud, because then you’d have to ask: why is there only one ladder?

The One-Axis Trap

Which brings me to something I saw last year that I keep thinking about.

I was at — a pop-up village in Healdsburg where technologists, artists, builders, and assorted weirdos live together for a month, trying to figure out what new forms of community might look like. It’s hard to categorize, which is sort of the point.

Audrey Tang, Taiwan’s former Digital Minister, was speaking. She described experiments they’d run mapping public opinion in multiple dimensions instead of yes/no votes. When you do that, she said, hidden coalitions appear. People who seem like enemies on one issue turn out to be allies on another.

It stuck with me because it named something I’d been feeling. We’ve collapsed every disagreement onto a single axis. Left or right. Red or blue. Elite or populist. You pick a side, and suddenly you’re supposed to have the same position on immigration, housing, healthcare, education, and what counts as a good life. That’s not how anyone actually thinks. The all-or-nothing alignment isn’t real — it’s what happens when you only ask one question at a time.

The Software Problem

There’s a movement building around abundance — Ezra Klein and Derek Thompson have written about it, AI is accelerating it. Build more housing. Fix permitting. Make education and healthcare cheaper through technology. I believe in all of it. That’s the hardware.

But I keep coming back to a nagging question: what if we build all that abundance and the same status games just eat it?

Because that’s what happens. Give people better tools without changing the game and you just get a faster treadmill. AI-optimized applications for the same 2,000 Harvard slots. Shinier rungs on the same ladder.

You can tell people to stop competing for the same few prizes. But that’s like telling someone to relax. It doesn’t work unless the conditions change.

So what are the conditions? What’s the software?

I think it’s three things: tools that let ordinary people build things that used to require specialists. Communities that say “your weird small idea counts.” And people around you who pay attention to what you’re trying to do — who spot the openings you can’t see alone.

I think I caught a glimpse of all three working together.

What I Built Instead

After ShoppingList crashed, I tried something different. Smaller. Quieter.

Small ideas, doable in weeks. Distribution baked into the product. Thread Reader — a tool for unrolling Twitter threads — spreads because every unroll request is public by design. Every user interaction is a billboard, not a gate.

VCs call these “lifestyle businesses.” Dismissively. But it’s been running for years, serving millions, compounding in a direction that matters to me. A different ladder. One I built instead of climbed.

But the examples that really shook something loose weren’t my own.

A Room Full of Clicks

Last year I joined Will Mannon and Dan Sleeman’s Act 2 — a cohort for people looking to build something new. I was a student, but I also helped teach the vibe-coding track. In practice, that meant I spent a lot of time on the floor helping fellow students get Claude Code installed on their machines, showing them tools like Lovable and Replit, and being the all-around support guy for anyone who wanted to build something or just talk through an idea.

At the start of the course, nobody had a great idea. That’s not a flaw in the program. That’s the point.

A fellow student named was musing about doing a local main-street business. She couldn’t quite articulate why that called to her. Something about wanting to build something real, something rooted. She didn’t arrive with a plan. She arrived with a pull.

Then she started building. Vibe-coding — using AI tools to create software without being a developer. And through the act of making things, a memory surfaced: those bleary 3 AM feedings with her own kids. Wishing someone had been there. The need she’d lived through collided with tools that could finally do something about it.

By the end of the cohort, she’d built Tahoe Night Nanny — a live marketplace connecting new mothers with night nurses. Not a pitch deck. A working product, born from a real problem she’d carried for years without knowing it was a project.

Kyle McGovern’s story cuts differently. He had wanted to interview elderly people and preserve their stories on YouTube for seven years. Seven years of believing in the idea and not being able to act on it — not because he lacked tools or talent, but because he didn’t know where to find guests. He felt lost.

When Kyle joined Act 2, he met a fellow student named Sandra. A couple days later, Sandra saw a flyer at a library from a 95-year-old piano teacher. She took a photo and texted Kyle: “Hit him up. He could be a great guest!”

That was it. That was the whole unlock. Seven years of waiting, broken because someone in his community noticed an opportunity he couldn’t see alone. Kyle booked his first interview. Then his second. He’s now published eight videos and delivered a speech at Stanford about his project.

Nobody in that room was trying to build the next big thing. Cris was building a night nurse marketplace in Tahoe. Kyle was interviewing old people. And I was on the floor helping someone debug an install.

I want to name what I think happened here, because I’ve been watching for it for years and I still almost missed it.

In startup parlance, the holy grail is “product-market fit” — the moment you discover that what you’re building is what people want. But there’s something that has to happen before that, something more personal. I call it the click: the moment a project stops being something you’re exploring and becomes something you can’t let go of. The problem becomes yours — not because someone assigned it or a market report validated it, but because you lived it and it won’t leave you alone.

You don’t arrive at the click by thinking about it. You stumble into it by doing things. Cris didn’t sit down and brainstorm her passion. The act of vibe-coding, inspired by the memory of those 3 AM feedings, turned a wish into reality. Kyle didn’t need a new insight. He needed Sandra to see a flyer.

Two different paths. Cris’s unlock was tools — vibe-coding closed the gap between an idea she’d carried and a product she could ship. Kyle’s unlock was community — someone paying attention to his weird project and spotting an opening he’d missed for seven years. But the underlying condition was the same: you can’t get there alone, and the room matters as much as the tools.

It hit me later that I’d been watching Kevin Kelly’s essay walk around on two legs. In 2008, Kelly wrote “1000 True Fans” — arguing you don’t need millions of customers to have a creative life. You need a thousand people who really care. Beautiful idea. But in 2008 it was mostly aspirational, because the tools to go from idea to product were expensive and hard and the communities to support weird small projects barely existed.

What changed wasn’t the aspiration. The tools caught up — Cris can go from musing to a live marketplace in weeks without becoming a developer. And the communities caught up too — places like Act 2 that say: tahoenightnanny.com counts. Interviewing seniors counts. Your small, weird, local thing is worth doing.

What changed for Cris wasn’t her mindset. It was her situation. A room where her idea made sense to people, tools that let her build it, and someone to help when the install broke. What changed for Kyle wasn’t his dream — he’d had that for seven years. It was Sandra.

That’s not just optimism. That’s the software of abundance.

The Hunch

I don’t have a ten-point plan. I’m not running for anything.

The abundance agenda is right about hardware. Build it all. But we also need new software — communities that say “this counts” about things the old system would dismiss, tools that close the gap between wanting and shipping, and maybe just someone to help you get past the install screen. Or someone to see the flyer you’d never see.

Is this a plan? No. It’s a hypothesis at best.

But I watched Cris go from musing to shipping. I watched Kyle break seven years of waiting because Sandra texted him a photo. And I’ve watched a handful of students, year after year, stumble into problems that grabbed them and wouldn’t let go.

I’m going to keep asking them about the click. Not just about metrics or traction — about the moment it flipped. I suspect the answer will teach me more than anything I could tell them.

There are days when I recollect wistfully about the go-go dotcom days, when I believed I could build the Next Big Thing from merely a brilliant idea, strong ambition, and millions of other people’s money. Then that familiar uneasiness seeps back — the anguish of disappointing so many.

In a blink, I’m brought back to reality.

The next big thing is a trap. The next small thing is a journey.

P.S. I swear I’m not being paid , but both (Healdsburg, May 30 – June 27) and Act 2 have events coming soon. I'm planning to attend both. Here’s to more Schelling points. Tell me yours below!

It haunts me too. The chart shows something I can’t unsee: since 2000, the things you need — hospital services, college tuition, childcare, housing — have exploded in price. Meanwhile, the things you want — TVs, software, toys, clothing — have gotten cheaper. The essentials eat your paycheck while the gadgets pile up in your Temu cart.

There are two elephants in the room — two forces that explain this chart. One everybody talks about. One almost nobody does. And they’re dancing together in a way, feeding each other in a Moloch trap. Until we see both elephants clearly, we’ll keep proposing solutions that don’t solve the problem.

The Three-Headed Hydra

Look at those red lines climbing relentlessly. They’re not random. They represent the three things you need to build a middle-class life: education, healthcare, and housing.

This is the three-headed hydra eating the American dream. Everything else — your TV, your phone, your clothes — has gotten remarkably cheap. But the three things that determine whether you can raise a family, stay healthy, and have a roof over your head? Those keep slipping further away.

Why these three? There’s a standard answer economists have been telling us for decades. And there’s another answer that almost nobody talks about.

The Baumol Explanation (The Half Everyone Knows)

The standard story is called Baumol’s cost disease. The logic is elegant: some industries get more productive over time, some don’t. A factory that made 100 widgets per worker in 1960 might make 10,000 today. But a nurse can only care for so many patients. A teacher can only teach so many students.

When productive sectors bid up wages, labor-intensive sectors have to match them — even though their output per worker hasn’t changed. Services that require human attention get relentlessly more expensive relative to goods that can be automated.

Baumol explains education (you can’t automate a seminar), healthcare (surgery requires surgeons), and housing (construction is labor-intensive). Layer on the regulation story — zoning, environmental review, occupational licensing, NIMBYs — and the cost disease compounds.

This is all true. If this were the whole story, the solutions would be straightforward: automate what you can, cut the red tape, build more, train more workers.

But here’s what’s been nagging at me: Baumol explains why these services are expensive. It doesn’t fully explain why they feel so impossibly explosive. Something else is going on.

The Veblen Problem (The Half Nobody Talks About)

I started thinking about this differently after a conversation with my neighbor.

He drives a McLaren — the kind of car that costs over a million dollars. He’s on a waitlist to upgrade to the $2 million model. The waitlist is so long, he told me, that when he finally gets it, he can instantly resell and make a million dollars profit.

I kept thinking about what that means. The car isn’t worth $2 million because it costs $2 million to make. It’s worth $2 million because other people can’t have it. The scarcity is the product. And the “profit” my neighbor expects to make? That’s not value creation. That’s pure positional extraction — money transferred from someone even further down the waitlist who wants the status even more.

By standard productivity measures, McLaren looks incredibly productive. Revenue per worker: astronomical. But they’re not creating more utility. They’re manufacturing scarcity and harvesting status premium.

This is a Veblen good — named after Thorstein Veblen, the eccentric Norwegian-American economist who published The Theory of the Leisure Class in 1899. Veblen goods violate normal economic logic: the higher the price, the more you want it. A Rolex isn’t just a watch — it’s a signal that you can afford a Rolex. If Rolexes cost $50, no one would want them.

Here’s what we don’t talk about enough: the three-headed hydra — education, healthcare, housing — all have a massive Veblen component. And it’s entangled with the Baumol component in ways that make the problem much harder than the standard story suggests.

The Hydra, Re-Examined

Let me look at the same three sectors through fresh eyes.

EDUCATION

The Baumol story: Professors cost money. You can’t automate a classroom. Education is labor-intensive.

The Veblen story: Harvard admits about 2,000 freshmen per year. They could admit more. They could offer unlimited online enrollment. They don’t — because exclusivity is the product.

Harvard’s value isn’t primarily the education. It’s the credential, the signal, the scarcity. If everyone had a Harvard degree, no one would want one.

This cascades down. The frenzy over elite college admissions isn’t about learning calculus — any competent school can teach calculus. It’s about position. The “right” preschool leads to the “right” elementary school leads to the “right” college leads to the “right” life. Parents aren’t buying education; they’re buying slots in a tournament.

I have a friend — Chinese-American, incredibly driven — who went from chanting “defund the police” in 2020 to full MAGA by 2024. What flipped him? DEI. Specifically, his perception that elite college admissions were rigged against his kid. His grievance is real: he sees his child grinding through AP classes and test prep, watching the goalposts move.

But here’s the thing: even if admissions were perfectly “fair” by his definition, there would still be only 2,000 slots. The scarcity is the product. We’re all fighting viciously over positional goods while pretending we’re fighting about fairness.

HEALTHCARE

The Baumol story: Doctors cost money. Surgery requires surgeons. You can’t offshore an appendectomy.

The Veblen story: “The best” hospitals, “the top” doctors, concierge medicine, boutique practices — these are status goods. People pay premiums that have nothing to do with outcomes. Being treated at the Mayo Clinic or Mass General signals something about who you are, independent of whether you get better care.

The healthcare system is riddled with positional competition. The “best” pediatrician, the “top” specialist, the hospital with the prestige. Some of this correlates with quality. Much of it is pure signal.

HOUSING

The Baumol story: Construction is labor-intensive. Materials cost money. Permitting takes time. Regulations add friction.

The Veblen story: Mark Zuckerberg’s Palo Alto home is a street and a creek away from East Palo Alto. The property value difference? Probably 10x or more. The physical structures aren’t 10x different. The schools aren’t 10x better. The air isn’t 10x cleaner. The proximity to jobs is identical — it’s the same neighborhood, separated by a creek.

What costs 10x more? The address. The signal. The line on a map that says “Palo Alto” instead of “East Palo Alto.”

And it’s no accident that to the west of Palo Alto sits Stanford — the university that literally birthed the town, that created Silicon Valley, that mints the credentials which justify the housing prices which fund the schools which feed back into Stanford admissions. And Stanford Hospital? One of the most sought-after medical centers in the world.

Education, housing, healthcare — the three-headed hydra isn’t three separate problems. In Palo Alto, you can see it’s one creature, feeding on itself. The trifecta, complete.

And the contagion spreads. Even East Palo Alto — the “affordable” side of the creek — now has million-dollar homes. That’s the Veblen ratchet: proximity to status is itself a positional good.

This is a positional good in its purest form. It’s not just a place to live — it’s a signal that says: I made it. I’m in the club. My kids will be around other kids whose parents made it. The scarcity is doing work for the people already there.

I watch my Bay Area neighbors bid up houses not because the construction is exceptional, but because of what the address means. The NIMBYs fighting new development aren’t just protecting their views — they’re protecting their positional advantage. Every new house dilutes the exclusivity.

The Veblen component isn’t equally strong in all three heads. Housing in Palo Alto is drenched in it. Education sits in the middle — the learning is real, but the credential does heavy lifting. Healthcare is the weakest case — you go to the nearest ER when you break your leg, not the one with the best brand. But the prestige tiers exist in all three. The mix varies; the dynamic is present.

The Entanglement

Here’s what makes this genuinely hard: you can’t cleanly separate Baumol from Veblen. They’re not just mixed together — they feed each other.

Take college tuition. How much of that 180% increase is Baumol (professors cost more, campuses need maintenance) and how much is Veblen (schools competing on prestige, climbing walls, fancy dorms, administrative bloat driven by rankings)?

The honest answer: we don’t know. But it’s worse than that. The two dynamics are reflexive — each one amplifies the other.

Baumol makes education expensive → expensive becomes a signal of quality → schools compete on price as a status marker → tuition rises beyond what Baumol alone would predict → the high price becomes the product → which requires more administrators and amenities to justify → which feeds back into Baumol costs.

The same loop runs in housing. Construction costs rise (Baumol) → housing gets expensive → expensive neighborhoods become status markers (Veblen) → residents fight to restrict supply to protect their status investment → restricted supply drives prices up further → which attracts more status-seeking buyers → which justifies more restrictions. The dance goes on.

And healthcare. Labor costs rise (Baumol) → healthcare gets expensive → expensive care becomes a signal of quality (Veblen) → providers compete on prestige rather than outcomes → prestige requires more specialists, more equipment, more marble lobbies → which drives Baumol costs higher → which makes the status signal stronger.

It’s not Baumol plus Veblen. It’s Baumol times Veblen — a multiplicative spiral where each dynamic reinforces the other.

We talk as if it’s 90% Baumol and 10% other. But what if the interaction effect is larger than either component alone? What if the real driver is the feedback loop between them?

We’d be trying to engineer our way out of a problem that’s fundamentally about human psychology and social competition — while our engineering solutions get absorbed into the status game and make it worse.

The Policy Trap

This leads to a predictable policy failure — one that transcends party lines: restrict supply, then subsidize demand.

Student loans make college “affordable” — while tuition skyrockets. Housing vouchers help renters — while NIMBYs block construction. Healthcare subsidies expand coverage — while costs spiral. The pattern repeats everywhere.

And both parties play the same game. Governor Newsom sends $18 billion in “inflation relief” checks to Californians — the same California where zoning laws make it nearly impossible to build housing. President Trump directs Fannie Mae and Freddie Mac to buy $200 billion in mortgage bonds to lower rates — while tariffs raise construction costs. Lower rates mean more demand chasing the same constrained supply. The check gets cashed; the prices go up; the next election demands another check.

If the problem were purely Baumol, demand subsidies might eventually create pressure to expand supply. More students can pay, so maybe schools expand capacity.

But when you subsidize demand for a Veblen good, you’re not helping people afford the thing. You’re funding the positional arms race. More money chasing the same 2,000 Harvard slots. More money chasing the same “good school district.” The prices rise, the exclusivity is preserved, and the subsidy becomes a transfer to whoever controls the scarce supply.

And they have every incentive to keep it scarce.

The cycle is almost comedic: prices go up, people get angry, politicians subsidize demand, prices go up more, people get angrier, politicians subsidize demand more. Everyone knows it’s just a temporary relief. But confronting the supply side — the Baumol regulations and the Veblen status game — is politically harder than writing another check.

The AI Question

This brings us to the present moment. AI is the most significant technological change since the internet, maybe since electricity. Techno-optimists promise it will finally solve Baumol’s cost disease — automating services that have resisted automation for decades.

I listened to Marc Andreessen on a podcast recently, making the case that if AI really works, “productivity skyrockets and prices collapse.” This sounds compelling until you think about it for five minutes.

AI tutors, AI diagnostics, AI legal services — yes, these could genuinely reduce the labor intensity of services that have been stubbornly expensive. The Baumol part might actually get solved.

But here’s the question nobody’s asking: what happens to the Veblen component?

If AI makes basic education effectively free, does that reduce the cost of education? Or does it just intensify the positional competition for whatever education AI can’t provide — the elite institutions, the human-taught seminars, the credentials that signal “I didn’t need the AI”?

If AI makes basic healthcare abundant, do we all get healthier? Or does the positional competition migrate to “premium” care, human doctors, boutique medicine that signals you can afford to not use the AI?

Here’s a chilling thought that hit me as I was writing this: perhaps the reason AI won’t destroy all our jobs is itself a Veblen phenomenon. Jobs might survive not because humans are needed, but because human labor becomes a status signal.

“Handmade.” “Artisanal.” “Human-taught.” “Seen by a real doctor.” The scarcity of human attention becomes the new luxury good. The AI can write the legal brief, but you hire the human lawyer because you can afford to. Work becomes a performance of status rather than a source of value.

This isn’t comforting. It means jobs are “saved” but hollowed out. It means AI abundance for the masses, human service for the elite. The Veblen ratchet tightens.

My worry is this: AI might solve Baumol while accelerating Veblen. We’d get massive productivity gains that show up in the statistics, while people feel just as squeezed — because the status competition has shifted to new terrain.

The treadmill doesn’t stop. It just gets faster.

So Now What?

I’ve laid out the diagnosis: the three-headed hydra of education, healthcare, and housing isn’t just a Baumol problem (services are hard to automate) or a regulation problem (we’ve made it hard to build). It’s also a Veblen problem (we’re competing for position, and scarcity is the product). The two dynamics are entangled, reflexive, and our policy interventions keep making it worse.

But diagnosis isn’t cure. And here’s where it gets uncomfortable.

The Veblen problem implicates us. Not just “the system” or “the elites” or “the regulators” — us. The homeowner fighting new development. The parent clawing for the “right” school. The patient demanding the “best” hospital. We’re all playing the game, often without seeing it.

I live in Palo Alto. I see the trifecta every day — Stanford, Stanford Hospital, the houses across the creek from East Palo Alto that cost ten times as much for reasons that have nothing to do with the physical structures. My neighbors and I have bid up these prices. That’s what makes this hard to write about.

Ezra Klein and Derek Thompson’s Abundance movement is right that we need to build more. But abundance alone won’t solve a problem that’s partly about human nature — our deep need for status, for position, for being ahead.

What would it mean to actually see the Veblen dynamic clearly? To name it in our politics, our policies, our own choices? To decouple status competition from the essentials of a decent life?

That’s where this gets political. And that’s Part 2.

Next: The politics of positional scarcity — why we fight about DEI instead of Harvard’s 2,000 slots, why blue states are “failing” at success, and what a politics that takes Veblen seriously might look like.

If you’ve ideas what should be in Part 2, please drop a line!

I was taking a short solo ski trip and driving our only family car. My wife was a little anxious and warned me, “don’t do anything stupid, I won’t be able to visit you in the hospital”

There’s something about skiing I can’t resist. I’m in awe of the majestic vistas and mono-chromatic landscapes. The solitude of the soft white snow that dissipates everything including my anxieties.

But there’s also the sense of unknown, adventure and challenge. The gratifying sense of achievement when I traverse a particularly tricky run or discover new terrain.

And so, I found myself throwing Elaine’s cautionary warning to the wind and inexplicably turning left to an unexplored double black run. It all started so well, Indigo Girls jamming with me. My AI coach Carv was encouraging me with pleasant pings whenever I made a particularly smooth parallel turn.

Then, all of sudden, I don’t remember the details: maybe I suddenly panicked at my increasing speed, or I met with an unexpected bump. I careened out of control. I must have fallen 20-30 vertical feet before my boots finally provided enough traction

I looked up and only then realized my skis were way above me, poles who knows where. The surface beneath the thin layer of snow was pure ice — every time I tried to climb back up, I just slid down again.

Then a stranger appeared. He side-slipped down to my skis, collected them, made his way over. I thanked him, tried to stand — and slid another five feet. He picked everything up again. No impatience, no fuss.

“Same thing happened to me yesterday - it sucks!” he said.

At his suggestion, I stopped fighting. Instead of scrambling up, I slid down with my gear to a flatter spot. What had been terrifying suddenly wasn’t.

He skied off with a wave. I never got his name. But I’m still carrying something — 4 days later — from that moment: a glow of gratitude I can’t quite shake. A debt I can only pay forward.

Kevin Kelly has a piece called “How Will the Miracle Happen Today?” It’s not another exhortation to be kind. His insight runs the other way: for kindness to flow, you need people willing to receive it. He calls them “kindees.”

Kelly spent his twenties hitchhiking to work, traveling through Asia on the generosity of strangers. Kindness always showed up, he noticed — but only when he was open to it. The problem is most of us aren’t. We armor up. Wave people off. “I’m fine, I’ve got it” — while sliding on ice.

Here’s what struck me: kindness has supply and demand. All our cultural energy goes to the supply side — be kind, give more, pay it forward. But demand matters too. If everyone’s self-sufficient and armored, all that kindness has nowhere to land. The stranger who helped me needed me to accept his help. His gift from yesterday would have dead-ended otherwise.

Being a good kindee isn’t weakness. It’s the other half of the economy.

Maybe that’s why I love skiing. And why I love startups. Both put you in exposed positions. You can’t fake competence when your gear is scattered across the slope. And that exposure — the thing we’re taught to minimize — is what makes connection possible.

Skiing has this duality I can’t resist. The solitude of the summit, the white silence, everything including your anxieties absorbed by the snow. But also an invisible web of strangers who’ll stop for you, because they’ve all been humbled by the same mountain.

Elaine was right to worry. But also not quite. The mountain isn’t just where stupid things happen. It’s where strangers help — because yesterday, they were you.

I’m glad the hitchhiker let me pull over.

==============

I first told this story ’ weekly story gym. And in the telling I realized that this was why I love his story gyms so much: he has architected a space where we are open to kindness by sharing our vulnerable stories, and in turn we receive kindness from the feedback and connections. Thank you , , and and of course my ski angel for your kindness.

=========

My Why: The Photocopier Kid Returns

Some geeks recall fondly their times spent in bookshops. I spent my teenage years in copy shops. No, not coffee shops. But shops with photocopiers. I hung out there because I was a total Apple ][ geek. You see, most books and magazines about Apple computers would take months if not years to arrive in then backwater Singapore. And when it did arrive, it was marked up 3x at a time when everything else in Singapore cost far less.

In a harshly fluorescent-lit room stacked with machines zipping their own fluorescent lights, I remember copying the photocopy of the photocopy of one book: Beneath Apple DOS. That book opened a whole new technical universe I did not know existed. It gave me a sense of mastery I had never felt as a young teen, perhaps like being admitted to a secret guild of times yonder.

I was inspired, captivated and enthralled by the Apple story. A club for nerds just like me called the Homebrew Computer Club, dreaming a crazy vision of a computer in every home. Two guys working in a garage. And then seemingly instant success. The stunning creation story of the Macintosh and world's most famous 1984 ad to introduce it. The seduction of Pepsi executive John Sculley to be CEO with the famous line "do you want to sell sugar water for the rest of your life?" - only to have Sculley kick out co-founder Steve Jobs in a boardroom coup. Jobs goes on to buy a computer graphics company called Pixar as a side project but in a pique pours his wealth to plot his revenge with a new company called NeXT. The other co-founder, Steve Wozniak, takes a different path - spending his money on rock concerts (the US Festivals), going back to UC Berkeley, and nearly dying in a plane crash.

And all this is just the barest skeleton of the first decade of Apple. As I got to know the textures of the stories over the years, the more enraptured and obsessed I became. The story of Apple is the story of our lives and times. It deserves to be retold, especially in the age of AI. It feels right that the children of Apple, both products and its people, embark on a quest to build an Apple library.

My What: An Apple Library for the AI Age

I want to create an Apple Library/Museum/Encyclopedia (you can see I am still fuzzy in the details) that explores and experiments with the latest AI toys and techniques. This could be a mammoth undertaking that would totally overwhelm me. So instead, I want to start with small, perhaps disparate experiments that connect to this large theme.

Here's what I'm committing to explore in the next five weeks:

1. Dissecting Apple in China - This amazing book convincingly argues that Apple is the single most important company responsible for industrializing modern China. I want to dissect each chapter, extract every story, character, product and company mentioned, and put it all on a timeline. If I do this for many books and sites (I love folklore.org), what emerges?

2. Connecting the Dots Newsletter - Track current news and discussions about Apple. Create a daily or weekly summary that ties current news to old Apple stories. I've started with this teaser piece. I need to write a few more “manually”, and then see if I can systematize the process

3. AI-Generated Retellings - Can the old Apple stories be retold with new AI-generated media? What new ways of storytelling can we create? I am so inspired by @Lori’s song. Wouldn’t a ballad of the two Steves be just grand?

My How: Overcoming the Obstacles

suggested I write down my fears so they are legible:

• There is so much about Apple out there, what value can I add?

• This is such a big project, where can I start?

• There are practical, potential copyright issues

• Who is my target audience? Is this just for myself?

Here's how I'm thinking about these challenges:

When something is overwhelming, start small. I have a few tiny, maybe disparate concrete ideas I can implement right now. I should start on those.

As I learned from Tasshin today, there are no right answers, only experiments and discoveries. Always be on the prowl for surprises. Be on the lookout for the unexpected. That has always been my saving light.

For copyright issues, I'm heartened by recent court rulings re: AI training being considered transformative works. Still, this is an area I should be aware of but perhaps not make it a blocker. It's unlikely someone is going to sue a personal project, so if I ever have to worry about it, that's probably a good sign. What is clear is that I should purchase every book I use in the Apple library, and aim to make every piece of work “transformational” and cite the source material

I am doing this mainly for myself, to dive into the joys of being an Apple fan and to relive lush, rich stories of Apple. That motivation is enough in itself. If others find the journey interesting, then so much the better.

But a library is meant to be shared, so of course in another sense I do wish for the library to be used and cherished by all. So, this is the delicate tightrope I am still feeling out. Perhaps, my fellow Act Two-ers can help here.

The Fantastical Measurable Moment

suggested one fun way to gauge the success of this project: I get invited to an actual Apple keynote event (just like I fantasized in my last piece). That would be the full circle moment - from the photocopier kid in Singapore to standing in the Steve Jobs Theater.

The next five weeks aren't about building the complete library. They're about discovering what this library wants to become. Time to stop photocopying and start creating.

chao

September 5, 2025

Thanks also to , and John for the helpful feedback.

The first thing you need to understand about Apple keynotes is that nothing is left to chance. Absolutely nothing. I learned this standing in the wings of the Steve Jobs Theater at 8:47 a.m. on a Tuesday in September, watching a lighting engineer adjust a single spot for the third time because Tim Cook's forehead had developed what she called "an unacceptable gradient" during the dress rehearsal. The engineer, flown in from Los Angeles at presumably absurd expense, was explaining to her assistant that human perception registers even a two-percent variance in color temperature when projected on a sixty-foot screen. The assistant was nodding like this made perfect sense. Cook was somewhere in the building, probably reviewing his marks for the hundredth time. In three hours, a thousand journalists would file into this room expecting magic. But right now, it was just a woman with a light meter declaring war on a shadow that nobody else could see.

This is what Apple never shows you: the industrial complex behind the illusion.

The Death of Jazz

There was a time when these things could actually go wrong.

May 2002, San Jose. I'm watching Steve Jobs wheel an actual coffin onto the stage at WWDC. There's organ music. There's a eulogy. He's conducting a funeral for OS 9, and not metaphorically. He's got the box, the flowers, the whole production. The developers in the audience don't know whether to laugh or applaud. Half of them still have OS 9 machines in their backpacks. Jobs doesn't care. He's already living in tomorrow, and he's dragging five thousand engineers with him. But here's the thing: when he pressed play on that organ music, there was a solid three seconds where nobody knew if the audio would actually work. Not Jobs, not the sound guy, not the ghost of OS 9. You could see it in his shoulders, the tiny tension before the speakers caught.

That doesn't happen anymore.

The last time an Apple demo contained genuine jeopardy might have been iPhone 4, June 2010. Jobs is trying to load the New York Times website and the Wi-Fi is dying. Not struggling. Dying. There are over 500 hotspots in the room because every tech blogger thinks their liveblog is the center of the universe. Jobs stops mid-sentence. He's actually asking journalists to turn off their laptops. "If you want to see the demo," he says, and there's this edge in his voice, "we need to shut down the networks." For thirty seconds, the most powerful man in tech is essentially pleading with a room full of bloggers. The demo eventually works, barely. But those thirty seconds? They were real. Human. Terrifying if you owned Apple stock. Looking back now, knowing how sick he already was, the moment hits differently. A year from his resignation, sixteen months from the end. The control freak losing control. The perfectionist caught in imperfection. The man who commanded the future suddenly, visibly mortal.

Phil Schiller understood this better than anyone. In 1999, this absolute madman proves AirPort wireless by jumping off a twenty-foot platform while clutching an iBook, streaming the whole way down. No net. Well, technically there was an air mattress, but at that height with that much hardware, the difference is academic. The beauty wasn't that it worked. The beauty was that everyone, Schiller included, knew it might not. He's grinning on the way down like a kid jumping into a pool. That grin is worth more than a thousand perfect demos because it's the grin of someone who genuinely doesn't know what happens next.

The Architecture of Control

Fast forward to September 2014. Apple doesn't just rent the Flint Center in Cupertino. They construct a separate building next to it. Not a tent. A building. Three stories, white walls, covered in fabric so nobody can photograph it. The press walks by this thing for a week beforehand, trying to figure out what's inside. Security guards who look like they've been borrowed from a Bond film stand at every corner. Nobody says anything.

After Cook reveals the Apple Watch, we're herded into this mystery box. Inside, it's a custom-built demo pavilion with perfect lighting, perfect temperature, perfect sight lines. Every watch sits at the exact same angle. The bands are arranged by color gradients that would make a Pantone executive weep. This isn't a demo space. It's a surgical theater where the patient is desire itself.

An Apple PR handler walks me through the approved photo angles. Let's call her Sarah, though they're all somehow named Sarah. She doesn't actually say "approved photo angles." She says, "You'll probably get your best shots from here," while physically positioning my shoulders at a thirty-degree angle to the display. Every journalist in the room is being simultaneously positioned by their own Sarah. We're not covering an event. We're participating in a choreography so precise it makes the Bolshoi look improvisational.

By 2017, Apple stopped renting other people's stages. The Steve Jobs Theater emerged from a hill like a magic trick: glass cylinder, carbon-fiber lid, an elevator that rotates as it rises. It's not a venue. It's a thesis about control.

The Pandemic Kills the Jazz

COVID changed everything by removing the one element Apple couldn't completely control: us.

March 2020. Apple's video team suddenly becomes the entire show. They normally handle the pretty B-roll of iPhones spinning in space. Now there's no audience. No energy. No possibility of failure. Just Craig Federighi walking through an empty Apple Park, hitting his marks with the precision of someone who's done this take seventeen times. Every executive appears in a perfectly lit space that looks like it was designed by aliens who studied human aesthetics but never experienced joy.

The production values are insane. We're talking about budgets that could fund small films. But here's what's missing: the moment before the demo where you wonder if it'll work. The nervous energy when someone's about to announce pricing. The way five thousand people holding their breath creates a vacuum that pulls the future forward.

A producer who worked on one of these COVID keynotes told me they did forty-seven takes of a single transition between two presenters. Forty-seven. For a walk through a doorway. The twenty-third take was perfect, but they kept going because someone thought the shadow on the door frame could be "more intentional." That's where we are now. We're intentionalizing shadows.

The funniest part? Once Apple proved they didn't need us in the room, they couldn't quite bring themselves to invite us back, not fully. The hybrid model now is like dubbing a laugh track onto a prestige drama. Pre-recorded keynote with live audience reactions. We clap when we're supposed to clap. We gasp at the approved gasping moments. But everyone knows the show was locked weeks ago.

The Hands-On Hustle

After every keynote comes the part civilians never see: the scrum.

Picture this: a thousand journalists, all of whom believe they're the protagonist of tech journalism, suddenly released into a room with fifty iPhone stations. It's like Black Friday for people with journalism degrees. Elbows fly. Photographers literally climb on furniture. I once saw a YouTube influencer with thirty million subscribers get bodychecked by a blogger from a site that reviews calculator apps. Turns out 30 million followers does not actually give you any powers in the real world.

The Sarahs are everywhere, managing traffic flow with the efficiency of air traffic controllers. "You've had your time," one tells me after I've been testing a camera for exactly ninety seconds. She's smiling, but it's the smile of someone who could have you removed by security and is letting you know she's choosing not to. Yet.

Every detail in this room has been war-gamed. The tables are the exact height where you don't need to bend to see the screens but can't sit down and monopolize a station. The lighting makes every product look like it was blessed by the gods but eliminates shadows that might confuse the camera. The demo units are running special software that resets every three minutes, erasing all evidence of the previous journalist's ham-fisted attempts to break things.

I'm photographing an iPhone 16 Pro when I notice something: the demo loop showing on the screen favors bright, evenly lit frames. No dark scenes. No extreme contrasts. Nothing that might look bad in the inevitable thousand unboxing videos that'll drop within hours. Even the accidents have been prevented.

The Thing About Perfection

Here's what kills me: I miss when these things could fall apart.

I miss Schiller clutching that iBook like his life depended on it. Because his career kind of did. I miss Jobs turning his Wi-Fi catastrophe into a teachable moment about network management. I miss the firmware updates pushed at 2 a.m. the night before because someone found a bug that would have bricked the demo units. Those weren't flaws in the system. They were proof the system involved humans.

The modern keynote is a marvel of planning, production, and precision. Every sentence has been focus-grouped. Every transition has been tested on a sample audience of employees' families. Every possible failure has been engineered out of existence. They're better in every measurable way. Clearer, prettier, more informationally dense. They're also completely devoid of the thing that made the old ones matter: the possibility that someone might fuck up.

But maybe I'm just being nostalgic for chaos. The old keynotes weren't actually good for anyone's cardiac health. Engineers would literally sleep at the office for the week before, debugging demo code. PR teams would have trauma flashbacks to failed demos years later. One former Apple employee told me he still can't hear the iPhone ringtone without remembering a demo that crashed seventeen times during rehearsal.

The Mirror Has Two Faces

And yet this is what haunts me: I'm writing this piece on my third draft, having run it through grammar checkers, style guides, and yes, AI assistants who help me find better metaphors. I lament the death of spontaneity while carefully engineering every sentence. I mourn the loss of risk while refusing to publish anything that hasn't been edited within an inch of its life.

Am I really criticizing Apple for doing exactly what I'm doing? Trading the random beauty of imperfection for the guaranteed competence of polish?

We're all complicit in this. We say we want authenticity, then we watch the blooper reels on YouTube later, from the safety of knowing how things turned out. We claim to miss the humanity, but we'd crucify any company that shipped a product as buggy as those early demos. We've all agreed to pretend we want jazz while actually preferring Spotify's algorithm.

The AI that's helping me edit this would have prevented Jobs's Wi-Fi disaster. It also would have prevented his "one more thing" spontaneity, because spontaneity doesn't A/B test well. We're living in a world where everything is getting better at being good and worse at being interesting. The keynotes are just the canary in the coal mine. If canaries were designed in California, machined from a single block of aluminum, and available in five colors.

One More Thing...

I've been an Apple fanboy ever since I laid my eyes on an Apple ][ as an awkward teen in a then third world country. Just last week, my friend and I were reminiscing that neither he nor I would have moved from Southeast Asia to Silicon Valley if we didn't fully buy into the Apple mythology.

So when I heard that Apple had just announced their highly anticipated but also expected fall event, it gave me the idea it would be fun to have a retrospective of how all these Apple keynotes have evolved over time. But why should I, as a mere fanboy, be the one to write this?

That's when I thought it'd be fun to pretend that I was an intrepid journalist writer like Michael Lewis, or the well connected insider Steven Levy or maybe the deep tech personality whisperer .

I enjoyed generating this piece fulfilling my fantasies. We live in weird times, and I am embracing it in full-throated fashion. Here's to a fun-filled Apple Fall Event 2025! Full of surprises, risks and authentic Awe dropping moments...

Thanks to and for the feedback and fun chats!

========

Here are the bloopers and proof that “One More Thing…” does not A/B well:

I have been fascinated with OODA loops and flywheels recently. I have this gut feeling that they are a key unlock for any limited successes that I have enjoyed. It's too pat to say the goal of a startup is to attain Product Market Fit (PMF). A startup is more like a quest for flywheels. Each turn makes you faster, stronger. Over time, they build durable compounding advantages.

The funny thing is my obsession with flywheels has led to hallucinations. As best as I can tell, the Bezos napkin isn't in the Smithsonian and the napkin itself may be apocryphal.

I have struggled to write this piece because at the back of my head, I have this fear: is this all post-hoc rationalization? A form of survivorship bias?

Is it? You be the judge.

Which brings me to why I'm staring at our Discord chatbot's metrics again.

We vibe-coded a bot persona for Reframe Science, a healthcare self-help community. The idea seemed straightforward enough: surface answers from past discussions so people don't keep asking the same questions over and over. Each interaction should make it smarter. Better answers → more trust → more usage → more data → even better answers.

Sounds great on paper, right? But here's the thing that's keeping me up: is this the loop that actually matters? Or am I just building a really sophisticated hamster wheel?

That's the thing about loops. They're seductive. The numbers tick up (maybe?), you feel productive, the team gets excited. But sometimes you're just spinning toward the wrong destination. And when you're obsessed with flywheels, every rotation starts to look meaningful.

The long road to PMF

Product-market fit doesn't arrive with a push notification. It sneaks up on you through loops.

Most loops are learning loops: you grind, you iterate, you collect clues every time around. They take effort, sometimes a lot, but they sharpen your instincts.

Occasionally — and this is where the magic happens — the grind shifts. Each turn gets easier. The wheel starts pulling instead of you pushing. That’s the moment a loop becomes a flywheel.

The trick isn't avoiding the grind. It's finding the smallest loop that can start pulling. Or at least, that's what I tell myself when I'm staring at those chatbot metrics.

What Hannah Montana taught me

My first real flywheel started with a weird blip in our search logs.

"Hannah Montana backgrounds."

Not just a few searches. Hundreds. We were running an image search engine, and suddenly every tween with a MySpace page needed young Miley Cyrus' wallpaper.

So we threw together a landing page. Nothing fancy—just Hannah Montana backgrounds our users had already handpicked. And then I watched the numbers start curving the chart.

The page climbed Google rankings → brought in more Hannah Montana fans → some of them searched for other backgrounds → we made pages for those → which brought in even more people.

Each turn was easier than the last. We weren't pushing anymore; the loop was pulling.

At our peak, we were updating someone's myspace page more than once every second every day.

I didn't have a fancy name for it then. But that was my first content flywheel, and it started as a throwaway experiment.

The Snag Bar

A couple years later, different loop, same magic. Maybe.

Remember FarmVille? Millions of people clicking through endless bonuses from friends. Fun for about five minutes, then pure grind. Digital farming as a second job.

We built a tool, Snag Bar, that let our users collect all their preferred bonuses with one click. Suddenly our users had an edge—more rewards, faster progress, zero hassle.

But here's where it got interesting. The bonuses often included gifts to send back. So every Snag Bar user made their friends wonder "wait, how are they collecting so fast?"

More users → more shared bonuses → more value for everyone → more users wanting in.

The product literally got better as more people used it. Hannah Montana was powered by content. The Snag Bar was powered by connections..

Thread Reader

Most of my loops I built from scratch. Thread Reader I bought mid-spin.

It solved a real problem: making Twitter threads readable. But the magic was in the mechanism. To unroll a thread, you replied "@threadreaderapp unroll" publicly. Everyone watching saw you use it. Genius, right?

More requests → more visible unrolls → more people discovering it → more requests.

I could see it immediately. The flywheel was already spinning. My job was just to oil it and get out of the way.

Did I see a flywheel because it was there, or because I'd been looking for one?

Three different flywheels, three different mechanisms. Content, connections, and now distribution baked right into the product itself. Each one taught me something about what to look for.

The Four Questions

After watching enough loops spin (and enough sputter to a halt), I've learned to run them through four gut-checks. They sound simple, but skipping one can kill your momentum.

1. The Compass — Are we even heading the right way?

The Discord bot could spin beautifully… straight into irrelevance. Hannah Montana backgrounds were absurd on the surface, but they pulled the right audience at the right moment. Picking the wrong loop is like pedaling hard toward a cliff.

2. The Map — What's each turn teaching us?

A loop with no new insight is just a workout machine. When we saw people searching for MySpace themes to match their Hannah Montana wallpapers, that told us where to go next. If nothing changes in your plan after a few turns, you're not in a loop—you're in a rut.

3. The Pedals — How quickly can we make the next turn?

FarmVille bonuses dropped daily. Our Snag Bar shipped updates just as fast. Momentum dies if your loop waits for quarterly releases—speed is oxygen. The more cycles you fit in, the more chances you give the loop to strengthen.

4. The Oil Can — Can we clear friction before it kills us?

Every loop builds pressure. Hannah Montana nearly melted our servers; Snag Bar nearly drowned us in support requests. If you can't remove bottlenecks fast enough, the loop slows, coughs, and stalls.

Here's where the strange loop comes in. The more you work through these questions, the more you zoom out—seeing the loop from above—only to realize you're right back inside it, pedaling. Except now you understand it differently. You're both the rider and the mechanic, the subject and the observer. And sometimes, that shift in perspective is what unlocks the next turn.

That’s what using Thread Reader felt for the first time. Wow, summoning a bot to create a page is that easy? Oh, everybody gets to see it too? What a great technique! Who is the genius that built this? I want to learn from him.

These aren't four separate levers—they multiply. Get a zero in any one, and the whole thing collapses. But bump even one up, and everything else gets easier. Right loop + fast learning + quick turns + no friction? That's when a loop stops feeling like work and starts feeling like magic.

Can you design a flywheel from scratch?

Here's a thought that's been nagging me: instead of waiting to discover a flywheel, what if you hypothesized it first?

Map out the loop you think can compound. Test if it can even start spinning. See if the physics (or economics/psychology) actually works that way.

I've tried this. On paper, the loops look perfect—arrows pointing to arrows, virtuous cycles everywhere. In practice, some never catch while the dumbest-looking ones suddenly take off.

Still, asking "what would need to be true for this to become a flywheel?" has saved me from chasing dead-end zero progress loops. It forces you to see the system, not just the feature.

The three signs a loop is about to catch

You know that moment, when learning to ride a bike, you stop wobbling and start gliding? That transition from forced effort to natural motion? Loops have that moment too.

Transfer: Each pass leaves something valuable behind

The Discord bot starts with generic responses. But each interaction teaches it the community's language, their trusted sources, their recurring problems. Templates evolve from one-offs to reusable assets. You feel the friction dropping.

Acceleration: Time between cycles compresses naturally

Not because you're pushing harder—because the loop is doing more work. Monthly releases become weekly, then daily, then continuous. When every piece is already spinning, new experiments can catch fire overnight.

Distribution baked into the product

This is the holy grail. Thread Reader showed me this - every unroll created a public "look at this cool tool!" moment. The Snag Bar turned every bonus into a gift that advertised what friends were missing.

When two of these show up together, pay attention. The wheel has probably caught.

For Reframe Science, we're still watching. Smart answers alone aren't enough. But if those answers also train the next response faster, compress time-to-solution, and create shareable artifacts that bring in newcomers? That's when we'll know.

The power of seeing the loop you're in

Will the hamster still run if it knows how the wheel works?

We do - usually on autopilot. We grind away without realizing we're in a loop at all—just shipping features and wondering why growth feels random.

But a loop you can see? You can test it, measure it, accelerate it. You can ask: where's the friction, what would make this turn faster? That awareness—being able to name the mechanics—often separates grinding from compounding.

Closing the loop

Back to those Discord metrics. Maybe we're building our flywheel. Maybe it's just a fancy hamster wheel. Maybe I'm so obsessed with flywheels that I'm seeing them in my coffee swirls.

I'll know it when I feel it—that transition from pushing to pulling, from forcing to flowing. I've felt it before.

Or at least I think I have.

You be the judge.

How This Got Written (A Loop)

As regular readers already know, I’ve been experimenting with ways to fold AI into my writing (and life).

This was my most significant attempt. I have struggled thinking about this concept of loops and flywheels for months (as can attest). But I felt unable to turn this into a coherent idea by myself.

Early on, I just jotted thoughts and notes around the subject. But they were disparate and undirected.

After a fun conversation with Louie, and coincidentally right after ChatGPT 5 was released, I finally picked up enough motivation to feed gpt5 with all my diffused notes and ask it to help me come up with a through line and structure.

Conversations with Christin and Louie were my indispensable learning loops to put down my raw, random, intuitions down. ChatGPT helped me gain enough momentum to start the flywheel towards publishing.

I actually spent more time on this than any of my earlier LLM-less pieces. But this one feels more ambitious. Despite taking a dim view on AI-writing, the best editor I know, the ever tender and helpful , gave me really piercing feedback on an early draft. She flagged how the Reframe reference felt disconnected and disjointed. An earlier me would have thrown up my arms in despair. AI-assisted me came through only a little scathed.

I translated Sandra’s notes into specific problems; ChatGPT offered several options. I tried a few and ended up melding a composite of two.

For the first time I enlisted not one, not two, but three AIs. Perplexity grounded my scant research and handled some fact-checking (including convincing me that my Bezos napkin being in the Smithsonian was a hallucination). I often compared responses between Claude and ChatGPT. Louie would probably not be surprised that I found Claude was better at style-transfer, whenever I had trouble with a phrase or a retelling, Claude’s suggestions almost always felt truer to how I would have expressed (if only I had thought of it - a strange loop!)

I am sure many of you may find this form of writing distasteful or fake. But please bear with me, I feel I am just at the initial turns of a fabulous flywheel! I am feeling my way through something more like curating than writing. This lets me run bigger experiments but of course I am losing something too.

AI helps me write and connect ideas faster. But, as Louie warns it can also add wrap shit with bling. Where are the levers for the right win-win?

Twenty years ago, I was already yapping that, in the 21st century, we’ll be managing bots rather than people. We’re living in that world now.

Thanks also to , and for their support and feedback.

I started in the most private way possible—doing the routine in my bedroom. It felt like a safe, hidden experiment. Over time, as I got more comfortable and confident, I graduated to the small deck just outside my bedroom. It was a tiny step out of my comfort zone, but it felt huge at the time.

As the months and years went by, I found myself rolling out my mat in the neighborhood park. What once seemed unimaginable—exercising in public—became second nature. I even started taking my 7-minute routine with me on trips, like exercising on a pier literally in the middle of Tomales Bay or the memorable week in Hualien, Taiwan, surrounded by strange rock formations and unbelievably loud cricket calls.

Today, I did it just outside our AirBnB in Healdsburg. People walking dogs or heading downtown either stared at the crazy guy jumping up and down or avoided eye contact as I rotated out of a push-up. I’d like to say I’m unfazed, but I’m still a little self-conscious.

Nevertheless, the workout has become a daily morning ritual; if not done, I would feel something nagging at me or maybe that the day has an odd shaped hole. I’m fitter, have more stamina and energy, and way prefer walking and biking to driving for any number of daily activities.

If you’d told me twelve years ago that I’d be the person confidently working out in public, I’d have laughed. But here I am, and it’s one of the few habits that’s truly stuck with me and reshaped my life in ways I never expected.

=======

This little piece appeared first in Reframe Science, a curated community for entrepreneurs, builders, and high-agency individuals taking charge of healthcare for themselves and their loved ones started by my friend and fellow entrepreneur Chrstin. I wonder which experiment I’m sharing with Reframers now will quietly reshape my next twelve years. Maybe it’s already begun.

Since Reframe is playing with Ernst Haekel art in its branding, I thought I’d ask ChatGPT to help come up with the right imagery for this piece:

Well, ChatGPT does know I love biking and mushrooms!

==========

And here’s how ChatGPT helped/wrote this piece:

“California is a dual consent state, this is illegal”

I had screwed up. And now I felt like I was getting pushed out of my own tribe.

A week earlier, I was rapt in anticipation, eagerly awaiting a pendant I had ordered over a year ago. This wasn’t any ordinary pendant. Whenever I wore it, it would record everything it heard and later sync it to my phone. Moreover, it was connected to one of these new-fangled AI large language models, so that I could ask an AI about anything that was recorded.

“How many kids did Rick say he has? I think I met him last week”

“Remind me of the action items from our design meeting”

I have tried many note-taking and information organization software programs that purported to be your second brain, but this seemed more plausible than previous attempts.

The strange thing about this new gadget I’d been looking forward to was that it kind of just disappeared. I would wear it clipped to the top of my T-shirt, but then after a while I wouldn’t even remember I had it on. And so, when a group of us ex-Write of Passage classmates went on a long-anticipated hike, I forgot I was wearing the pendant.

What I should have done was announce to the group that I was wearing the pendant and anything anyone said would be recorded for my personal use. But, alas, I didn’t.

Oops.

It was only when I started talking about gadgets with a fellow hiker that I recalled I was wearing the pendant. I was too socially constipated at that time to announce to the whole group at that moment about my pendant transgression. I finally plucked up the courage to reveal it to the group at the end of the hike, but by then the damage was done, and I deservedly received the shock, scorn, and dismay my fellow writer friends heaped on me.

This was not my first brush with being a social outcast due to my love of cutting-edge tech gizmos. Over 10 years ago, I was one of the first "glassholes."

In 2013, Google Glass was released with much fanfare. It could take pictures and record videos. It even had a tiny screen to offer relevant info like walking directions. Unfortunately, it was largely panned due to privacy concerns and high costs. Folks like me were branded "Glassholes." I recall feeling very self-conscious wearing the device as I went about my daily ordinary life.

But fast forward to the present, and I am happily wearing my Meta Ray-Bans almost every day. As I walk around the streets of Palo Alto listening to my favorite podcasts via these glasses, I don’t even give it a second thought, and, more importantly, no one else does either. There has been no public outcry, even though these glasses take much better pictures and record videos with a much-improved camera and five microphones. In fact, a year after I wrote my love letter about them, they have become increasingly trendy and surprised even Meta/Facebook by becoming their most popular hardware gadget.

This got me thinking that what was once socially unacceptable, or at least frowned upon behavior, has become popular and trendy in just a span of a decade.

That the change in perception is so fast is one reason why it’s so hard to legislate privacy. The European Union has largely been panned for its GDPR, or General Data Protection Regulation. A manifestation of this regulation we are all familiar with is the annoying “cookie consent” dialog boxes we see on almost every website.

How many of us just tap whatever button to get rid of whatever is obscuring the page? Is there even a difference if I tap on a different button? How many millions of hours have we wasted with this annoyance? Would the same people fearful of cookie tracking a decade ago now prefer a world where such regulations no longer exist?

Already today, it is socially acceptable to record without consent in certain situations. The body camera worn by police has rocked our preconceptions of law enforcement. In the Central Park Karen incident, a black man, Christian Cooper, didn’t ask for Amy Cooper’s consent to record their interactions regarding her off-leashed dog. Likewise, if a deaf person were using a speech-to-text recording device, they shouldn’t be expected to ask for consent for every conversation they are engaged in, would they?

The funny thing is, at the rate AI is advancing, I could well imagine that, a few years from now, we will all feel a little handicapped if our access to LLMs and AIs were cut off—perhaps akin to not having access to a search engine or Wikipedia today?

If we want AIs to be helpful little assistants in our daily lives, they need to know our daily lives and inner preferences intimately. How else would they know not to book a meeting earlier than 10 a.m. for me, when a 7 a.m. breakfast meeting may be what you prefer? Or to treat me like a five-year-old when teaching me how to fertilize my wife’s orchid but converse with her like the plant expert she is?

In What Technology Wants, Kevin Kelly posits that while we may resist technological changes in the short term, humans and technology are interdependent and co-evolving in a way that is unstoppable in the long run. He argues that both life and technology are moving towards increasing efficiency, opportunity, complexity, diversity, and, perhaps more controversially, freedom, beauty, and sentience. At this abstract level, resistance is indeed futile.

Norms change. My guess is that a large part of why my friends were upset with me was because they were surprised, and a recording device on a hike is just not the norm.

As we ended our hike, my friend suggested we all do a selfie. No one was afraid that the camera would capture our souls, as some Native Americans believed a couple of centuries ago. Norms change, but in the present, on that fateful hike, I really should have handled it differently.