Categories
AI Business Consulting

The Toll Bridge and the Terrain

For fifteen years of my life, I lived inside the fortress of information asymmetry. I was part of a payments consulting business, and our model was exactly what Andrew Feldman described on a recent Moonshots episode when he pointed a sharp finger at traditional professional services.

His observation was simple, cutting, and entirely true:

“Their role today is to stand between ordinary people and obscure knowledge. And the application of that obscure knowledge to everyday problems.”

When I heard him say that, it landed with a quiet thud of recognition.

For a decade and a half, my colleagues and I were the ones standing in that gap. The payments industry—with its labyrinth of interchange fees, compliance structures, clearing networks, and legacy tech stacks—is a monument to obscure knowledge. Clients didn’t come to us because we possessed some divine, unreplicable wisdom. They came to us because the map was locked in our heads, and navigating the terrain without us was a recipe for an expensive disaster.

We charged for our time, and we earned it. We untangled complexity and solved real, everyday business problems for people who just wanted to move money safely from point A to point B.

But looking back now, I can see the architectural flaw disguised as a premium service. The economic foundation of that entire era relied on friction. It relied on the fact that it took an immense amount of human energy to retrieve a piece of obscure data and map it onto a specific business dilemma. You weren’t just paying for strategic guidance; you were paying a premium on artificial scarcity.

We are living through a moment where the marginal cost of intelligence is rapidly trending toward zero. When the barrier of “obscure knowledge” evaporates, the traditional toll bridges begin to look absurd.

For anyone starting a consulting business today, the playbook would have to be entirely different. When an LLM can parse thousands of pages of network operating rules, interchange tables, and regulatory compliance frameworks in a handful of seconds, the gatekeeper’s standing ground liquefies.

If your value proposition is merely standing between a client and a hidden database, your business model isn’t just flawed—it’s obsolete.

Yet, this collapses into a fascinating paradox. You might assume that when you democratize expertise, you eliminate the need for the expert. But as Dan Shipper recently observed, the reality of AI is completely counterintuitive.

Shipper points out that AI effectively packages up “yesterday’s competence” and makes it cheap and ubiquitous.

Suddenly, anyone can generate a complex contract, a software pull request, or a payments flow strategy with the click of a button. But when cheap competence skyrockets, adoption explodes, resulting in an unprecedented glut of generic output—what the internet has collectively taken to calling “slop”. It’s the default, lazy answer that lacks soul, context, and nuance.

When everything begins to look and smell the same, a strange thing happens: the market’s demand for genuine difference sky-rockets.

The shift we are facing across all professional services—whether legal, financial, or consulting—isn’t about eliminating the expert. It is about changing the expert’s job from data-retriever to orchestrator and judge. The floor has been raised. Yesterday’s ceiling is today’s baseline.

What remains is the ability to read a room. To watch a client’s shoulders tighten when you present an option that’s technically correct but organizationally impossible. To notice the glance exchanged across the table before anyone speaks. No LLM parses that. The map is universal now; the guide still has to be in the room.

We don’t need fewer guides; we need fewer toll booths. The future of consulting doesn’t belong to those who hoard the map. It belongs to those who use a universally available map to help people actually walk the terrain.

Categories
AI Business Investing Technology

The Scarcity Portfolio: Navigating Sovereign Debt, Wafer Bottlenecks, and Orbital Compute

Today I was watching the interview of Gavin Baker by Patrick O’Shaughnessy on his Invest Like the Best podcast. Like prior conversations this was another fascinating excursion into the mind of a sophisticated and very successful tech venture investor.

During the conversation, Patrick asked Gavin what agents he was using that were especially helpful and he mentioned one which summarizes YouTube podcasts and videos for him. Like most of us Baker just doesn’t have the time to watch or listen to them himself so good summaries are really helpful.

Turns out I’ve been working on a Google Gemini Gem that does this for me. When Baker mentioned his I fired up the new Gemini 3.5 Flash model and asked it to summarize the Baker interview.

Later in the conversation Baker used the term “battlefield AI” which caused me to go back to Gemini again to learn more about that. The results were so interesting that I asked Gemini to create a syllabus for a semester class on these subjects. After that I asked it to convert our whole conversation into a Markdown file so I could share it. You’ll find it below.

I found this whole experience pretty stunning. I came away very impressed with Gemini 3.5 Flash both for the quality of the responses but also the sheer speed. Wow!

Anyway I hope you enjoy the following!

Continue reading “The Scarcity Portfolio: Navigating Sovereign Debt, Wafer Bottlenecks, and Orbital Compute”
Categories
Haiku Living Reading

The Presence We Keep Deferring

I have so many unread articles saved to Instapaper that I’ve stopped checking the count. Each one felt, in the moment of saving it, like something I needed. A long piece on urban planning, a profile of someone interesting, a reported essay I fully intended to sit with.

The app is beautifully designed for exactly this — the frictionless capture, the clean reading interface waiting patiently on the other side.

What it can’t do is manufacture the attention I didn’t have when I saved it and still don’t have now. The articles aren’t the problem. The premise is: that presence is something you can bank.

There’s a haiku I keep returning to, from Natalie Goldberg’s Three Simple Lines. It’s by a poet named Fumiko Harada:

Morning chill
I savor this moment —
one meeting one lifetime

Eleven words. No verb in the third line, which makes it feel less like a thought and more like a verdict.

The Japanese concept underneath it is ichi-go ichi-e — loosely, “one time, one meeting.” It’s a Zen idea with origins in the tea ceremony, the understanding that each gathering is singular and therefore irreversible. You cannot archive it. You cannot search for it later. When it ends, it doesn’t go anywhere you can retrieve.

This is what the Instapaper queue is, at scale: an archive of moments I decided to experience later. The article about urban planning was written by someone who spent months reporting it, on a day when some editor thought it was ready, and landed in my feed on a morning when something about the headline caught me. That constellation doesn’t reassemble. Later is a different article.

The tools I use every day are getting astonishing. There are systems that can summarize, translate, recall, explain, anticipate. I use them. I find them genuinely useful.

But there’s a habit of mind they reward — a kind of perpetual deferral of full attention — that I haven’t fully reckoned with. The promise, always, is that you can engage more completely later, once the summary is ready, once the transcript exists, once the notes have been taken. Presence becomes a productivity tax you pay while waiting for a deliverable.

Harada’s haiku doesn’t moralize. The speaker isn’t lecturing herself into awareness. She’s just cold, and awake, and choosing to notice. I savor this moment. The word “savor” does a lot of work. It implies effort. You savor things that could be missed.

The pivot in the third line is what stays with me. One meeting one lifetime. Not “this meeting will last a lifetime” — that would be sentiment. It’s more like a mathematical statement: the cardinality of this encounter is one. There is exactly one of them. This morning, this particular chill, whatever conversation or solitude is happening inside it — that set has one element. By tomorrow it has zero. No amount of documentation changes that arithmetic.

I’m working on believing that.

Categories
Architecture Infrastructure

The Architecture of the Indestructible

We are conditioned to look for the center of things. When we try to understand an organization, we ask for an organizational chart. When we look at a nation, we look to its capital. Traditional architecture—whether of a building, a company, or an army—relies on a classic playbook: a strong hub, radiating outward. You find the center, you secure it, and the system holds.

But what happens when you try to decapitate an enemy, or a technology, that has no head?

In 1964, a brilliant engineer named Paul Baran sat at his desk at the RAND Corporation, trying to solve a Cold War nightmare: How do you maintain a communications network after a catastrophic nuclear strike? Baran realized that traditional networks were centralized—like a wheel with spokes. If you destroy the hub in the center, every single spoke becomes useless.

His solution was the distributed network, the foundational blueprint for what would eventually become the Internet.

“Under the proposed system, each station would need to be connected to only a few of its nearest neighbors… The system would be highly reliable, even if a large fraction of the stations were destroyed.”

Baran mathematically proved that if you remove the center, the edges don’t die. They simply reroute. A few decades later, telecom engineers used a remarkably similar logic to build cellular telephone networks. Instead of one massive, high-power radio tower serving an entire city, they broke the terrain into a grid of small, low-power cells. If one tower goes offline, the network degrades gracefully rather than collapsing. It bends, but it refuses to break.

There is a profound, poetic irony buried here. The United States government originally funded Baran’s research to create a distributed network so that its centralized monolith could survive. Decades later, asymmetric adversaries across the globe adopted that exact architectural philosophy for their physical defense doctrines—creating “Mosaic Defense” systems designed specifically so that when you destroy the center, the edges keep fighting.

They copied our homework to survive our strength.

I find myself thinking about this tension far beyond the realms of military strategy or software engineering. It is a metaphor for how we construct our lives. We often build centralized lives—anchored entirely to a single identity, a single career, or a single institution. We project a monolith of strength to the world. But monoliths are brittle. When the center is struck, the whole architecture crumbles.

The lesson of our modern architecture is becoming increasingly clear, whether you are managing a network, building an organization, or navigating the quiet complexities of a human life. The fragile monolith is an illusion of safety.

The future belongs to the web that knows how to reroute.

Categories
Etiquette Founders

How to Be in a Room

I happened to notice this in a post on X: Slow Ventures ran a 4-hour etiquette class for founders. This is actually a brilliant idea. A non-trivial amount of founders (and VCs) could use some help in this department.

So I asked AI: what would a course syllabus for something like this look like? Below is the response (which I quite like!):

This is a genuinely good idea — etiquette as a business skill is perennially under-taught, especially for technical founders who may have gone deep on product/engineering but shallow on navigating rooms. Here’s how I’d structure a day-long or modular course built around the Slow Ventures premise.

Founder Etiquette — A Field Guide

4-hour intensive · 5 modules · lecture + role-play format

Morning Block — First Impressions

Module 1 · 45 min — The room before the meeting

Arrivals, introductions, and presence under observation

  • Timing: the cost of early vs. late
  • Handshakes, eye contact, the graceful exit from a name you missed
  • Business card and phone etiquette
  • How to work a room without appearing to work a room
Module 2 · 45 min — Table stakes: dining fluency

The business meal as performance and ritual

  • Place settings, utensils, and the BMW rule
  • Who orders first, who orders what
  • Wine: fluency without performance
  • Dietary restrictions — yours and theirs
  • The check: the ritual, the timing, the dodge
  • When and how to talk business at the table

Mid-Morning Block — Communication

Module 3 · 45 min — Communication that doesn’t leak desperation

Written and verbal signals investors actually read

  • Email: subject lines, length, tone, and the follow-up cadence
  • Reading the room on Slack/text vs. formal channels
  • The thank-you note — when, how, why it still matters
  • Saying no, disagreeing, and delivering bad news
  • How to receive feedback without deflecting or over-agreeing
  • The exit from a conversation: graceful, not abrupt

Afternoon Block — High-Stakes Scenarios

Module 4 · 45 min — Investor relations: the long game

Board rooms, pitch meetings, and the unwritten VC code

  • Board meeting protocols — who speaks, in what order
  • Managing up to your board without managing them
  • The warm intro — how to ask, how to use, how to close the loop
  • Confidence vs. bravado: what the room hears
  • What to do when the meeting is going badly
  • Conference behavior: panels, hallways, and the after-party
Module 5 · 60 min — The edges: cultural fluency and hard situations

Where etiquette meets judgment

  • Cross-cultural norms: Japan, Middle East, Europe, India — the high-risk differentials
  • Power dynamics: how to behave when you have more (or less)
  • Navigating someone else’s social gaffe gracefully
  • Gift-giving: what to give, what never to give
  • The fine line between casual and sloppy in casual settings
  • Live role-play: difficult scenarios, coached debrief

A few things I’d stress if actually running this:

Module 3 is where the real behavioral work happens for founders specifically. The desperation signal leaks through email more than anywhere else — the too-fast reply, the over-explanation, the follow-up that arrives four hours after the first message. That’s the stuff that’s hard to see in yourself.

One thing Slow Ventures probably got right: four hours is about the right ceiling. Any longer and you start to feel like you’re at finishing school, which kills the energy.

After reviewing this, I thought it might be useful to add two role playing scenarios to the course. Here’s what the AI recommended:

First, add this role play in after Module 2:

Role-play 1 · 20 min — The wrong bread plate

Recovering from a gaffe without losing the room

A founder is taken to lunch by a partner at a firm they’ve been trying to get in front of for six months. It’s going well — good rapport, the partner seems genuinely interested. Then the founder accidentally takes the partner’s bread roll, realizes it mid-bite, and has to navigate the next ten minutes without letting the gaffe collapse the energy in the room. Secondary layer: the check arrives and it’s ambiguous who’s hosting.

This tests everything from Modules 1 and 2 simultaneously — table fluency, recovery under embarrassment, reading whether to acknowledge the mistake or absorb it silently, and the check ritual. Low-stakes enough to be funny, high-stakes enough to feel real.

Next, add this role play after Modules 3, 4, and 5:

Role-play 2 · 30 min — The meeting that’s going sideways

Holding the room when the numbers missed

A founder is in a board meeting presenting quarterly numbers that missed. Midway through, one board member interrupts with a pointed question the founder doesn’t have a clean answer to. A second board member piles on. The founder has to hold the room — not deflect, not over-apologize, not go defensive — while also reading which of the two is genuinely concerned versus performing concern for the room. Cultural wrinkle: the more aggressive questioner comes from a context where directness reads as respect, not attack.

This pulls from Modules 3, 4, and 5 simultaneously. It’s the hardest scenario in the course and should feel that way. The coached debrief afterward is where most of the learning actually happens — ending here means it lands with weight, which is the right note to close on.

More commentary from the AI:

The dining module is the most load-bearing. A remarkable amount of professional signaling happens at the table and most people don’t realize they’re broadcasting anything. The BMW rule alone (bread-meal-water, left to right, so you don’t eat someone else’s roll) is the kind of thing nobody tells you until you’ve already grabbed the wrong bread plate in front of a Series B lead. The live role-play after Module 2 brings this to life.

The live role-play in Module 5 is what separates a good version of this class from a mediocre one. You can lecture about graceful exits from bad meetings all day. Watching someone flail through one in real time, then getting coached on the spot, is what actually moves the needle.

Categories
AI Consulting

The Judgment Layer

An analyst’s note about the CEO of one of the largest consulting companies making comments at an investor conference includes a line that deserves more attention than it got: “token volume used on a project isn’t a proxy for AI maturity.”

Translation — clients are burning money on frontier models for problems that don’t need frontier models, and they’re not getting the outcomes they expected.

This firm’s CEO offered this as a business opportunity. I read it as a confession.

The old consulting model was simple: client has a technology problem, firm deploys humans to solve it. Billing followed effort. The new problem is different in kind — clients have an AI strategy problem. They know they’re supposed to be using AI. They’ve heard the word “frontier.” They’re spending accordingly. They just don’t know why, and the outcomes are showing it.

So the CEO is right that there’s an opportunity here. The value proposition shifts from implementation to judgment — not deploying AI, but knowing when not to deploy the expensive one. Matching capability to problem. Being trusted enough to tell a client that their $50M frontier model contract is solving a $500K problem.

Here’s the irony that the comment skates past: that advice is structurally difficult for a large consultancy to give.

The business model that built consulting firms was billing for doing. The more you deploy, the more you bill. Helping a client spend less, or choose the cheaper model, or run a narrower project, is genuinely good advice that the incentive structure actively works against. You don’t grow a $70 billion professional services firm by talking clients out of scope.

The judgment layer, if it becomes the real value, requires something closer to a doctor’s relationship with a patient than a contractor’s relationship with a client. Doctors get paid whether they prescribe or not. The value of the visit is the diagnosis — including the diagnosis that says you don’t need the expensive intervention. Consultants, historically, get paid to prescribe, and paid more when the prescription is larger.

There’s a reason we trust doctors with that asymmetry and not contractors. Licensing, malpractice, professional norms built over centuries — all of it exists to align the incentive. Consulting has none of that infrastructure. What it has instead is reputation, which is slower-acting and easier to game.

Whether the large firms can actually make the shift — rather than just reframe the same billable-hours model in the language of AI optimization — is the real question the market is wrestling with. The CEO’s comment is genuinely perceptive about where client value lies. It’s less clear that consulting firms are currently built to capture it honestly.

Categories
Business History IBM Infrastructure Nvidia Programming Semiconductors

The Half-Life of Moats

Prompted by an article on X by @magicsilicon on the CUDA moat. Research and drafting assistance from my AI intern assistant Clark.

The NVIDIA H100 looks, in retrospect, like an inevitability. It wasn’t.

What Jensen Huang built is more accurately understood as a sixteen-year accumulation of optionality — a platform investment made in 2006 for a market that wouldn’t fully materialize until 2022. NVIDIA intros the G80 architecture in November 2006, laying the groundwork for CUDA’s release a few months later. The stated ambition was to let scientists write C++ that ran on GPU cores without needing to understand 3D graphics pipelines. The unstated bet was that parallel computation would eventually matter for something bigger than rendering shadows in video games.

For sixteen years, it mostly didn’t. Not at scale. Not commercially. CUDA lived in research labs and HPC clusters. It attracted a small, devoted, and economically marginal user base — the kind that papers cite but investors ignore. NVIDIA kept investing in it anyway: cuDNN for deep learning operations, cuBLAS for linear algebra, a layered ecosystem of libraries that made CUDA not just accessible but nearly irreplaceable for anyone doing serious numerical computation. When TensorFlow and PyTorch emerged as the standard frameworks for neural network research, they didn’t adopt CUDA because it was the only option. They adopted it because CUDA was where the optimized kernels already lived.

AlexNet won the ImageNet competition in 2012 and did it on two NVIDIA GPUs. The deep learning community noticed immediately. The financial community largely did not.

Then ChatGPT launched in November 2022, and suddenly everyone needed H100s they couldn’t get.

The parallel to Intel is instructive and also undersells how strange this kind of story looks while you’re living through it. Intel was founded in 1968 as a memory company. DRAM. The founders — Noyce, Moore, Grove — were materials scientists and engineers who believed the future was in silicon memory chips. They were right, briefly: in the early 1970s Intel dominated the DRAM market. By 1984, that share had collapsed to 1.3%, ceded almost entirely to Japanese manufacturers who had commoditized the product.

What saved Intel wasn’t a pivot so much as a realization that a stopgap had become a foundation. The 8086, conceived in 1976 as an internal hedge and launched in 1978 was never supposed to matter. It was a 16-bit processor designed to hold off Zilog while Intel finished its ambitious 32-bit iAPX 432 architecture. The 8086 was assigned to a single engineer. “If management had any inkling that this architecture would live on through many generations,” its designer Stephen Morse later recalled, “they never would have trusted this task to a single person.”

IBM chose the 8088 — a cost-reduced variant — for the original IBM PC in 1981. That decision wasn’t destiny, it was simply a procurement. And yet from that accident of selection, Intel’s x86 line became the backbone of personal computing for four decades. The Pentium in 1993 was Intel’s Wintel moment — the flag bearer the @magicsilicon tweet gestures at — but the flag had been quietly sewn since 1978.

What these histories share is not just a pattern of “slow build, explosive payoff.” The structural similarity is subtler: in both cases, the moat was a software abstraction layer built on top of hardware. Intel’s real lock-in wasn’t transistor count or clock speed. It was backward compatibility — the commitment, formalized with the 80386 in 1985, that every future Intel chip would run software written for older ones. That promise created a flywheel that trapped developers and buyers in a virtuous (for Intel) dependency loop for decades.

CUDA is the same architecture at a different layer. The lock-in isn’t the H100’s 80 gigabytes of HBM3. It’s that switching to an AMD MI300X or Google TPU means potentially rewriting training pipelines that have been optimized against CUDA kernels for years. AMD’s ROCm platform exists. It is, by most accounts, maturing. Engineers who have tried the migration report that it costs months and hundreds of thousands of dollars. The moat isn’t a wall. It’s accumulated friction — the switching cost of a decade of engineering decisions baked into codebases that no one wants to touch.

But to find the actual origin of this pattern, you have to go back further than Intel. To 1964, and to a decision IBM made that Fred Brooks — its project manager — called a bet-the-business move.

The IBM System/360 was announced on April 7, 1964, after five years of turbulent internal development. What it introduced wasn’t just a new computer. It was a new concept: the separation of architecture from implementation. Before the 360, IBM ran five incompatible product lines simultaneously. A customer who outgrew their machine had to scrap all existing software and start over. The 360 replaced all five lines with a single unified architecture — six models covering a fiftyfold performance range, all running the same operating system, all sharing the same instruction set. The name itself encoded the ambition: 360 degrees, all directions, all users.

Gene Amdahl, the 360’s chief architect, had a precise formulation for what this meant: the architecture was “an interface for which software is written, independent of any implementation.” The Principles of Operation manual described what the machine did; separate Functional Characteristics documents described how each model did it. This distinction — separating the contract from the execution — was genuinely new. It’s the conceptual root of everything that came after.

The 360 generated over $100 billion in revenue for IBM and established the first platform business model in computing. Jim Collins would later rank it alongside the Model T and the Boeing 707 as one of the three greatest business achievements of the twentieth century. But its deepest legacy was architectural: the insight that if you make your abstraction layer the standard, the hardware underneath becomes fungible. Customers didn’t buy specific IBM machines. They bought into OS/360. The machines were an implementation detail.

Intel understood this by the 1980s, even if implicitly. The 80386’s backward compatibility commitment in 1985 was IBM’s 360 insight applied to microprocessors — the architecture is the product, the silicon is the vehicle. CUDA is the same insight applied to GPU compute. What NVIDIA sold researchers in 2006 wasn’t the G80 card. It was the abstraction: write parallel code in C++, run it on any NVIDIA hardware, trust that the next generation will be faster and compatible.

The pattern is now sixty years old. It has reproduced in every major platform transition. And it keeps working for the same reason it worked in 1964: when you own the layer that developers write to, your customers’ switching costs compound every year they stay.

There’s something worth sitting with here. Neither Jensen Huang in 2006 nor Gordon Moore in 1968 could have specified exactly what the payoff would look like. What they shared was a willingness to build infrastructure for a demand they could sense but not yet see — and the discipline to keep investing in it through the long years when it looked like a research project rather than a business.

The question that doesn’t resolve cleanly is whether that kind of patience is a strategy or a personality. And whether, in an industry that now moves faster than the cycles it’s lived through, sixteen-year moats are still the kind that get built.

Which raises the uncomfortable corollary: the same AI tools that CUDA enabled may be what ultimately erodes it.

The attack on CUDA’s moat is now structurally different from anything AMD or Intel could mount before. OpenAI’s Triton compiler lets developers write GPU kernels in Python without touching CUDA at all, and generates optimized machine code that often matches hand-tuned CUDA performance. MLIR — Multi-Level Intermediate Representation, originally from Google — provides a compiler infrastructure that can target any hardware backend from a single codebase. AMD’s ROCm has historically been dismissed as immature; ROCm 7, released this year, delivers meaningfully better inference performance than its predecessors. And perhaps most directly: Claude Code reportedly ported a CUDA codebase to AMD’s ROCm in thirty minutes — work that previously took months of engineering time.

The irony is almost too neat. CUDA’s moat was built on accumulated switching costs: the friction of rewriting code, the library dependencies, the tribal knowledge encoded in a decade of kernel optimizations. AI coding tools are specifically good at exactly that kind of mechanical, high-context translation. The weapon is attacking the wall it was built behind.

That said, it’s worth being careful about the speed of this. Abstraction layers that “should” erode moats often take far longer than expected, because the moat isn’t just the code — it’s the ecosystem of tooling, documentation, community knowledge, and hardware-software co-optimization that took eighteen years to compound. Triton and MLIR are real. They’re also early. The question isn’t whether the moat is vulnerable; it’s whether it erodes before NVIDIA’s next generation of chips makes it irrelevant to argue about.

As for what comes next — which company is building the IBM 360 of this decade — the honest answer is that it’s too early to call with confidence. But there’s a candidate worth watching.

Anthropic’s Model Context Protocol, launched in late 2024, has the structural fingerprint of a platform play. MCP is a standard for how AI agents connect to external tools and data sources — a common interface layer, hardware-agnostic (or rather, model-agnostic), that any system can implement. By late 2025 it had been donated to the Linux Foundation, adopted by OpenAI and Google, and was tracking 97 million monthly SDK downloads. There are now over 10,000 MCP servers. It is becoming the way agents talk to the world.

The parallel to OS/360 is imprecise but instructive. What IBM built in 1964 was a standard interface between software and hardware that decoupled what you wrote from what you ran it on. MCP is attempting something similar one abstraction layer higher: decoupling what an agent does from the specific models, tools, and data sources it does it with. If it becomes the standard — the layer that developers write to — then whoever owns or most deeply shapes that standard controls the integration tax of an industry whose applications we can’t fully specify yet.

The counterargument is that open standards, once donated to foundations and broadly adopted, don’t generate the same lock-in as proprietary platforms. OS/360 was IBM’s. CUDA is NVIDIA’s. MCP is now the Linux Foundation’s, with OpenAI and Google as co-stewards. The historical pattern suggests the moat accrues to whoever owns the layer, not whoever invented it.

Which may mean the next great platform play is still being assembled in a room we haven’t seen yet — the way IBM’s System/360 was being architected in a Connecticut motor lodge in 1961, three years before anyone else knew what was coming.

Categories
AI History

The Arrival

Yoshua Bengio spent forty years building the foundation of modern artificial intelligence. He won the Turing Award for it. And he didn’t think he’d live to see it work.

That’s the quiet fact buried inside Stephen Witt’s New Yorker profile of him. Bengio — one of the three researchers whose decades-long bet on neural networks eventually became the architecture underlying every large language model running today — had made peace with the idea that the thing he was building was a multi-generational project. Something for his successors to finish. Then Witt writes: “one day in late 2022, the technology had simply arrived. He compared it to meeting an extraterrestrial.”

Hemingway once described bankruptcy happening two ways: gradually, then suddenly. He meant ruin. Bengio experienced something harder to name — not ruin but arrival, which carries its own vertigo. The gradually was four decades of work that most of his peers considered quixotic. The suddenly was a Tuesday in November when a chat interface went live and the world quietly changed.

What unsettles me about the extraterrestrial comparison isn’t the strangeness it implies. It’s the distance. You meet an alien; you don’t meet something you made. The metaphor suggests that even its creator couldn’t fully recognize it — that the thing, once arrived, belonged to a category that exceeded its own origins.

We don’t have good language for this. Breakthrough, inflection point, paradigm shift — these are words people reach for after the fact, when they’re building timelines. What Bengio seems to be describing is the experience of standing in front of a threshold you spent your life approaching, and finding it already behind you.

The technology didn’t ask permission. It didn’t announce itself.

It arrived.

Categories
AI Technology

The Bathwater Problem

Gary Kamiya was writing about the Tenderloin when he said it, but the line has been following me around: “The problem is that by saving the baby, you also save the bathwater.”

The pattern is remarkably consistent across every major information technology. Each one arrives promising to liberate the deserving — the faithful, the learned, the civic-minded — and each one immediately, inevitably, arms everyone else too. Gutenberg’s press was understood by its champions as a device for spreading the true Word; within decades it was the primary infrastructure for Protestant schism, Catholic counter-propaganda, astrological almanacs, and pornography. The reformers got their Bible. They also got their pamphlet wars.

The telegraph was greeted as a force for peace — shared information would make war irrational, commerce would bind nations. It also became the nervous system of commodity speculation, financial manipulation, and the first truly industrial-scale news hoaxes. The telephone: connection and the crank call, the crisis line and the threatening voice in the dark. Radio: FDR’s fireside chats and Father Coughlin. Television: Murrow taking down McCarthy, and also fifty years of manufactured consent. The internet: the largest library ever assembled and the largest sewer.

The pattern isn’t coincidental. It’s structural. Each technology expands what’s possible for human expression and coordination — and human expression and coordination contain both the noblest and the worst of us in roughly fixed proportion. The tool doesn’t change the ratio. It scales both sides of it.

What’s interesting historically is how each generation believes their technology will be different — that this time the architecture can be designed to select for the good. The internet era produced the most elaborate version of this belief: algorithmic curation would surface truth, network effects would reward quality, the wisdom of crowds would outcompete misinformation. Instead it turned out that engagement was the attractor, and outrage was the highest-engagement content. The bath got hotter.

The AI moment is the same belief system, restated with more technical sophistication. But the Kamiya line stands. You are saving a baby, and you are saving bathwater, and no one has yet designed a tub that can tell the difference.

The question isn’t whether the bathwater comes with the baby. It always does. The question is whether you turn on the tap.

Categories
AI AI: Transformers Books

The Updating Machine

Tom Chivers puts Bayes’ theorem in plain English and it sounds almost obvious: “the probability of event A, given event B, equals the probability of B given A, times the probability of A on its own, divided by the probability of B on its own.” A formula for revising what you believe when new evidence arrives. You started somewhere. Something changed. Now you believe something slightly different. Repeat.

The obvious part is the mechanics. The hard part is the loop.

Most reasoning errors I catch in myself aren’t failures of logic — they’re failures to update. I hold a view, evidence accumulates against it, and I find reasons the evidence is flawed rather than reasons the view might be.

Psychologists have a name for this: confirmation bias. But I’ve always found that label a bit too clean, like it describes a bug rather than a feature.

The prior isn’t wrong to be sticky. It represents everything you’ve learned up to this point. The problem is when it becomes load-bearing — when the prior stops being a starting position and starts being a conclusion.

“Strong opinions, loosely held” is supposed to solve this. It’s a useful phrase — it captures something true about the right posture toward your own beliefs. But in practice the second half is harder to honor than it sounds. The strong opinion gets stated, new evidence arrives, and changing your mind in public feels like losing. The “loosely held” part quietly becomes decorative.

What Bayes actually demands is something closer to epistemic humility with arithmetic attached. You don’t get to say I don’t know. You have to say I estimate 0.4, and here is what would move me to 0.6. That’s harder. It requires you to specify not just what you believe but how you’d know if you were wrong.

This is why Bayesian thinking keeps surfacing in AI conversations. Modern language models do something structurally adjacent to this — not consciously, but mechanically. Every token generated is a probability distribution revised forward by context. The model doesn’t know the next word; it updates a prior over all possible words, given everything that came before. It’s not reasoning the way humans reason, but it’s updating the way Bayes updates: continuously, contextually, without the luxury of certainty.

Whether that’s comforting or unsettling probably depends on your own prior.

The deeper thing Chivers is pointing at, I think, is that Bayesian reasoning is essentially a description of intellectual honesty as a process rather than a trait. You can’t just decide to be open-minded. You have to build the loop: form a belief, assign it a probability, watch for evidence that should move it, and then actually move it. Most of us do the first three. The fourth step is where it gets expensive.

I’ve been wrong about enough things by now that I’ve started to treat my own confident views with mild suspicion. Not paralysis — you have to act on something — but a background awareness that the prior I’m acting on was formed by a person who had less information than I do now, and less than I’ll have next year.

Strong opinions, loosely held, sounds right. The trick is meaning it.