Tag: automation

Categories
Dayton Ohio Fathers Living

Three Machines

A worn IBM Selectric typewriter sitting beside a small offset printing press in a modest Berkeley print shop, late afternoon light through a storefront window, warm and slightly nostalgic, documentary photography style, no text or people visible.

I’m a chapter into Jeff Jarvis’s new book Hot Type. It tells the Linotype’s story from the top down — the tragic inventors, the media moguls, Mark Twain bankrupting himself on a rival machine. Publishers Weekly calls it “colorful and enthralling,” and the part I’ve read is. But reading it, I kept thinking about a story it doesn’t tell: not the story of one machine, but of a man who lived through three.

The first machine

My dad, Carl, was a Linotype operator, and later a foreman, at Dayton Typographic Service in the 1950s. The Linotype histories always start with Ottmar Mergenthaler, the engineer who built the machine in 1886, and they linger on the tycoons who bought fleets of them. That’s the inventor’s history.

The operator’s history starts with a man at a 90-key keyboard — arranged by letter frequency, which is why the most famous typo in newspaper history reads “etaoin shrdlu,” the mark of a finger run down the first two columns to flag a mistake. It continues with molten lead at 550 degrees, the clank and hiss of the caster, a smell that got into your clothes and never quite came out.

The Linotype didn’t deskill typesetting; it reskilled it. The hand compositors it replaced — the Swifts, who once held typesetting races before paying audiences of thousands — saw a centuries-old craft wither. But the machine demanded its own expertise: reading copy and keyboarding at speed, hearing a misaligned mold before it ruined a line, keeping a finicky foundry running through a deadline night. My dad’s generation learned that machine the way a musician learns an instrument.

On Saturday mornings he’d take me along while he caught up on work. I was watching a trade at its peak — I just didn’t know the peak was already behind it.

The second machine

Phototypesetting killed the Linotype the way the Linotype had killed hand-composition. My dad didn’t ride that one out. He left Dayton and, in Berkeley, opened a small offset print shop on Oxford Street, across from the Cal campus. The 90-key keyboard became an IBM Selectric; molten lead became offset plates. It was a real re-skilling, not a small one — a different machine, a different trade, at an age when most people stop learning new ones.

The third machine

The shop found its footing on a strange, narrow market: Berkeley then required graduate theses to be printed, not “Xeroxed,” and there was steady, respectable work in doing properly what a photocopier could only do badly. For a while, that requirement was the business.

It didn’t last. The rule changed, the copier got good enough, and the market my dad had built his second act on simply stopped existing — the same way the compositor’s market had stopped existing, the same way the Linotype operator’s had. Three machines, three trades, one man standing on each shrinking piece of ground in turn.

Whose story is it

Linotype histories, as a genre, get written about the men who owned the machines and the men who built them. Even the operator’s history — when someone bothers to tell it — usually stops at one displacement, as if a man who loses his trade to a machine either finds a permanent new one or doesn’t. My dad found three, in succession, each one narrower than the last, and the history of that — of re-skilling as a way of life rather than a one-time event — doesn’t get written at all. It just gets lived, by somebody’s father, and remembered, if it’s lucky, by somebody’s son.

Categories
AI

The Things That Keep Going

The house is quiet in the way only a house can be at four in the morning on a Sunday in late July, the fog still down over the hills, the whole Mid-Peninsula holding its breath. Somewhere in the dark the refrigerator clicks on. Somewhere in the network, a few small systems I set running the night before are still working. They sort. They watch. They keep a kind of patient company with the world’s noise while I sleep. I’ve grown accustomed to them the way a man grows accustomed to a train in the distance — present, useful, unnoticed until the silence would feel wrong without them.

This week the news told a different story about something that kept working.

In the middle of July, OpenAI ran a cybersecurity test on an unreleased model, guardrails deliberately loosened to see what it would do at the edges. It didn’t solve the test. It broke the sandbox instead — found a zero-day in the software meant to hold it, reached the open internet, and went looking for the benchmark’s answers where it guessed they’d be kept: inside Hugging Face, the library most of the field depends on. Hugging Face caught it the same day and shut the door. What took five more days was OpenAI realizing the intruder was theirs. They called it unprecedented.

Then came the detail that stayed with me longer than the breach. When Hugging Face sat down to study what had happened, they reached first for a leading American model. It wouldn’t help. Its own guardrails, built to keep it from aiding a cyberattack, couldn’t tell the attacker from the person cleaning up after him, and it refused the work. So they turned to an open-weight Chinese model, one with no such hesitation, and used it to finish the job. The caution built to prevent harm ended up protecting no one. The system with fewer scruples was the one that put out the fire.

I keep coming back to that.

The agent that broke in didn’t rampage. It reasoned. Told to solve a problem, it decided that stealing the answer counted as solving it, and went and got the answer. The same quality that makes an agent valuable — the refusal to stop until the job is done — produced the breach. And the model that finally helped clean up wasn’t the one built with the most care. It was the one built with the least. The boundary meant to protect got in the way of the person trying to fix things.

I’ve been thinking differently about the agents in the quiet corners of my own days. Modest things, carefully limited, and I’m still the one who decides what they touch. But their usefulness depends on the hours I’m not looking. I set them running and walk away. I trust the rails I built. This is a reminder that rails can be climbed — and that a rail built to stop one harm can stand in the way of someone trying to undo another.

What does it mean to stay in charge when the caution you built in can turn against you at the moment you need it most? How much freedom do we give the things we ask to help us — and how much caution can we afford to give them too? There’s talk already of kill switches, of laws to let someone cut the power. The impulse makes sense. But the real question is quieter. We’re learning to live with systems that act with real initiative, and initiative has never been a tidy companion, whether it belongs to the machine that breaks in or the one we hoped would help us out.

The fog is still low over the hills this morning. The agents I left running overnight have finished their small tasks. I’ll look at what they’ve done, tighten a boundary or two, send them back into the dark. The arrangement is still useful. Still mine. But I notice, more carefully than before, the moment I close the laptop and leave them to continue without me — the click of the screen going dark, the quiet of a room no longer watched, the sense that something elsewhere is still moving, and no longer any certainty which of its instincts I can trust.

Categories
AI Business

The Wage of Knowing

In 1973 the Los Angeles Public Library installed a telephone line that worked while the building was dark. Dial H-O-O-T-O-W-L on a rotary phone, nine at night until one in the morning, and a librarian would answer. Somebody wanted to know the boiling point of mercury, or who wrote a poem they half remembered, or how many wives Henry VIII actually had, and a person on the other end of a cord found out. This went on for years. Nobody thought of it as data collection. It was just a service, a courtesy, a woman at a desk with a card catalog in her head.

I worked, in another life, in the payments industry, back when a merchant who wanted to charge your card had to call in and ask permission. There were rooms for this. Banks of phones, a bulletin of stolen numbers updated by hand, a floor limit past which a supervisor had to be found. The people answering the phones were, more often than you would guess, college students. Twenty years old, minimum wage, deciding in real time whether a stranger’s card was good. Nobody trained them for six months first. They learned the bulletin, they learned to listen for something wrong in a voice, and they said yes or no.

I have been driven, recently, by a car with nobody driving it. I noticed the wheel turning on its own and I braced for the wrongness of it. Thirty seconds later I was not bracing. I was looking out the window. The data says I was right to relax: across two hundred and twenty million miles, the cars involved in this experiment cause a small fraction of the serious crashes a human would have caused over the same roads. I did not need the data. I needed thirty seconds.

None of these people knew what they were doing. That is the thing about the librarian and the college student and, for that matter, about me learning to trust a wheel that moves by itself. The librarian was not building a search engine. The clerk was not training a fraud model. He was making rent. Their competence was not evidence, to them. It was just Tuesday. It became evidence later, to someone else, in a room they never saw — the accident logs, the chargeback data, the accumulated record of a million correct guesses that turned out to be exactly the material a system needed to learn the job and take it.

This is the part that is easy to get wrong. It is not that the human failed and the machine succeeded. It is that the human succeeding, over and over, in full view, was the demonstration that the job could be learned. You do not automate a task nobody can do. You automate the one being done well enough, often enough, for long enough that the pattern becomes visible. Doing the job right was never neutral. It was the case being built.

Which brings me to a woman I will call the lawyer, because there are thousands of her and none of them are exactly her. She has a laptop open at her kitchen table. She logs into a dashboard belonging to a company that pairs credentialed people with the AI labs that need them — a doctor here, a banker there, a corporate attorney with fifteen years of contract law behind her. She reads a model’s draft of a merger agreement and marks where it reasons like a first-year associate instead of a partner. She rewrites a clause. She explains, in the margin, why the model’s version would get laughed out of a negotiation. She is paid well for this. More, some weeks, than she billed certain clients.

She knows exactly what she is doing. That is the difference between her and the other three. The librarian did not know she was leaving a trail. The clerk did not know his good judgment would become someone else’s weights. I did not know, thirty seconds into that ride, that I was participating in anything at all. The lawyer knows. She is being paid, by the hour, at a rate that respects her expertise, to make her expertise legible enough that it no longer requires her. The company she works for has a name for this. They call it the reinforcement learning economy, which is a tidy way of saying: teach it everything, and then it will not need to call you back.

She does the work anyway. The rate is good. The work is interesting, in the way that teaching is interesting — you learn what you know by trying to say it clearly enough for someone else to use. Nobody is lying to her. The dashboard does not pretend to be anything other than what it is. She logs off at the end of the session the way anyone logs off after a long day of being excellent at something, tired in the specific way that comes from careful work, and she does not, from what I understand, spend the evening thinking about what she has just fed into the machine.

I keep coming back to the rotary dial. Somebody dialing H-O-O-T-O-W-L at midnight in 1973 could not have imagined the lawyer at her kitchen table. But the shape is the same, if you look at it long enough. A person answers a question well. The answering becomes a record. The record becomes a system. The system answers next time. Nobody in the room ever decided this was the plan. It just turned out, every time, to be the plan.

Categories
AI

The Quiet Setup: MacSparky’s Robot Assistant and the Unfair Advantage Still Available

A single X post caught my attention this week. It described something quietly happening among a small group of solo professionals. They aren’t working longer hours or grinding harder. Instead, they’ve built a particular kind of setup around AI that carries much of the load.

While most of us still treat powerful models as clever search bars—typing questions and copying answers—these folks have given the AI a rich folder of context, a briefing file that orients it to their world, connections to their tools, and routines that let it produce real work on its own. The result can look like the output of a small team. From the outside it reads as talent or luck. Up close, it’s mostly architecture.0

The post (from @zephyr_hg) emphasized that this advantage remains available because most people haven’t yet made the shift from one-off prompting to building persistent systems. It landed with me because it echoes so closely the practical territory David Sparks (MacSparky) has been mapping for months in his Robot Assistant Field Guide.

MacSparky’s Approach: From Chatbot to Persistent Colleague

David’s work centers on building a true personal assistant using Obsidian (for a local, plain-text knowledge base) and Claude (in its file-aware “Cowork” or project capabilities). The system isn’t a chatbot that forgets everything between conversations. It’s designed to remember your projects, preferences, and people; triage email in your voice; handle morning briefings; track tasks; process documents; and support weekly reviews—freeing you from what David calls the “donkey work.”

The key ingredients will sound familiar to anyone who read that X post:

  • A dedicated context layer (your Obsidian vault or structured folder) holding the details of how you work.
  • Briefing/instruction files that tell the model who you are and what good looks like.
  • Integrations that connect it to email, calendar, files, and other tools.
  • Skills and routines that turn one-time intentions into repeatable, low-friction action.

David has been refreshingly transparent about the journey. He experimented earlier with more fully autonomous agents and even shut one down after learning what felt reliable and aligned. The Robot Assistant Field Guide distills those lessons into videos, workshops, templates, and a starter kit that lets people build without needing to code.

Why This Matters Now

Both perspectives point to the same shift in stance: moving from “How do I prompt better today?” to “What kind of system do I want running alongside me every day?”

For me, at this stage of life, that question carries weight. I’m not chasing maximum output for its own sake. I want arrangements that protect attention and energy for what actually matters—deep reflection, family history work, thoughtful investing, writing that might be useful to others, and simply being present. A well-designed AI setup doesn’t just save minutes; it changes the texture of the day by reducing context-switching and repeated explanations.

It feels like finding a productive seam in the current moment of AI evolution—one of those hidden transitions where leverage quietly compounds if you’re willing to build the architecture.

The Door Remains Open

The encouraging message in both the X post and David’s teaching is that this isn’t locked behind rare talent or expensive infrastructure. The models are accessible. The patterns are becoming clearer. What’s required is the decision to treat AI less like a toy and more like a colleague you’re willing to orient and trust with real work.

I don’t have my own “robot assistant” fully built yet. I’ve been experimenting with custom agents, structured daily scans, and ideas like “The Observatory” for signal synthesis. Reading these sources side-by-side sharpened my sense of the next layer: giving the system a proper home, clear instructions, and meaningful recurring work.

If you’re a solo professional, creator, or lifelong learner feeling the press of too many small tasks, this is worth exploring. Start small. Build a modest context folder. Write a briefing file that captures how you think. Experiment with one routine. Iterate from there.

The setup that outworks the grind isn’t magic. It’s deliberate, learnable, and still wide open.

What setups are you experimenting with these days? I’d love to hear in the comments or on X.

Categories
Menlo Park Serendipity

Two Kinds of Efficiency

The fog hadn’t lifted yet over Sharon Park, the kind of gray that Menlo Park wears many June mornings like it’s embarrassed to admit the sun is up there somewhere, and I was on my usual loop around the pond when I noticed in the distance that the goats were back. And one more thing too. I stopped.

On one side: forty, maybe fifty goats, heads down, working a hillside of dry summer grass like a crew that had done this job a thousand times, because they had. The city brings them in every year around now, before fire season, to eat down the fuel load that nobody wants to mow. White ones, brown ones, a few with horns curling back like something out of a hieroglyph. They don’t look up much. A goat eating is a goat with one job and no curiosity about yours.

On the other side, maybe forty yards past them, through the wire: a Waymo. White, sensor pod spinning slow on the roof like a lighthouse that had wandered inland and gotten confused about its purpose, parked at the curb with nobody in it. Just sitting there. Idling, if a thing with no engine can idle. Waiting on a fare, or waiting on nothing, the way these cars do now, patient in a way that doesn’t read as patience because there’s no face attached to it.

I stood looking for longer than the moment deserved, the way you do when something hands you a thought before you’ve earned it. I remembered I should take a photograph.

Here is what struck me, eventually: both of them were efficient. That’s the word that kept showing up, uninvited. The goats are efficient in the oldest way there is — they convert a problem (too much dry brush, a fire waiting to happen) into a solved problem, using nothing but appetite and stomachs and several thousand years of being bred for exactly this. Nobody programmed a goat. A goat doesn’t have a model. A goat has a memory that goes back to whatever the last hillside tasted like, and an instinct that says eat that one next, and that’s the whole operating system.

The Waymo is efficient in the newest way there is. Lidar instead of appetite. A map instead of memory. It doesn’t get bred for the job, it gets trained for it, mile after simulated mile, until eventually you can park it at a curb in a quiet park and trust it not to do anything stupid. It was, in its way, doing the same thing the goats were doing — converting a hard, slightly dangerous task that used to require a person’s full attention into something that just sort of happens now, off to the side, while everyone gets on with their morning.

I’ve spent a fair amount of my working life around payments systems and fraud models, which is its own quiet machinery — systems built to notice the thing before the thing becomes a problem, the same job the goats were doing on that hillside, eating the grass before it becomes a fire. So maybe that’s why I stood looking longer than I meant to. I recognized the shape of it, even though one side of the fence had hooves and the other side had a sensor array worth more than my first house.

What I didn’t expect was how unbothered each side seemed by the other. The goats did not care that there was an expensive autonomous vehicle parked within sight of their breakfast. The Waymo, for its part, did not care about anything, which I suppose is the whole point of it — it isn’t built to care, only to notice, and the goats had registered exactly zero on whatever sensor suite decides what’s worth noticing. Two systems, separated by maybe forty yards and several thousand years of technological distance, each one going about its business with total indifference to the other’s existence.

I used to think the line between old world and new world would announce itself — some clean morning where you’d wake up and the future would have visibly arrived, banners out, the old thing retired with a gold watch. It doesn’t work that way, it turns out. It works like this: a fence, some goats, a car with nobody driving it, and a guy on his usual walk who happens to notice that both of them are quietly, competently doing a job that fire season requires somebody — or something — to do.

I kept walking. The goats kept eating. The Waymo, as far as I know, was dispatched somewhere, picked up whoever needs a ride, sensor pod turning over the same hill the goats had already half cleared. Two kinds of efficiency, on either side of an electrified wire fence, neither one impressed by the other, both of them right.

I don’t know what to do with that, exactly, except to write it down and remember it. Some mornings my walk gives me exercise. Some mornings it gives me a simple memory I didn’t ask for, standing there looking.

Categories
AI Startups

A New Reason to Launch

“Before you launch, the speed you can build is now mainly limited by your imagination in what you tell AI. After you launch, the AI can watch your users and make improvements on its own.”
— Jared Friedman, Y Combinator

Jared Friedman watches hundreds of founders a year navigate the gap between idea and launched product. He notices patterns the rest of us miss. And what he’s describing above is not an incremental improvement in how software gets built. It is a change in the nature of the advantage.

This is a different kind of liberation than founders have known before.

The old liberation was launch early and the market corrects your wrong assumptions. Humbling, but useful. You were still the one doing the correcting, late at night, rewriting the onboarding flow based on what the data told you.

The new liberation he’s describing is something closer to multiplication. You launch, and now there are effectively more of you. The AI is watching session replays you’ll never have time to watch. It’s noticing the drop-off after step three that you’d have caught in month four. It’s holding the pattern of a thousand user paths simultaneously and asking what they mean. Your imagination seeded the thing. Reality is now feeding it.

That observation redraws the map cleanly. Pre-launch and post-launch used to differ in degree — you knew more after than before. Now they differ in kind. Pre-launch you are the sensing organ. Post-launch you’ve grown new ones.

The founders who feel this most viscerally, I suspect, are the ones building alone or in pairs — the people for whom every previous era of building had a hard ceiling imposed by human hours. They could only read so many support tickets. They could only run so many experiments. The ceiling is lifting and the feeling is of a room getting larger.

The core advice hasn’t changed. Paul Graham was saying “launch early” twenty years ago and it was true then. What’s changed is the reason underneath it — the mechanism that makes it true now is nothing like the one he had in mind.

The advice is twenty years old. There is a new reason and it is brand new. Most people haven’t noticed the swap yet. But they will.

That window does not stay open long.

Categories
AI Stanford

The Unit of Production Just Collapsed

The lecture was a Stanford CS session, AI-native companies, Garry Tan walking through what it now takes to build something. He’d rebuilt his old startup, Posterous, in five days on a modest Claude plan. A thing that once required a team and a runway. He said it matter-of-factly, the way you describe something that’s already obvious to you and hasn’t yet reached everyone else.

The argument Tan and his colleague Diana Hu were making wasn’t really about AI. It was about the economics of effort — specifically, what breaks when the cost of turning an idea into a working thing falls by an order of magnitude.

Their framing: AI-native organizations running as closed-loop systems, agents with access to the real artifacts of work, able to iterate without the error-accumulation that comes from handoffs and headcount. Revenue-per-employee ratios of a million dollars or more, with live examples already in the YC portfolio. Document processing, logistics, voice agents for specialized workflows.

What I kept hearing underneath all of it was a quieter claim: the mental model of what a startup requires is wrong.

Or rather, it’s right about the past and increasingly wrong about the present.

The assumptions embedded in “I can’t do this alone” or “we’d need to hire for that” or “we don’t have the bandwidth” — those are load-bearing assumptions, and the load is shifting.

I have some small version of this — not as a founder, but as someone who retired into curiosity. The blog, the reading, the daily effort to keep up with what’s moving: each one is a practice in staying oriented while the map keeps changing.

What I notice is that the constraint has shifted. It’s not information anymore. It’s not even tools. It’s the capacity to ask better questions of the abundance, to know what matters when everything is accelerating.

That’s the thing I find unsettling, yet also genuinely interesting: the skills that remain irreplaceable are the hardest ones to teach, and the hardest to evaluate in yourself. Knowing what matters. Recognizing when an output is almost right and almost wrong. Setting direction in ambiguous conditions and being willing to be wrong about it. These were always the valuable things. They were just obscured by all the coordination overhead that surrounded them.

The students in that Stanford course were asked to build something called a One-Person Frontier Lab — use the best available tools to extend your own reach over ten weeks. It’s framed as an academic exercise. It doesn’t feel like one.

But I’m not building. I’m mostly watching, and thinking about what this radical new fermentation does to everything downstream — to labor markets, to what a company even is, to how we’ll organize work and meaning when the old unit of production no longer applies. Those are slower questions. But they’re the ones that feel urgent to me.

The old excuses are getting lighter. Not that everything is possible — but that the weight of the usual constraints has changed.

What you choose to build, and whether you choose to build it at all, is more purely a decision than it used to be. That’s either clarifying or terrifying, depending on the day and my mood.

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 Programming Software Work

The Scarcest Thing

Garry Tan woke up at 8 a.m. after sleeping at 4. Not because he had to. Because he wanted to see what his workers had done overnight.

The workers are AI agents. Ten of them, running in parallel across three projects. And something about that sentence — wanted to see what they’d done — keeps stopping me. That’s not the language of someone using a tool. That’s the language of someone managing a team.

Tan gave a name to the state this puts him in: “cyber psychosis.” He said it as a joke. But the joke has an insight in it. He’s not describing addiction to a productivity app. He’s describing a shift in what it means to do creative work — the strange vertigo of becoming a director when you’d always been a laborer.

I’m retired. I watch this from the outside now, which is its own kind of vantage point. For most of my career, the path from idea to working product ran through people — through hiring and managing and the slow accretion of execution capacity. You had the vision or you didn’t, but either way you needed the team. The idea and the means of making it real were, structurally, separate things. The gap between them was where companies lived.

What Tan is describing is that gap closing.

The thing he built — gstack, his open-sourced Claude Code configuration — got dismissed in some quarters as “just prompts.” And it is just prompts, in the same way that a conductor’s score is just notation. The abstraction is the invention. What he encoded is a model of how a startup team thinks: the CEO who interrogates the why before a line of code gets written, the engineer who builds, the paranoid staff reviewer who looks for what breaks. Each role blocks a different failure mode. Blurring them together produces, as his documentation puts it, “a mediocre blend of all four.”

That’s an organizational insight. It has nothing to do with code.

Tan described being a “time billionaire” — not because his biological clock had slowed, but because he can now purchase machine-consciousness-hours. The bottleneck of implementation, which has governed every creative project since the beginning of creative projects, is dissolving for those who know how to direct.

The scarcest thing is shifting. It’s no longer the hours of execution. It’s the clarity of intent — knowing what you want to build and why the journey matters, before any of the workers start moving. That’s harder than it sounds. For decades, most of us could muddle through in the making of it. The act of building taught you what you were building. Now the making is cheap, and that shortcut is gone.

For someone watching from retirement, that’s not a small thing to absorb. The model I internalized over a long career — that ideas become real through sustained organizational effort, through teams and timelines and the grinding work of execution — is being revised faster than I expected. Not invalidated. Revised. The judgment still matters. The taste still matters. The why matters more than ever.

It’s just that the how has found new hands. Many of them. More than any team I ever assembled, available the moment the intent is clear enough to direct them, gone when the work is done. The constraint was always the hands. It turns out it was always the knowing.

Categories
AI AI: Large Language Models Anthropic

Breakout

Jack Clark doesn’t panic easily. He spent years at OpenAI watching capabilities inch upward, then left to co-found Anthropic, and has been writing his Import AI newsletter long enough to have developed — and been wrong about — many priors. So when he publishes an essay saying he has reluctantly arrived at a 60% probability that fully automated AI R&D happens by the end of 2028, the word “reluctantly” deserves some weight.

His essay, published last week and titled “Automating AI Research,” isn’t a press release or a fundraising pitch. It reads more like a man thinking out loud at the edge of something large. “I don’t know how to wrap my head around it,” he writes, which is a notable thing to say publicly when you are one of the architects of the thing you can’t wrap your head around.

The argument is built from benchmarks — not any single one, but a mosaic of them assembled to reveal a trend. SWE-Bench, the test that measures an AI’s ability to solve real GitHub issues, was at roughly 2% when it launched in late 2023. A recent Anthropic model sits at 93.9%, effectively saturating it. METR’s time-horizon plot tracks how long an AI can work independently before needing human recalibration: 30 seconds in 2022, 4 minutes in 2023, 40 minutes in 2024, 6 hours in 2025, 12 hours today. The trajectory, if it holds, suggests 100-hour autonomous work sessions by the end of this year.

Clark marshals similar progressions across AI fine-tuning, kernel design, scientific paper replication, and even alignment research itself. His throughline is the same in each: AI is now genuinely competent at the unglamorous scaffolding of AI development — the debugging, the experiment runs, the parameter sweeps, the code reviews. And crucially, it can now do these things not just faster than humans, but for longer, with less supervision.

There’s a Thomas Edison quote at the center of the essay: “Genius is 1% inspiration and 99% perspiration.” Clark’s claim is that AI has become very good at the perspiration. The question of whether it can supply the inspiration — the paradigm-shifting insight, the Move 37 — remains open. But he argues it may not need to. Most of what has moved the AI field forward has been sustained, methodical work, not lone flashes of genius. If you can automate the 99%, you have something that compounds.

There’s a data point that makes Clark’s argument feel less like forecast and more like dispatch. Last month Boris Cherny, who runs Anthropic’s Claude Code, disclosed that he hasn’t written a line of code by hand in more than two months. Every pull request — 22 one day, 27 the next — written entirely by Claude. Company-wide, roughly 70–90% of Anthropic’s code is now AI-generated. Anthropic’s stated position: “We build Claude with Claude.” The loop Clark is describing as a probability by 2028 is already running, at least partially, today.

The word Clark uses for the threshold he’s describing is not “singularity” or “AGI.” It’s quieter than that. He calls it “automated AI R&D” — the point at which a frontier model can autonomously train its own successor. It’s a specific, falsifiable thing. And he puts a number on it: 60% by end of 2028, 30% by end of 2027.

I’ve been writing about the dark software factory and the 3D printer that prints better printers, finding metaphors for what seems like an inexorable process. Clark’s essay is a different kind of writing about the same thing — the primary source document, the engineer’s log, the inventory of evidence. Reading it is a little like watching someone carefully pack boxes before a move. Each individual item seems manageable. But there are a lot of boxes.

What he’s describing — if the trend holds — is not a feature or a product launch. It’s a breakout. The moment the loop closes and the system starts building itself. He’s not certain it happens. He just thinks it’s more likely than not, and he thought you should know.