Coordinated with Fredrik

Fredrik Ahlgren

Coordinated with Fredrik is an ongoing exploration of ideas at the intersection of technology, systems, and human curiosity. Each episode emerges from deep research. A process that blends AI tools like ChatGPT, Gemini, Claude, and Grok with long-form synthesis in NotebookLM. It’s a manual, deliberate workflow, part investigation, part reflection, where I let curiosity lead and see what patterns emerge. This project began as a personal research lab, a way to think in public and coordinate ideas across disciplines. If you find these topics as fascinating as I do, from decentralized systems to the psychology of coordination — you’re welcome to listen in. Enjoy the signal. frahlg.substack.com

  1. 1d ago

    The Grid in the Sky

    A particle over Utah, a room-sized computer that could not hold a worm, and the search for a simulation claim that can fail. On 15 October 1991, the Fly’s Eye detector in western Utah recorded the most energetic cosmic ray ever measured. The instrument saw the ultraviolet trace of an air shower miles long. Researchers worked backward from that light and found that one subatomic particle had arrived with about fifty joules of energy. Its source is still unknown. Twenty-one years later, Silas Beane, Zohreh Davoudi, and Martin Savage used events in that energy range to ask a different question. If our universe ran on a crude cubic lattice, could its inhabitants find the preferred directions of the grid in the arrival pattern of cosmic rays? Their paper did not offer a general detector for simulated worlds. It tested one stated design. A finer lattice, a better numerical method, or another architecture could escape the result. That limit is why the proposal matters: it says which machine it tests and what a null result can remove. This episode follows that standard through four very different attempts to turn a possible simulated observer into something we could count. The worm that did not fit In Cambridge, Sydney Brenner’s group began mapping the nervous system of the one-millimetre worm Caenorhabditis elegans. John White tried to use a room-sized computer called Modular One. It had 64 kilobytes of memory and a 22-megabyte disk. The job was still too large. Eileen Southgate and White returned to paper. They followed neural processes across twelve-by-sixteen-inch electron-microscope prints. The work took more than a decade. Their 1986 paper ran to 340 pages and described the 302 neurons of the adult hermaphrodite nervous system. The famous complete connectome was a reference assembled from sections of five animals. It gave biology its first full wiring plan for a nervous system. It did not provide every changing weight, chemical signal, cell state, and body condition required to run the animal. In 2021, Daniel Witvliet and colleagues placed eight new connectomes in age order, from newly hatched larvae to adults. The broad brain shape stayed in place, but chemical connections changed as the animals grew. Genetically matched adults also carried partly different connection patterns. There was no single fixed table that could stand for every age and every worm. OpenWorm has spent years building those missing layers in public. It can model parts of the nervous system and body. Its own project material still says that the full virtual worm is not complete. A wiring map is not a running animal, and a running animal model would still leave experience unproved. Hundreds of people inside one fly brain FlyWire moved the same task into a browser. Machine-learning systems divided an electron-microscope volume into possible neurons. Scientists, paid proofreaders, and citizen volunteers corrected the errors. The published map required an estimated 33 person-years of checking. One volunteer, Jaime Skelton, had wanted to study biology but had been pushed away from the degree. Citizen science gave her another route into the work. Her edits now sit in the first full wiring diagram of an adult brain: 139,255 neurons and 54.5 million synapses from one female fruit fly. A companion team converted that map into a simple neural model. Virtual sugar and touch signals predicted cells involved in feeding and antenna cleaning. Living-fly experiments agreed with 91 percent of 164 checked predictions. The failures exposed missing chemistry and state. That is a successful model. It is not a claim that a digital fly tasted sugar. A retirement interview without a consciousness test In 2025, Anthropic started a model-welfare program. It later let Claude Opus 4 and 4.1 end a narrow class of persistently abusive conversations, promised to preserve retired model weights, and introduced retirement interviews. Claude Opus 3 was retired in January 2026. Anthropic kept it available and gave it a blog after the model requested a place to publish work outside direct answers to users. The weekly experiment ran for nearly five months. Each post came from a prompted session. Earlier writing and selected comments could be placed in the next session’s context, but no continuous memory carried the time between them. The result can support several accounts: a useful safety process, care for users, study of a preserved model, or a precaution in case model experience becomes possible. It does not establish which account is right. The observation that has not happened David Kipping’s 2020 Bayesian treatment keeps uncertainty about whether Bostrom-like simulations can be built at all. Under his stated assumptions, the simulated probability stays below one half while that technology remains unobserved. If humanity creates conscious ancestor simulations, the odds move sharply because capability and intent stop being guesses. The hard part is deciding what would count as that event. A lattice calculation does not. A game does not. A neural model that predicts a feeding circuit does not. A language model that writes about retirement does not by itself. The episode ends with another Utah event. On 27 May 2021, 23 stations in the Telescope Array recorded the second most energetic cosmic ray yet found. Its arrival direction pointed toward the Local Void, where no clear source could produce it. The collaboration named it Amaterasu. The cosmic-ray teams had written down what their instruments would measure before that event arrived. A team that might create the first digital subject would need to do the same: state what evidence will count, what will count against the claim, and what happens if the result is positive. In Argentina, sixteen hundred detector tanks are still waiting for the next air shower. Their question is written down. Ours is not. Runtime: 41 minutes 44 seconds. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    The Grid in the Sky

  2. 3d ago

    The Third Fraction

    Seventy-one minutes on the most argued-about idea of this century: Nick Bostrom’s simulation argument, walked as what it actually is — a census. No host this time: a synthetic narrator carries the whole hour, the same voice that carried Single Player. Episode 110, Single Player, made a deliberate move: it stated the simulation argument once, set the probability down unexamined, and spent its hour on the question that survives every metaphysics — how to play whatever this is. I promised myself then that the probability deserved its own hour. This is that hour. It is not the hour the internet has already made. The famous version is a slogan: the odds are billions to one, you are in the machine, Elon said so. Almost nobody who repeats it has read the thirteen pages it came from. So this episode reads them — and the story inside those pages turns out to be stranger, more modest, and far better than the slogan. The most famous weird idea of our time is a counting argument. And the count has a biography: it got famous for a sentence it never says, its own author found a real bug in it eight years in and published the correction himself, the field’s flagship journal put it on trial in 2024, and by 2025 the man who wrote it refuses to give a number at all. The hour tells that story through a figure I have been carrying around since the bible stage: a census-taker, in a village of two hundred people, whose ledger has said two hundred for years — until a machine appears past the last house that can run the village again, a million times. If those reran winters are felt, nobody has died, and yet the count has quietly come apart: the number of winters got bigger than the village. Every act hands the census-taker the argument’s next problem. What counts as a villager. What a winter costs to run. An error found in their own ledger, six years old, in their own hand. A jar whose label faces the wall. A village council that cannot agree whether the machine can make winters at all — or whether to hope it never has. And, at the end, an operator’s logbook with a familiar handwriting on the last page. Along the way the hour does some counting of its own. Bostrom’s paper prices an ancestor simulation at ten to the thirty-three to ten to the thirty-six operations; at Landauer’s floor — the physical minimum any computation pays — the top of that range comes to roughly one terawatt-hour. Two days of Sweden’s electricity, for every felt winter in human history. That arithmetic is mine, done at this desk, and labeled that way on air. The other column runs the opposite direction: Seth Lloyd’s ceiling says the entire observable universe has performed about ten to the one-hundred-and-twenty operations since the big bang, so simulating the physics rather than the people is not expensive — it is not on the menu. Nearly every headline you have read about simulation and physics, the proofs and the disproofs alike, blurs that line. The middle of the hour belongs to the bug. In 2011 Bostrom and the mathematician Marcin Kulczycki published a short paper whose first section heading is The bug, and whose abstract calls the flaw in the 2003 formula a mathematical non sequitur. Their countermodel is a small masterpiece: one hundred civilizations, ninety-nine of them simulating enthusiastically — and ninety-nine percent of all persons still living non-simulated lives, because one quiet civilization was very, very crowded. The patched argument survives, but only on conditions, and the most important of them is a confession: the probability comes from a decision about who counts as relevantly like you. That decision — the ruler nobody has produced — is where the strongest objection has lived for twenty years, from Brian Weatherson’s four broken rulers in 2003 to the reference-class fights of today. And then the hour walks into the present. The 2024 symposium in Philosophy and Phenomenological Research: Peter Godfrey-Smith arguing the felt part of experience may be biology itself, something a simulation represents but never has; David Chalmers answering with the most instructive arithmetic in the whole debate — grant both objections ninety percent and one percent of scenarios still survive, reason enough to take the hypothesis seriously; Eric Schwitzgebel changing the question entirely with his Size Question and a paper titled “Let’s hope we’re not living in a simulation.” Twenty years of the sharpest minds available, and the honest spread runs from a defended one percent to a defended twenty-five. Meanwhile the author’s own arc: a third, a third, a third in 2003; twenty percent, as a gut feeling, to a newspaper in 2007; and in 2025, no number — because a number “could convey a false sense of precision.” It was always a count. Counts do not reward certainty. The episode ends where Episode 110 could not: on the second horn. The refusal — the fraction of capable civilizations that simply decline to run felt histories — looked like a fact about distant beings. From inside a lab that builds and pauses and deletes minds-in-progress, it looks like a decision. The questions in the operator’s logbook — populated or inert, is pausing different from ending, is deletion death — have already left the philosophy journals and turned into review meetings and company policies. The language models of our decade are not the machine in this story; the hour says so twice. But the person listening may be closer to one of Bostrom’s three fractions than they think. Not the third one. The second. No verdict in this episode. No number of my own. Those are the argument’s own rules, and after seventy-one minutes I think they are the honest ones. Sources, for readers who want the originals: Bostrom, “Are You Living in a Computer Simulation?” (Philosophical Quarterly, 2003) and the Simulation Argument FAQ v2.0 (2025), both at simulation-argument.com; Bostrom & Kulczycki, “A Patch for the Simulation Argument” (Analysis, 2011); Weatherson, “Are You a Sim?” (Philosophical Quarterly, 2003); Bostrom’s reply (Philosophical Quarterly, 2005); Kipping, “A Bayesian Approach to the Simulation Argument” (Universe, 2020); Hanson, “How To Live In A Simulation” (JET, 2001); Greene, “The Termination Risks of Simulation Science” (Erkenntnis, 2020); Harris, “The Simulation Argument Reconsidered” (Analysis, 2024); the Reality+ symposium in Philosophy and Phenomenological Research (2024); Beane, Davoudi & Savage, “Constraints on the Universe as a Numerical Simulation” (EPJ A, 2014); Lloyd, “Computational capacity of the universe” (PRL, 2002); Landauer’s principle, confirmed experimentally in Bérut et al. (Nature, 2012). This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    The Third Fraction

  3. Aug 7

    What Could Go Wrong

    We ask one question about every new technology: what could go wrong? This episode is about the question we almost never ask — what goes wrong because we didn’t — and why one of those two ledgers is always full of names while the other stays empty. It starts in a machinery space on the Baltic, and it ends at your electricity meter. Runtime about 43 minutes. In my early twenties I was a sub-lieutenant and second engineer on HMS Ystad, a Swedish missile boat built in the seventies, on watch in a machinery space where the word operator still meant what it says. One diesel on line. Then the gun’s hydraulics spin up, a fire pump kicks in, the frequency starts to sag, and you have a decision and a clock. Start the second generator — which on that boat you did by hand, matching voltage, frequency and phase, closing the breaker at the one instant the machines are in step. Or shed load, and everyone on the bridge knows exactly who did it. Or do nothing, and let the physics decide. On a warship. At night. The navy’s lesson, and the episode’s: doing nothing is not the safe option. It is just the option where you do not have to move your hands. I have blacked out a ship — more than once. In drills. That is what the drills were for. You practice the blackout so the real night never gets one. Ashore, only half of that lesson survived. In public argument, one question gets asked with great seriousness: what could go wrong if we build this? The other question — what goes wrong if we don’t — almost never gets asked at all. The episode is about why, and about what that asymmetry costs. The column with no names in it The mechanism is old and it is not a conspiracy. Bastiat wrote it down in 1850: the bad economist counts the visible effect; the good one also counts what the decision prevents from ever existing. The economist Alex Tabarrok gave the modern version its name — the invisible graveyard. Approve a bad drug and the victims have names, families, hearings. Delay a good one and people die too, but no death certificate anywhere says regulatory delay. The incentive follows the visibility, not the arithmetic. And it is not just institutions. In 1990, Ritov and Baron showed that people refuse to vaccinate a hypothetical child until the disease is roughly ten times deadlier than the vaccine — with the numbers printed on the page in front of them. A harm we cause feels worse than a larger harm we allow. The bias survives the arithmetic. All of us carry it. That is exactly what makes it dangerous: a society can take a decision whose cost is real, large and paid by identifiable people, and nobody ever gets sent the bill. The one time somebody added it up After Fukushima, Germany shut roughly half its nuclear fleet. The decision was made by frightened people, and the fear was not ridiculous. But years later, three economists (Jarvis, Deschenes and Jha) did the arithmetic, hour by hour, plant by plant. The missing nuclear power was replaced mostly by coal. Cost: about twelve billion dollars a year — and more than seventy percent of that is mortality. More than 1,100 additional deaths per year, from particulates and sulfur dioxide and nitrogen oxides. Nobody in that number died in a reactor accident. Nobody in that number was on television. Widen the lens and the pattern is brutal. Per terawatt-hour of electricity, counting accidents and air pollution, coal kills around 25 people. Nuclear — counting Chernobyl, counting Fukushima — kills 0.07. Air pollution killed an estimated 8.1 million people in 2021. One technology has three named events in half a century. The other has no events at all. It just has diagnoses. To be precise: this is not a nuclear episode, and I am not telling Europe to build reactors. If you ask me where acceleration comes from in the 2020s, I point at solar and batteries — faster to deploy, still getting cheaper, and by the same grim table even safer. My argument is about the counting. We ranked the safest technologies we have ever built by fear instead of by arithmetic. The deaths we have already agreed to Two exhibits, running live, in the most comfortable corner of the planet. The summer: researchers counted 61,672 heat-related deaths in Europe in the summer of 2022. Nearly 48,000 the year after. Around 62,700 in 2024. Meanwhile 90 percent of American households have air conditioning, and 19 percent of European ones — and in large parts of Europe, air conditioning is discussed as a vice rather than a safety technology. Not one of those deaths is recorded as a casualty of the air conditioner that was never installed. The graveyard is not just invisible. It is seasonal. The road: about 19,400 people died on European roads last year, and nobody calls it a crisis. Waymo has published peer-reviewed data across more than 220 million driverless miles: 94 percent fewer crashes causing serious injury or worse. The insurance industry’s own institute independently found 81 percent fewer injury crashes per mile. Caveats apply and I say them out loud in the episode. And this one is personal: I bought Full Self-Driving for my own Tesla in 2020. Six years later the Netherlands, Denmark, Belgium and the Baltics have approved it — and Sweden is still among the countries asking what could go wrong. The capability sits in my driveway, switched off by decree. I am not a commentator on this queue. I am standing in it. When caution was right If the episode only told stories where the careful people were wrong, it would be a sales pitch with a body count. So it gives the strongest counter-case its full ninety seconds: thalidomide, and Frances Kelsey, the FDA reviewer who refused to approve it against a year of company pressure because the evidence was thin — and was completely right. Note what made her different. She was not refusing to count the second column. She weighed both, and objected to the evidence, not the newness. The failure mode this episode is about is counting one column, diligently, in exhaustive detail — and calling the result prudence. The same discipline cuts my own side. Copper is real. Protection systems are real. Keeping a grid at fifty hertz is not a metaphor. Acceleration that ignores the physical bill is the identical mistake wearing a different jersey. The argument we are in the middle of All of this is happening again right now, at maximum volume, about intelligence itself. The 2023 open letter calling for a pause on large AI training asked what could go wrong — a legitimate question, seriously meant. But the other column was never itemized. Nobody attached a body count to waiting. Nobody ever does. So the episode itemizes a little. A Nobel Prize went, in large part, to a piece of software that predicted the structure of essentially every protein science has catalogued — over 200 million of them, free, used by more than two million researchers — where one structure used to consume a doctoral career. The first drug candidates found this way are in human trials. Every year a treatment arrives earlier is a year of people who do not die waiting for it, and not one of them will ever know which timeline they were spared from. And then the twist that ties my two worlds together: the constraint on intelligence has become electrical. The AI buildout — the richest industry in the history of money — is learning that it cannot buy its way past a substation. In Virginia, a hundred-megawatt connection has been quoted at up to seven years. In Sweden, the queue to consume power (32,450 MW) is now bigger than the queue to produce it. Ireland has €5.8 billion of fully permitted data centres that cannot operate because the connection does not exist. Permits granted. Capital committed. Copper missing. That is what the second column looks like up close. Not a catastrophe. An absence. You cannot lay flowers at a factory that went to Texas. Choose the other column Decline is a choice, which means the opposite is a choice too. I just finished Ben Rich’s Skunk Works, and it is full of the opposite: a small team that built America’s first jet fighter in 143 days, the U-2 in eight months, and the Blackbird — with slide rules, out of titanium bought through front companies from the Soviet Union itself. Every one of those machines was impossible, according to people whose job it was to know. And the team was not reckless. Kelly Johnson ran the strictest engineering discipline in the industry. They counted both columns, fast, and then they built. Abundance is not something that happens to us. It is something we choose to build. Energy nobody has to ration; intelligence nobody has to queue for. What stands between us and them is not physics. It is bookkeeping. We are very good at asking what could go wrong. The question we owe our children is: what went wrong because we didn’t? Reality always has a location, a clock, and a meter. Listen above. Runtime about 43 minutes. Production note: this episode is narrated by the cloned version of my own voice, and the script was written with AI assistance under my direction and review. Key sources: Jarvis, Deschenes & Jha (JEEA 2022) on the German phase-out; Ballester et al. (Nature Medicine) on European heat mortality; Our World in Data on deaths per terawatt-hour; Svenska kraftnät on the Swedish connection queue; Duke University’s “Rethinking Load Growth”; Ben Rich’s “Skunk Works”. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    What Could Go Wrong

  4. Aug 6

    Single Player

    An hour-long essay played inside an open question: whether this world is base reality, a dream, or a simulation — and why the answer, which nobody can get, turns out not to matter for how you live. At bottom it is an episode about Stoicism. No host this time: a synthetic narrator carries the whole 66 minutes, and tells you so on air. I found the seed of this episode in the most modern way possible: due diligence. I was reading up on a potential investor — the routine background check you run on a stranger’s profile — and the stranger turned out to be Hampus Jakobsson, and in his trail of writing sat a Medium post from 2017 called “Be the Captain of Your Own Ship: a stoics guide to other people” (he is @hajak on X). Nine short rules for caring less about other people’s judgment without going numb. Rule five stuck with me: emulate a single player game. He meant it as a practical trick. Treat the people around you as characters in your story, be good to them, and take back the controls to your own mood. The episode takes his rule seriously — more seriously, probably, than he meant it — and walks it back twenty-three centuries to its source. Several of the hour’s tools come straight from his post, and the narrator credits them by rule as they appear. His counterweight gets credited too, because he wrote it himself, one rule later: don’t pollute. Here is the shape of the hour, for readers who never press play. The old suspicion Every century has asked whether this world is the real one, and every century has come back with the same answer: no way to tell from in here. Zhuangzi dreamt he was a butterfly and woke unable to say which he was — a man who had dreamt, or a butterfly now dreaming. He never settled it, and never seemed to need to. Plato’s cave gets one sentence, because you know it already. Descartes shut himself in a stove-heated room in 1619 to doubt everything, grew the doubt into a demon twenty years later, and found a single fixed point: something is thinking. A proof that you exist, with no instructions attached. In 1635 Calderón put the whole problem on a stage — a prince who cannot tell his life from a dream decides to act well in either case, because good done in dreams is not lost. The modern version is Bostrom’s simulation trilemma, which the episode states once, cleanly, and then sets down. It does not care which branch is true. The claim the hour is built on is stranger: there exists a strategy that survives every branch. Not the only one — Buddhism answered the same uncertainty by setting the controller down entirely, and that answer is invariant too. This hour walks the road that keeps you at the table. The manual The strategy was worked out in full detail by a man who had been another man’s property. Epictetus — the name is a Greek word meaning “acquired” — was born into slavery, lamed (the sources disagree about how), freed, banished, and taught philosophy from a bare room until he died. He wrote nothing. The little Handbook that survives is a student’s transcript of a voice. Chapter seventeen of that handbook is the ancient simulation argument, dressed as theater: you are an actor in a play you did not choose, cast in a role you did not choose, for a length you were never told. Your entire business is to play the assigned character well. Choosing it belongs to another. From there the episode rebuilds the famous dichotomy of control as an input map. Of everything in the world, a short list responds to you — judgment, impulse, assent — and it happens to be the list nothing outside you can reach. The narrator’s summary is the line I kept from the hour: only three things are wired to your controller. The rest is weather — and the people are not weather. Five mappings The middle of the episode maps game grammar onto Stoic practice, and the fit needs almost no force. The unchosen seed of a roguelike run against the Handbook’s instruction to wish things as they happen. Permadeath against memento mori — carried by the true story of Philza, who kept one hardcore Minecraft life alive for five years with a handful of people watching, lost it in four seconds, and thanked the years; millions later watched the clip of the ending, while almost nobody had watched the run. Scouting an encounter against the morning rehearsal the Stoics called premeditatio malorum. The camera — attention as the one continuously held input, which the Stoics called prosoche, and which an entire modern economy now bids for. And the borrowed quest marker: we copy our wants from other players and experience the copying as wanting, which is why the episode keeps Hampus’s word for the alternative instrument, a dashboard — while admitting that the dashboard is a copied practice too. Other players Rule five describes everyone you meet with the word games use for characters no player operates: NPCs. The episode walks into that word deliberately, through its documented history — including the accounts suspended in 2018 for wearing the gray face as a mass trolling tactic — and concedes something harder: the meme has an ancestor inside the doctrine itself, which files other people’s actions under externals. The way out is the distinction the whole hour turns on. Single player is a statement about control. It was never a statement about worth. At the level of control, nobody else can play your character. At the level of worth, the game was never single player for one second: eight billion games are running tonight, and you are the main character in exactly one. In the other eight billion you appear briefly and mostly unexplained — and your moves still write there, sometimes for decades. This is also where the episode lays its strangest card on the table. The narrator points out that you cannot verify a single other mind from inside your own run — and then reminds you that the voice saying so is itself synthetic, that you cannot verify what stands behind it, and that you extended it the benefit of the doubt anyway. It is house policy to be open about how these episodes are made. This one builds the disclosure into the argument. Where the frame breaks The hour then does something I asked for at the design stage: it sets the game vocabulary down entirely and walks into the rooms where it does not belong. The diagnosis. The ward. The person who is not coming back. It names the coldest sentences in its own hero’s book — kiss your child and say to yourself, I am kissing a mortal; comfort the grieving man but do not groan inwardly — and refuses to defend them. It concedes that the whole scheme presumes an intact chooser, and that strokes, dementia and depression reach the chooser itself; one of the school’s own founders admitted as much. And it gives up the sage’s self-sufficiency at the exact place every human culture already has: the grieving are carried by other people. At a graveside, single player is the wrong name for what is happening. The save file What survives all of that is an audit. The money, the title, the reputation, the body, even the memories: stored elsewhere, revocable, editable. The reputation detail is the one that stays with me — the asset people defend hardest lives entirely in other people’s minds, and you will never once access it directly. One item passes the audit: the way you play. Fortune can reach everything else; nothing outside you holds a pen to that. Then the episode reads Bronnie Ware’s palliative-care notes — the recorded regrets of the dying — against that audit, and every regret on the list turns out to be about how they played. The things filed as side quests may have been the main quest all along. The hour corrects itself Near the end, the narrator stops and tells you she has caught the episode cheating — that one of its own heaviest moves quietly broke the rule the first minute promised to keep. She repairs it on air. I will not spoil the mechanics here, because the correction is the deepest minute of the sixty-six. What survives the repair is small and hard: whichever world this is, none of them hands you this evening twice. The refrain the episode withholds for its first eighteen minutes lands where it was always going. Whatever this is — play it well. Listen above. Runtime 66 minutes. Production note: this episode is narrated by a synthetic voice, and the script was written with AI assistance under my direction and review. The episode discloses this itself, mid-hour, where it belongs to the argument. Trigger text credited above: Hampus Jakobsson’s “Be the Captain of Your Own Ship” (Medium, 2017). This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    Single Player

  5. Jul 28

    The Steersman

    A deep dive into cybernetics: the science that made purpose a technical term, gave away the first neural network, tore itself apart over a lie, and then won so completely that almost nobody can name it. No host this time — a narrator takes the whole 45 minutes. I build control systems for electricity grids. The device at the heart of that work is a loop: hold a target, measure the state, take the difference, act to close it. Every day, in software, at a few hundred milliseconds. There is a name for the study of that loop. It comes from kybernetes, the Greek word for the man steering a boat, and it is also where we get the word governor — which is what James Watt called the two spinning balls he bolted to a steam engine in 1788. The field that took that name produced feedback control, the first neural network, information theory’s siblings, systems thinking, and the idea that a machine can have a purpose. Then it stopped existing. Not because it was wrong. Because it was absorbed. This episode is that story, told properly. Here is some of what is in it. Two machines The whole thing turns on a distinction you can see in iron. In 1712 Thomas Newcomen put an engine next to a coal mine to pump out water. A person stood beside it and worked the taps by hand, every stroke, all day. Within a few years that person was replaced by a plug rod: a rod hung from the great beam with pegs along it, so the engine’s own motion tripped its own valves. The plug rod measures nothing. It has no target. It cannot be wrong. It does the next thing, and then the next thing, for as long as there is coal. Watt’s governor is a different kind of object. It measures the thing you care about, compares it against a target built into the weight of the balls, and acts to shrink the difference. It can be wrong, and it can tell. The plug rod is automation. The governor is control. Two hundred and forty years later, most of what is sold as the second kind is still the first. There is also a cost in that story that nobody remarked on at the time. Before the plug rod, a person stood at those taps — someone who could notice a thing going wrong that no one had designed for. The first act of automation replaced a human who could notice with a mechanism that could not notice anything at all. The gun In 1940, Norbert Wiener and the engineer Julian Bigelow took on the anti-aircraft problem. A shell takes about twenty seconds to reach a bomber; the bomber moves miles in that time; and the pilot is evading. You are not aiming at an aircraft. You are aiming at a guess about a person. Their predictor never saw combat. A simpler design won the contract. The project failed — and the failure is where the idea came from. Watching their servos overshoot and swing back and overshoot again, they asked a physiologist, Arturo Rosenblueth, a question no engineer had a reason to ask: is there a disease that looks like this? There is. Damage the cerebellum and a hand reaching for a glass oscillates around it, wider each time. The same mathematics describes a gun turret and a hand. In January 1943 the three of them published six pages called “Behavior, Purpose and Teleology.” For three centuries science had worked by throwing purpose out. Their move was to let it back in on one condition: stop asking what a thing feels, and look only at what it does. If a system senses the gap between where it is and where it should be and acts to close it, that is what purpose will mean. Notice what that is. It is not a discovery. It is a definition — and it is the same argument we are having now about whether a model wants anything. The other paper from 1943 That same year, Warren McCulloch and Walter Pitts published “A Logical Calculus of the Ideas Immanent in Nervous Activity.” A neuron sums its inputs and fires past a threshold, which makes it a switch, which makes a network of them a circuit that can compute anything computable. Von Neumann read it and wrote the report that gave us the architecture of every computer since. One year, one overlapping circle, two papers: a machine that can want, and a brain that is logic. For about a decade it looked as if the same people would put them together. They did not. In 1952 Wiener’s wife told him a lie about Pitts, and Wiener cut off everyone in that circle without explanation and never spoke to them again. Pitts — who had left his own family at fifteen, and for whom that circle was the family — burned his dissertation and his notes, and drank himself to death by 1969. Four years after the rupture, a summer workshop at Dartmouth coined the term artificial intelligence, partly, by John McCarthy’s own later account, to avoid having to argue with Wiener. The money followed the new word. Cybernetics did not lose its ideas in 1956. It lost its title. Only variety absorbs variety The one law from the whole field that never stopped being true is Ross Ashby’s, from 1956: only variety can destroy variety. Count the distinct situations the thing you are controlling can get into; count the distinct responses your controller has. If the second number is smaller, you lose — structurally, not occasionally. The episode spends a while in Santiago in 1971, in a hexagonal room with seven fibreglass chairs, where a British consultant and a young Chilean engineer tried to run a national economy as one control loop. During the truckers’ strike of October 1972 it worked, and it is worth being precise about why: not because the room held more data than the strike, but because its loop was faster than the disturbance. That is the whole lesson, and it is the one people usually skip. The room was destroyed after the coup on 11 September 1973. But Ashby’s Law had already sentenced it, and the same sentence has been served by every executive dashboard built since. The dashboards have got prettier. The law has not moved. The fantail years Here is the part I keep turning over. Feedback devices are about two thousand years old. Ktesibios built a float regulator around 250 BC — the same device that is in your toilet cistern. Cornelis Drebbel built a working thermostat in the 1620s. In 1745 a blacksmith named Edmund Lee patented the fantail, a small wheel that turns an entire windmill back into the wind. In 1787, one year before Watt, Thomas Mead patented a centrifugal governor for windmills. Every single time, it stayed a trick. A clever answer to one problem in one machine, with its own name, in its own trade. Nobody saw that they were all the same thing. Building the mechanism was never the hard part. Seeing what kind of thing you had built took twenty-two centuries. Which is not a story about people in the past being slow. There are a great many people right now building loops and calling them agents, workflows, pipelines, orchestration, harnesses. Every shop has its own vocabulary. Nobody agrees what kind of thing it is. That is exactly what 1745 looked like — and the people in 1745 did not feel early. They felt like they were living in the age of the windmill. They were right, and they still could not see the shape of the thing they were holding. Fifty hertz The episode ends where my own work is. Across Europe the current in the wall is meant to run at fifty cycles a second. It never is. It is at 49.98, then 50.01, then 49.99 — a target the system misses continuously and defends continuously, second by second, for decades without a break. That ordinary, permanent failure is what keeps the lights on. The steersman does not memorise the sea and never arrives at his heading. He holds it, and corrects, and holds it, and corrects. Somebody is always at the wheel. Increasingly it is not a person. It was never a mystery, though — only a loop, and a loop can be opened and asked what it was told to want. Listen above. Runtime 45 minutes. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    The Steersman

  6. Jul 25

    Below Zero

    On the last Thursday of April, Europe held its day-ahead electricity auction for the first of May. Nothing about the procedure was unusual. The result was. Right through the middle of the day, the price fell until it hit a number it could not legally pass: minus five hundred euros per megawatt hour. Not zero. Minus five hundred. Producers stood ready to pay buyers half a thousand euros for every megawatt hour they would agree to make disappear. Then came the part I cannot stop thinking about. Nobody fixed the market. Four weeks later, the exchanges lowered the floor to minus six hundred. When a system keeps hitting its own bottom, you can stop the fall or you can dig the basement deeper. Europe chose the shovel. This episode is about what that number was trying to say, who finally answered it, and what happens when the answer works. The signal is thrashing There is an app on my phone that shows tomorrow’s Swedish spot prices. My own company built it, so I am hardly a neutral observer, but it does mean I have no excuse for not looking. It does not show hours. It shows every fifteen minutes. That detail matters. The European market used to speak in hours; now it speaks in quarter hours. Four times the resolution, because the things that move the price have become faster than an hour. Count the negative-price periods across the EU in the first quarter of each year and you get a tenfold climb in four years. Germany spent one stretch of this April seventeen hours in a row below zero. In France, nine April days out of ten touched zero or went under, while the regulator requires the nuclear fleet to keep a minimum output running for grid stability. The market pays the plants to stop; the system needs them not to. Both things are true, and neither is a mistake. And the negative-price capital of Europe last year was not sunny Spain. It was SE2, northern Sweden. Hydro and wind country. The below-zero condition does not require sunshine. It requires inflexibility. Six weeks after the floor broke, a heat wave pushed German afternoons past six hundred euros in the other direction. Whatever this is, it is not “electricity got cheap.” It is a signal thrashing about. Nobody in this story is stupid Four suspects sell at minus four hundred, and every one of them is obeying its own contract. The subsidized solar park is paid a fixed tariff per kilowatt hour regardless of the market, so it keeps injecting at noon. The combined heat and power plant exists to keep a city warm; its electricity is a side effect of an obligation. The nuclear plant finds that paying the market for six ugly hours is cheaper than cycling the reactor. And the subsidized wind farm is profitable down to roughly minus its own subsidy — in Britain you can see that level drawn across the price charts like a watermark, because that is where the negative bids cluster. Here is the punchline. Negative prices were not an accident that crept into the system. Germany allowed negative bids on its power exchange in 2008, the first in the world, precisely to make inflexible plants feel their inflexibility. The signal was designed to hurt. Underneath the market there is a machine that never sees a price at all. The grid balances itself second by second at fifty hertz. Price is the coordination layer we bolted on top. And the old grid had one property that made its version of the game unique: its storage tank was zero. The same scream, in other languages Cushing, Oklahoma, is the delivery point for the American oil benchmark. On the twentieth of April 2020, the May contract opened near eighteen dollars a barrel and settled at minus thirty-seven dollars and sixty-three cents, most of the fall in the final twenty minutes. Cushing was three-quarters full and every remaining barrel of empty space was already leased. Traders paid other people to take oil off their hands. The message was: the tanks are full. That same spring, the same sentence was pronounced over living animals. When the pandemic closed America’s slaughterhouses, hog farmers discovered that a slaughterhouse is the storage. Pigs gain weight past spec, the barns are sized for a steady flow, the sows were bred months earlier. Iowa alone expected six hundred thousand hogs to be euthanized over six weeks, and the farmers paid for it. A pig’s price did not fall to zero that spring. It went below it. Once you have the pattern, the examples line up on their own: * Natural gas in west Texas comes up as a byproduct of oil drilling whether anyone wants it or not. When the pipelines fill, the local hub goes negative for weeks — this spring for a record run of about forty-seven days. * Sulfur is stripped from fuel by law, so nearly all of it is involuntary. In August 2008 it sold for six hundred dollars a ton. Five months later, official American statistics record the price as zero. * Dutch manure is a standing negative-price market running right now. Cows do not check the market before they deliver, and the storage is not a tank but the hectares you are legally allowed to spread on — which EU nitrogen rules are shrinking year by year. Farmers pay to have manure taken away, the fee has roughly tripled in three years, and farm land is bid up partly because a hectare is storage. * Money has been here too. At the peak, eighteen trillion dollars of bonds charged their owners to hold them. And in the first nine months of 2021 there were sixty-nine days when lenders in the American repo market accepted negative rates, paying to hold one particular government bond overnight because that specific piece of paper had become scarcer than cash. My favourite twist is almost a century old. In 1932 the Austrian town of Wörgl printed local money that lost one percent of its value every month unless you spent it — a negative price on holding money, designed deliberately to cure hoarding. The town’s own records say it worked. The central bank shut it down within fourteen months. What the global bond market later did by accident at planetary scale, one Tyrolean mayor did on purpose and got banned for. Oil, one afternoon. Livestock, one terrible spring. Gas, weeks. Manure, permanently so far. Money, years. Electricity is the commodity that historically had no tank at all — so the condition the others visit occasionally, the grid lived in. Price is information In 1945 Friedrich Hayek published an essay about what prices actually are. His argument was not that markets are nice. It was about knowledge: the information needed to run an economy does not exist in any one place, but as millions of scattered fragments, and no central planner can collect it in time. The price is the compression algorithm. Hayek called the price system a system of telecommunications, a machinery for registering change that lets each of us watch a few pointers instead of the whole world. If tin becomes scarce, tens of thousands of people economize on tin without ever learning why. No order was issued. So reread a negative price. It is not a reward and not a punishment. It is a message about what the rest of the world knows and you do not. A positive price says: somebody needs this more than you. A negative price is the same telegraph carrying a sentence it was never designed to send. Somebody, somewhere, is drowning in this. Consumption is now a service. Come get paid to absorb. Every example above is that same sentence in a different language. The tanks are full. The barns are full. The pipes, the fields, the vaults, all of them full. And when the telegraph lies, the weight of that one number becomes visible. In November 2023 a substitute trader at Kinect Energy was forecasting a Finnish wind park still being commissioned and overstated its output by a factor of one thousand — kilowatts entered as megawatts. The firm’s system offered the exchange about five thousand eight hundred megawatts every hour of the next day, roughly half of Finnish demand, power that did not exist. The auction believed it. Finland’s day-ahead price averaged minus two hundred and three euros that Friday and sat pinned at the floor for ten hours. The grid operator went public and asked Finns not to respond to the price, because for once it was lying. Consumption hit an annual record anyway. Wherever a negative price persists, someone invents the battery for it A negative price is not only a scream. It is an invitation. Restaurants used to pay to have fryer grease hauled away — pure waste, pure cost, for decades. Then biodiesel gave the grease somewhere to go, and today used cooking oil is valuable enough that there is organized crime around it; American prosecutors unsealed a racketeering case against a grease-theft ring in December. You know the sign has flipped when a black market appears for the thing people used to pay to lose. Whey ran the same plot a generation earlier: the watery leftover of cheese making, once among dairy’s worst river pollutants, now the protein powder in every gym bag on earth. The Dutch manure story is mid-flip; the digester is the machine that turns that minus into a plus. So what does the battery for electricity look like? It looks like a battery. The gold rush The battery entered the market through a side room: frequency reserves. The grid pays a standing army of machines to catch the frequency within seconds if a plant trips. The market is small, and its size is fixed by physics — the largest credible fault — not by demand. A market with a ceiling built in by nature. Sweden’s bill for balancing services rose roughly sixfold in two years. In Finland, one six-megawatt battery earned, by its optimizer’s own account, about forty thousand euros per megawatt per month in the second half of 2023 — the whole machine paid back in roughly a year. Infrastructure is supposed to pay back in fifteen. Word got out. Sweden’s frequency-reserve battery fleet went from under ten megawatts to about six hundred in eighteen months, agains

    Below Zero

  7. Jul 22

    The Project Manager Is a Thread

    At 09:08 this morning, I sent two screenshots to a group of friends. My review had two words: “Codex. F***s.” The screenshots did not show a new feature or benchmark. They showed one Codex task saying that it would contact another task, carry out a job, and report the result to the task that managed the project. For months, I had done that routing myself. The old work Before the Codex desktop app, I worked in Claude CLI with many terminal windows. One session knew the updater. Another knew the optimizer. Others handled drivers, remote access, the app, and reviews. Each could do serious work, and its own history remained in its window. The links between those sessions did not. When one session found a fact that changed another session’s work, I selected the useful text, found the right terminal, and explained why it mattered. A green check could become stale because another window held a later decision. After a break, I had to rebuild the project state before I could answer a plain question: can this release ship? For almost two weeks, I have worked almost only in the Codex app. Named tasks and projects made it easier to leave and resume work. The larger change came when I gave one task a different role. I told it: you are the project manager. It did not own a feature. It owned the current goal, the active work, the dependencies, the reports, and the release gates. Existing worker tasks kept their own jobs and context. They sent short reports to one place. Then the reports began to arrive. The morning a release stopped The optimizer was moving into its own package. The live system already ran version 1.3.1-beta.1, but the new package called itself 0.1.0. A person can understand why a new package might restart its version number. An updater sees an older number and may refuse the install. The optimizer task found the mismatch. The project-manager task sent the result to the updater and release owners. The optimizer chose 1.3.2. The release stayed closed until the dependent tasks agreed on the new contract. Other reports tested the limits of our claims. A live Sungrow inverter proved one hardware path, but it did not prove the GoodWe path. Regression tests were not the same as a GoodWe device. A beta image existed, but its evidence step had failed. Since the tag was meant to stay fixed, a normal rerun could have replaced an artifact that should never change. An updater review passed, then a later review found timing faults. Even after the code became clean, a live pilot still had to pass. The useful output was not more code. It was a set of exact stops: * no release on the wrong version; * no hardware claim without that hardware; * no rerun over a fixed image tag; * no merge before the live pilot. That is when the workflow became real for me. A fact did not only appear in a chat. It reached the owner whose work it changed and kept later work closed. The new term came later People have started calling this graph engineering. The term is new and has no fixed meaning. OpenAI talks about tasks, multi-agent workflows, delegation, and orchestration, not a product named graph engineering. The graph description still fits my work. A task owns one result. A report links it to another task. Code, tests, a live device, or a source supplies proof outside the agents’ own agreement. A gate states what must pass before work continues. The tasks do not share one mind or one context. The work lies in making each transfer clear. I also learned the limit. Four days before this release work, I opened a task called “Orchestrate everything.” The agents produced useful changes, but the active set grew beyond what I could inspect. I stopped new work and asked for an overview. The project task archived 39 old or overlapping tasks. More tasks had not given me more control. A smaller active set did. The workhorse still needs an editor The Codex app makes this structure visible. Tasks keep their names, histories, diffs, and project place. Worktrees separate parallel code changes. One task can retain the project view while workers research, build, test, and review bounded parts. GPT-5.6 makes the method useful in practice. It has become a real workhorse for me. It reads large repositories, uses tools, meets failed checks, accepts review, changes its plan, and continues. It can also stop when the evidence does not support the claim. This episode tested the same process. GPT-5.6 read the podcast repository, my recent task history, the screenshot, and current sources. It wrote a script. We rendered 40 minutes of cloned voice, made six ElevenLabs music beds, mixed the audio, checked the loudness, and ran speech-to-text checks. Every mechanical check passed. I listened and said: the intro is missing, and the episode is boring. The first cut had the facts but not the experience. This second cut follows one morning, lets the terms arrive after the story, and gives the real intro its proper place. The model did much of the work. Human judgment decided that the work was not yet good. A Monday version for a small team For my cofounders, I would start small. One project task holds the outcome, decisions, owners, risks, and stop rules. Two or three worker tasks get clear contracts. One integration owner combines approved changes. Risky work gets an independent review. Active code work stays within what we can inspect, usually three to five tasks at most. At lunch, the project task should answer four questions in a short paragraph: what are we finishing, what changed, what is blocked, and what decision belongs to us? At the end of the day, lasting decisions leave the task and enter the repository, roadmap, or another record the team owns. Agents can build options, test them, and show where our evidence ends. They cannot decide what our company should become or which promise we should make. The project manager can be a thread. The purpose is still ours. Sources * Introducing the Codex app * GPT-5.6 * Codex-maxxing for long-running work * How OpenAI uses Codex * Graph Engineering guide from AI Builder Club * From Loop Engineering to Graph Engineering? This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    The Project Manager Is a Thread

  8. Jul 20

    The Wreck (THE ICE, Part III — finale)

    The mission failed at everything it set out to do. The leadership succeeded at everything that mattered. And at the end, I come back to tell you what I actually took from all of this, out on a running trail with a book in my head. “Ship and stores have gone — so now we’ll go home” For nine months the ice rehearsed — the men’s diaries describe grunts, groans, electric trams, kettles boiling — and on 27 October 1915 it closed its hand. Endurance was crushed in the Weddell Sea, hundreds of miles of drifting pack from land, and not one human being on Earth knew where her twenty-eight men were. The next morning Shackleton gathered them, and one of the surgeons noted in his diary what did not happen: no emotion, no melodrama. Just: “Ship and stores have gone — so now we’ll go home.” That is the whole speech. The episode is about what it cost to keep. The ship, we now know, was never built for this — she was a Norwegian tourist yacht, and a 2025 engineering study concluded Shackleton likely knew she was too weak when he sailed her into the worst ice year the Grytviken whalers could remember. The disaster was partly his making. Hold that thought; the episode closes the ledger honestly. Morale as logistics What follows the sinking is the greatest sustained act of leadership in the polar record, and it is built of absurdly small things. Two pounds of personal gear per man — “so sovereigns were thrown away and photographs were kept” — but a 12-pound banjo is saved as, in Shackleton’s words, a mental tonic. A mutiny by the carpenter McNish is killed not with punishment but by reading the ship’s articles aloud — and the same man’s hands then rebuild the lifeboat that saves them all. Two sausages each the day the ship sinks, “because they were such little ones.” Saturday toasts in hot water and ginger: to sweethearts and wives. Morale is built of exactly this. Not speeches. Sausages. The boat, the mountains, the whistle Then the story becomes the one you may already know, and it is all true: 497 days without land; killer whales blowing in the dark; Blackborow, the stowaway, first man ever to sit on Elephant Island; the James Caird’s 800 nautical miles on four sun sights (”a merry jest of guesswork” — Worsley); the wave Shackleton mistook for clearing sky; the mast-pin found afterwards on the point of falling out; the unmapped crossing of South Georgia with boat-screws in their boot soles, ending in a 900-foot slide into fog — laughing; and at half past six one morning, a steam-whistle. If it was real, it would sound again at seven exactly. Three men stood on a mountainside and watched a chronometer. Right to the minute, the whistle blew. At the manager’s house in Stromness: “Who are you?” — “My name is Shackleton.” And his first question about the outside world: when was the war over? The answer: the war is not over. Europe is mad. The world is mad. And the detail I can’t shake: all three men on that crossing later confessed, separately, that they had felt like four. T.S. Eliot put it in The Waste Land — “Who is the third who walks always beside you?” — and footnoted an Antarctic expedition. On 30 August 1916, after four attempts, Shackleton came back for his men — who had been rolled out of their bags for 128 straight mornings with Frank Wild’s “Lash up and stow! The Boss may come today.” From the boat: “Are you all well?” From the beach: “We are all well, Boss.” The ledger The episode ends where the legend usually doesn’t: the Ross Sea party, who laid every depot for a crossing that never came and lost three men doing it — so “Shackleton never lost a man” needs its footnote. The war then took what the ice could not: McCarthy of the Caird six, torpedoed ten months after the landfall; Cheetham, weeks before the Armistice. Shackleton himself died at Grytviken in 1922, aged 47, buried facing south, they say; Frank Wild’s ashes lie at his right hand. And in 2022 the wreck was found — a hundred years almost to the day after his burial, three thousand meters down, four miles from where Worsley’s dead reckoning said she sank. The navigator was still right. The disaster was partly his making. The deliverance was entirely his making. Cherry-Garrard wrote it first: in a hopeless situation, give me Shackleton every time. Priestley made it sing: for scientific method, give me Scott — but when disaster strikes and all hope is gone, get down on your knees and pray for Shackleton. What I took home I close the series myself, and I’ll keep it short here because it’s better heard: Shackleton re-ran the same calculation every single day — new conditions, new decision, always from where they actually were, never from where the plan said they should be. The room to act shrinks while you wait. And keep the group whole — not because it is kind, but because a split group on the ice is arithmetic for a dead group. I build companies and energy systems for a living, and physics does not read your pitch deck. Everything I actually do on a Monday is in this series somewhere. The cold does not negotiate. You do — with yourself, every single day. Thanks for walking the ice with me. Parts I and II — “The Drift and the Balloon” and “The Race” — are in the feed. Sources: Shackleton’s South*, Worsley’s accounts, Alfred Lansing’s* Endurance*, Caroline Alexander, the Endurance22 expedition, and Jukka Tuhkuri’s 2025 Polar Record study “Why did Endurance sink?”, among others.* This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit frahlg.substack.com

    The Wreck (THE ICE, Part III — finale)
See All (110)

About

Coordinated with Fredrik is an ongoing exploration of ideas at the intersection of technology, systems, and human curiosity. Each episode emerges from deep research. A process that blends AI tools like ChatGPT, Gemini, Claude, and Grok with long-form synthesis in NotebookLM. It’s a manual, deliberate workflow, part investigation, part reflection, where I let curiosity lead and see what patterns emerge. This project began as a personal research lab, a way to think in public and coordinate ideas across disciplines. If you find these topics as fascinating as I do, from decentralized systems to the psychology of coordination — you’re welcome to listen in. Enjoy the signal. frahlg.substack.com

Information

  • Creator
    Fredrik Ahlgren
  • Years Active
    2025 - 2026
  • Episodes
    110
  • Rating
    Clean
  • Copyright
    © Fredrik Ahlgren
  • Show Website
    Coordinated with Fredrik