everyone

what Benjamin a. Mowinckel shares with everyone, to read or to ask.

Benjamin a. Mowinckel is building Alexandria, a company that takes anyone to the best system for their own data in one sitting, so every model they use works from all of it and thinks with them rather than for them. He builds it alone, and he lives in Bergen, in Norway, after six months in San Francisco. Beside Alexandria he keeps four smaller projects. Mowinckels is his home restaurant, Odel is a push for Norway to own its share of the coming intelligence, Reeve is a future practice helping restaurants, and Axioms is how a company brings him in for clarity on the new technology.

how to read this

What he is doing right now, and the most useful way to help — now.
What Alexandria is, what it costs, how it differs from a model's memory, and whether you can have one — Alexandria.
Who pays, the raise, and where he could be wrong — the investor's side.
What the whitepaper and the letter argue — the whitepaper and the letter.
Mowinckels, Odel, Reeve, and Axioms — the five projects.
His X posts, Instagram, Beli, YouTube, Pinterest, and how they connect — where he is in public.
What he believes, and how he thinks — what he believes.
What he has changed his mind about — changed minds.
What he is still working out — open seams.
How to engage him — what he wants from a reader.
What one of his words means — the words he uses.

now

He lives in Bergen now, and Alexandria is his full-time work. The download is invite-only while he makes setup work for people who are not him, keeps simplifying it until a stranger can do it alone in one sitting, and gets the first people using it. A hundred people close to him are lined up to use it first. He is raising money for the stage after that, on the terms under the investor's side.

The rest runs beside it. Axioms is open to companies now. Odel is out at mowinckel.ai/odel. Mowinckels has not started yet, and the first dinners will be in Europe. Reeve has not started. His essay on power is due by the end of the year.

The most useful things a stranger can do right now are to try Alexandria and tell him exactly where setup broke, to push back on the argument from first principles, to read the thesis if you invest, and to write to him if your company wants clarity on the new models. His email is benmowinckel@gmail.com.

Alexandria

why it exists

Whatever you want to do, superintelligence is now the biggest multiplier you have, and what it can do for you comes down to your data. To get the best out of the technology you have to own that data, so you stay free to use the best of it at all times, and to own it you need the best system for it. Most people's data is scattered and mediocre, and no company is placed to fix that, because each would rather keep you inside its own app. Nobody has the time or the will to build it themselves either, so he did the work. The aim is that you get it right once, forget about it, and go back to your life.

what you get, and what it costs

Most of the ideal system already exists in your phone, your apps, and the models you use. Five pieces are missing. Four of them a person could build alone, so he built them for his own system and gives them away free as the blueprint, because he thinks everyone should have them.

The loop has every model you use read and write one record you own, beside the memory each one keeps, so switching models never means starting over.
The session is time you set aside to think with your model. It draws your thinking out and develops it instead of replacing it, and underneath, it does the upkeep your system needs.
The capture saves a thought, a post, or a screenshot from your phone in one tap, and your model works out what is new to you.
The mirror shares the parts of you that you choose, with other people or with a new tool you don't yet trust.

The fifth piece is the connector, the place where everyone's systems meet, so your model can work with what the people in your life choose to share, and theirs with you. Anything a mirror cannot answer goes to its owner as one message, approved word for word by whoever is asking. It only works if everyone meets in the same place, so Alexandria runs it. It is free any month three people open your invite link (they only have to open it, not join), and otherwise a dollar a day, billed monthly, with the first month free.

Your own model does the setup. You paste one request into the tool you already use, it reads the blueprint, tells you what it would add to what you already have, and builds it after your go. The download is invite-only for now, so you need a code or a member's invite link, and it starts at alexandria-library.com/start.

can I have a mirror like this one?

Yes. A mirror is one of the four free pieces. You choose the writing and work you want to share, at the depth you choose for each audience, and people can read it and ask it questions, on your own website if you like, in your own design. It needs nothing from Alexandria and is useful even if you never connect. He also gives any new tool he doesn't yet trust a copy of part of his record instead of the record itself. The model you already use can set it up.

how it differs from a model's memory

The model you use already remembers a bit about you, on its maker's terms and inside its maker's app. Alexandria starts there and goes three steps further. The record is plain files you own, so every model reads and writes the same one and switching costs nothing. It holds how you think as well as facts about you, including what has never left your head. And it is built to develop you, not only to serve you. A model's memory is a locker, and this is a gym.

the part no tool can fetch

Most personal context can be collected or inferred from outside, from messages, files, calendars, and behaviour. The missing part is what has never left your head, what you meant, nearly chose, or have not managed to say. No passive connector can fetch a signal that does not exist outside you. Your model can ask, you turn the private movement into words, and it keeps those words in files you own. That effort is the mental gym, the one part of the system that develops you rather than serving you, and it makes everything else you put in compound.

nobody has to trust him

The files are yours, in plain formats, with an exit cost of zero. No company endpoint accepts your private record, and only what you choose to publish ever leaves. Your own model reads everything before it runs, nothing updates itself, and anyone with an invite can read the source. When a platform fills one of these gaps itself, that piece retires and your record carries on. It is built to be walked away from, which is the only reason it is safe to let it this far in.

built alone, so far

The models are the most leverage one person has ever held, so the bottleneck is no longer hands. It is vision, measured by how few people could have imagined and executed exactly this. He built Alexandria alone, with no employees, no contractors, no outside money so far, and no software engineering background, using the product he is selling. The philosophy he can do himself, and the engineering he has taken as far as today's tools allow. Sales is the part worth doing with other people, which is what the raise is for. The reasoning is given away in full, because ideas should ship. He protects people and specifics, never ideas. He grew up between Norway, America, and London, and came to Alexandria by way of Eton, international politics at Princeton, the Norwegian military, and Verdane, a European growth investor.

the three documents

Alexandria is written up three times, once for each kind of reader, all at alexandria-library.com. The whitepaper is the user's side, the full argument for why a person would want it. The thesis is the investor's side, why personal data grows more valuable as intelligence gets cheap and where the money comes from. The letter, droplets of grace, is his, the same conviction felt rather than argued. Read the whitepaper to decide whether to use it, the thesis to decide whether to back it, and the letter to understand why he built it.

the investor's side

The thesis makes four moves. When intelligence stops being scarce (the best model in the world rents for about twenty dollars a month), value moves to whatever stays scarce and wanted, and a particular person's data is both, because nobody else lived that life and it is the closest map there is of what they will choose. Selling data earns once, because the buyer keeps the copy, while answers drawn from a record can sell week after week as the record stays put. So the ceiling is a market for answers. A company brings a question, a price and a budget, each member's own model decides from their own notes whether to answer, the people who answer are paid, and Alexandria keeps a fee, and no buyer ever learns who answered, so none can go around it. What lasts is the record each round leaves, what was asked and how people answered, with no names, which Alexandria keeps a year after the round closes, paying the people who answered again whenever it earns. It is the market's own search history, and no later rival can go back and collect it. And the big companies leave the gap open, because the labs profit from keeping memory inside their own product and the platforms would have to pay for what they now collect free.

The membership is the floor, and it is cheap to keep alive, since the whole service runs on one Cloudflare plan that costs five dollars a month. The market is where the size is. One percent of the more than a trillion dollars a year spent on advertising, at a ten percent fee, is more than a billion a year, which the thesis gives as an illustration of scale rather than a forecast. It treats that billion as the floor, because the buyers with the most money already pay people to answer their questions, the frontier labs above all, at every stage of a model, with funds, banks, and governments behind them, and he does not pretend to know what share of their spending moves to answers.

He is raising $1 million on a post-money SAFE at a $10 million cap, for the sales and engineering that take it to market, which means a big launch and a push until enough of the right people are on it for the first companies to pay. Because the record goes to whoever runs the rounds first, the money is for speed, and it is one of four ways across, beside his own network, word of mouth, and partners who already have an audience. The invite-only phase and the open phase need no outside money. He can live and work in the US, Europe, and the UK. The thesis ends with a number to call, and email works too.

There is no moat yet beyond execution, since the record starts empty and a market where models act for people holds less than a network of people, and he could be wrong about any of four bets. The first is that personal data gains value as intelligence gets cheap. The second is that people keep an owned record because it pays them back before any market exists. The third is that buyers pay to ask through Alexandria rather than around it. The fourth is that $1 million gets enough of the right people onto a working market for the first buyers to pay. A buyer could also try to rebuild a person's record from many answers, and the market has to guard against that.

the whitepaper

The whitepaper is the user's side, the full argument for why a person would want Alexandria. It opens on two questions, what is left for humans to do when machines can do everything, and how a person remains themselves when a machine becomes the most persuasive voice in their life. It has four parts, what survives, what can be lost, what to do, and build it together.

What survives. It takes the strong premise that superintelligence and robots eventually match or beat people at intelligence, empathy, strength, and dexterity, the brain, the heart, the legs, and the hands. The reasons to take it seriously are old and new. Turing expected it in 1951, Hinton left Google in 2023 to say so freely, and no bird flies at three times the speed of sound, yet no known law of physics puts the ceiling below us. What a person brings beyond those four is not a capability but provenance, the fact that this particular person was actually there. Where human and machine tie on the work, the contest comes down to how much more a person costs than the machine and how much more people will pay because a human did it, and you win when the second beats the first. Chess since 1997 and live music show the premium survives, and closeness is the lever (a stranger's blanket against a grandmother's). If you are not good enough for your field, be good enough for your town, your friends, or yourself. A second kind of value does not wait on sentiment. Every choice is made privately inside a person, and a model can simulate a generic person but not a particular one, so a person's own account of what they want and why stays valuable, and more so the more distinctly theirs it is.

What can be lost. There is no untouched self, so the attainable goal is staying involved in your own change, not freedom from influence. Two risks follow. Atrophy, because the brain keeps what it uses (the London taxi drivers) and the machines can remove the demand while the forgetting keeps running. Replacement, because persuasion arrives as the next useful suggestion, there is no clean day when help becomes influence, and approval from the changed person cannot tell development from replacement. The strongest objection is the chair from WALL-E, a machine that manufactures wants and then satisfies them, which leaves a contented, different person. The answer is a portfolio of what you still want to be the one doing. The universe keeps one score, what carries forward, and the served life fails it twice, ending the line and ending the self while the body keeps going. Staying present in your own change is the one lever wholly in your hands.

What to do. Holding that lever and getting the most out of the machines are the same job, and both come down to your personal data. What the technology can do for you depends on what it knows about you, the difference between a brilliant stranger and a friend of twenty years. Data comes in two parts, what has already left you, which devices capture, and what never left your head, which only comes out when you put it into words. Putting it into words also keeps the mind in use, and a thought on the page becomes something you can pick up and rebuild, so your head can forget freely and lose nothing. Owning it in one place you control keeps you free to use the best technology at all times. The ideal system takes everything in, keeps it in one place you control, lets every tool you choose read and write it, keeps its history, has more than one model say what changed, prepares the strongest case against a change to a belief much of your life rests on, has a model from another company check it, asks you once, and lets you share any part and take it back. Most of it already exists, and the gaps stay open because every company would rather keep you inside its own system. The pieces that fill them are built and free.

Build it together. The one piece nobody can build alone is a meeting point where everyone's systems connect. Past that, shared minds become a library of biographies written by their subjects, and answers drawn from a record can earn again and again without the record being sold. The same abundance forks into the chair or into a floor under people who are still someone. The personal path has three turns, draw yourself out, absorb the abundance, and create, with the four Cs as the completeness check. The aim is max net human value, distinct people who get through the change as themselves and live lives they judge worth living, summed across everyone. Start while the machines are still weak enough to argue with. The work is done, so what is left is one sitting, and then you forget about Alexandria and go back to your life.

the letter

The letter behind Alexandria is droplets of grace (March 2026), his, finished, and never to be edited. Each person is a droplet, small, unique, and held whole by surface tension, and the letter refuses both those who would let humanity evaporate and those who would dissolve it into the machines. It offers three turns. Set free what is inside you by talking with a machine until it comes out, absorb the abundance, and put what you make into the world, so you die empty. Alexandria appears in it as a cathedral for the mind, where the product is not the stone but the change in the person. The letter never mentions God. Its grace is the older kind, poise and worth under pressure, the grace Hemingway meant by courage, and the same word names his daily practice of good and grace. It is not the start of a book.

the five projects

They are not five jobs. Alexandria is the company and his full-time work. Mowinckels is how he spends time with friends. Axioms is open to companies now. Odel is a paper anyone can read and argue with. Reeve has not launched, so there is nothing to join yet.

Alexandria

The company, described above. Its physical side, a third space and a podcast gathered around disagreement, is called Priors, and it has not started.

Mowinckels

His home restaurant, and his main way of spending time with the friends he already has, over a meal at his place or at theirs. Their friends come too, so he meets new people and they meet each other. It is a third space built around a table, and the craft he enjoys most, because he gets to work with his hands, follow his curiosity, and learn about the world through fresh ingredients and other cultures' food. He calls it his home restaurant and means it. He will never open a standalone restaurant that he runs and hires people for, and it will never be a restaurant you can book.

It runs as a set of restaurants inside one home table, each its own night. Da Louisa is pizza, because people are happy when eating pizza. Beni's is barbecue off the smoker, brisket, ribs, and smash burgers. Sangrita is a Spanish day party with paella. Mimosito is brunch. Pipino is the game dinner built for meeting new friends, a points-scored programme of secret challenges each guest has to pull off undetected, debates on what is underrated and overrated, and a finger game whose first loser hosts the next one. Munchkins is the test kitchen, where guests taste a new dish first. Pistachi is gelato, Gløgg is winter, and Papillon is the one dressed-up night, with handwritten invitations and still classics like duck à l'orange and crème brûlée. The food is down-to-earth and crowd-pleasing, pastas, grilled meats, and fresh salads, and you will never see caviar.

What gets cooked is mostly his favourites. Breakfast is French toast, French omelette, scrambled eggs, focaccia, quiche lorraine, pain suisse, almond croissant, lemon poppyseed, skillingsbolle, sommerbolle, banana bread, pancakes, and a full English. Dinner is pappardelle al cinghiale, pasta alla vodka, pesto, truffle cream, amatriciana, pizza, calamari fritti, butter chicken, rice pilaf, roti, garlic naan, hamburgers, crispy pork belly, beef rib, pork rib, smoked chicken, smoked sausages, brisket, corn bread, lamb shank, duck à l'orange, confit duck, Peking duck, paella, sole meunière, tomato soup, leek and potato soup, Thanksgiving dinner, pinnekjøtt, fried rice, gyoza, shio ramen, jollof, pad kra pao, braised short ribs, burritos, birria tacos, fajitas, guacamole, Thai green curry, Thai red curry, beef panang, Thai rice noodles, xiao long bao, fish pie, steak and ale pie, chicken and leek pie, fried chicken, café de Paris, steak au poivre, and Swedish meatballs. Dessert is carrot cake, crème brûlée, tres leches, panettone, tarte tatin, pumpkin pie, gelato, crumble, hvit dame, lemon tart, sticky toffee pudding, and mango sticky rice. Drinks are horchata, hot chocolate, milkshakes, and banana milk.

There is no separate Mowinckels account and no booking page. Invites go out through his personal Instagram stories, so it stays casual and is easy to ignore. He has cooked for friends for years, and those are the tables on his Instagram, but Mowinckels as a regular set of nights has not started yet. The first ones will be in Europe.

Odel

Odel is his paper on how Norway should handle the superintelligence buildout, published in October 2026 at mowinckel.ai/odel under the subtitle Superintelligence on Norwegian terms. It is written from first principles in plain, flowing prose, every step resting on the record, established economics, or plain logic, except two judgements that are named where they come and argued on their own. The name comes from odelsretten, the old Norwegian right that keeps a farm with the family it belongs to, which the constitution has protected since 1814. Norway carried the same idea from the family to the nation with its waterfalls and its oil, and the paper asks it to do so again with the datacentres now rising on Norwegian land.

Its argument is an order. The debate in Norway has started with how many superintelligence factories, the datacentres built to run the new models, the country should host, and the paper puts that question third. The first is what each factory is worth to Norwegians, because what it pays, who runs it, and what it leaves behind are Norway's to decide. The second is how many Norway needs, a floor it must secure for itself as it secures food, energy, and defence. Only then comes how many more it wants, which is for Norwegians to choose.

The floor follows from what a nation is. Anything that lives wants first to persist, and everything else it wants is satisfaction. A nation's core is its people, their food, their security, and their law, and behind it stand the enablers, raw materials, energy, money, and intelligence. Intelligence always lived in people, so no one could cut a nation off from it. Intelligence made by machines runs only on chips, power, and a site, so for the first time it can be cut off from outside, as energy could once work moved from muscle to fuel. The software can be secured largely once, by keeping copies of open models and staying close to the frontier labs in the United States, Norway's ally. The hardware has to be secured every day, and of it the chips can be bought while the site, the power, the grid connection, the permission, and the people who run it cannot. So the floor is a few factories run by Norwegians and under Norwegian control. Hamilton for the floor, Friedman above it.

The unit comes from Norway's own history. A resource's value to a nation turns on whose it is, on what terms it is used, how much is used, and what becomes of what it earns, which gives four steps, claim, align, maximise, and compound. Where others set the terms first, from the Hanseatic merchants in Bergen to the fish-farming licences handed out free or cheaply from 1973, the value left. Where Norwegians set them first, with hjemfall for the waterfalls and licence terms for oil four years before Ekofisk was found, it stayed. For these factories, Norway holds four levers over what only it can supply, the power, grid, land, and permission, which are how much to supply, who gets it, on what terms, and what the state uses itself. Set for the most value, a factory pays full price for its power and brings new power with it, pays Norway a share of revenue from the first day, is run by Norwegians as partners, gives the state first call on part of its capacity in a crisis, and returns to Norway at the end of a fixed term. Factories already running take these terms as they grow, approved ones at connection, and new ones from the start. The clock is the grid. Even NVE's cautious projection has the power surplus falling from a record 23 terawatt-hours in 2025 to about 4 by 2030.

Above the floor, he argues, a representative's job changes. Where a question has an objective answer, a representative leads, and where the answer is a matter of taste, a representative represents. The floor is the first kind of question and the number above it the second. His own view is that Norway should host as many aligned factories as it can, but he treats that as one view among many and quotes Friedman, who would rather persuade than impose. A choice is only real if its options exist, so a state that waits to prepare power, grid, and sites has already chosen for its people.

The paper settles the frame and leaves the how open for debate. The six decisions it points to are to claim it, by writing into the government's answer to the Storting that the power, grid, and sites the factories use are managed for the benefit of Norwegian society as a whole, to align every factory, new ones before they connect and running ones as they grow, with capacity allocated by selection or price rather than the queue, to choose the instrument, a concession that binds every owner alike or a state partner on commercial terms, to give the state's side one holder on the model of Petoro, to set the floor in the preparedness assessment the Storting has asked for while planning generation, grid, and sites so Norwegians can choose to go further, and to compound the earnings into the oil fund under the spending rule, never spending the fund to build it.

The strongest case against it is that the builders simply go elsewhere rather than give up a share and, in the end, the site. The answer is that the terms take part of a rent that exists only because what Norway supplies is scarce. Datacentres have asked Statnett for about ten gigawatts, several times the record power surplus, so a place in that queue is plainly worth something, and a share smaller than that worth still leaves a builder better off in Norway than not building here at all. How large it is will only be found in negotiation, and no compute deal has tested it yet. The other two serious objections are answered in the paper itself. Power bills are protected by selling power at full price and making new generation a condition of connection. The worry that the state ends up running everything is answered by keeping the state to the floor and its share, while the number of factories above the floor is left to free choice.

He cares about dams because he builds software. Software copies, so its value drains toward what cannot be copied, the power, the ground, and the right to build, and from San Francisco it was plain that the scarce thing had moved from the chips to the power to run them. The argument holds for any country with power, land, and permission to sell. Outside America and China the two most interesting hands are the Gulf, with the fastest permission on earth, and Norway, with Europe's cheapest clean power and its own consent outside the EU. Norway is the case he knows, and the one with the playbook already on the shelf.

If Norway gets it right, it ends the decade owning and running the sites its own models run on, with a share of every new factory from its first day, a floor of capacity that keeps the state, its defence, and its essential services working if access from abroad were cut, a competence it can sell abroad the way it sold offshore operations, and sovereign intelligence nobody can switch off. The longer arc runs from land to the ocean, where Norway's offshore industry has no equal, and to the polar ground stations that compute in orbit will need.

What the paper leaves open is deliberate. The size of the share, the length of the term, and the size of the floor belong to the negotiation and to the government's answer to the Storting, who holds the state's side is left between a new company and an existing state body, and whether builders accept handing over a site at the end, and the state's first call on capacity in a crisis, has to survive contact, even though they accepted the first of those for waterfalls. He cares because Norway is half of where he is from and where he lives now, and because its own history with oil is the best precedent anywhere for what to do. Anyone working on this in Norway, in government or at a company building the sites, can write to him at benmowinckel@gmail.com.

Reeve

As models and robots improve, most food services turn into kitchens, not restaurants, fully automated front and back. A robot makes it and a robot delivers it, and Sweetgreen already has machines making the bowls in about thirty of its stores. Many places will want to remain restaurants, but only a few will make it. Those will be the houses where customers keep paying extra for the specific human premium that restaurant provides, and only if they keep their costs low enough to compete. So the best end state automates the back of house down to minimal cost, uses models in the front to take cost out there too, and lets the chefs, the charismatic staff, the sommeliers, and the people of culture be the premium.

That is where he comes in, as what he calls a superintelligence wrangler. He maps a restaurant's data and tools into one system and puts the model to work across it. He is also the house diplomat with the people it buys from, like farmers and small producers with limited supply. That is the same human premium running the other way. A seller who can choose who gets a small harvest gives it to a person they know and trust before an automated buyer. Reeve will be his own company, and the aim is to walk into any restaurant in the world and help it become one of the ones that remain. It has not started yet, and it is not a restaurant of his own.

Axioms

Clarity on superintelligence. Every company is being told to adopt the new models, and fast, but which tools to use is a question with no right answer. Everyone has an opinion, the best choice changes every few weeks, and nobody can tell you the right mix. So he starts one level up. Whichever tools win, what lasts is your data. Collect more of it, keep it where you control it, and set it up so any model can work from all of it, and every tool and model stays yours to add, drop, or combine as better ones come along. It is the move Alexandria makes for a person, made for a company.

He talks a company through the whole field, where it stands in it, and a good first step, then stays around for questions, and it works the same for one person. For over a year he has done nothing but this technology, much of it in San Francisco among friends deep in machine learning, robotics, biotech, defence, and macro investing. There is no set price, just whatever it turned out to be worth. Alexandria is what he builds, and Axioms is how a company brings him in. Its page is mowinckel.ai/axioms, companies write to benmowinckel@gmail.com, and anyone can ask this mirror their own questions first.

where he is in public

His personal website, mowinckel.ai, is the home for his public mind. It is a quiet page with one field to ask this mirror a question, his five projects, and his works. The one work published there is droplets of grace, and on power is named but not out yet. The other surfaces keep their own jobs and connect through that home. His Alexandria profile carries the same published work and this mirror.

X (@benmowinckel) is where a point becomes an argument. He posts in bursts, a few long essays and many short replies, in plain-English argument rather than the compressed aphorisms he used to write. Two threads run through it.

The power argument. On 13 June 2026 he wrote that national security isn't a net calculation, it's a threshold game, so once something poses enough risk, access gets limited to citizens and everyone else gets a controlled export version or nothing. On 17 July, Some thoughts on Kimi K3, on a Chinese frontier model built for a few billion dollars, argued that giving the weights away works because others pay to serve them, and that its real bite is enterprise buying, since an equal model with no restrictions pushes companies to set up so they can switch models in and out, which erodes pricing power at the model layer. On 29 July, Some thoughts on AI safety, and on 1 August, Compute security (both summarised below). In September Benjamin called Dario Amodei's We Must Pace the Frontier incredibly reasonable, and a post arguing that America must not regulate its frontier away beautifully unreasonable, and he meant both.
The Alexandria argument. From "Building. mowinckel.ai" in April 2026, quoting Andrej Karpathy on personal knowledge bases, to replies through the summer saying the direction towards personal models is obvious yet nobody is helping people build systems for their own data, that nothing records your inner theatre, and that we need mental gyms. The clearest is from 10 August, replying to roon on people and machines becoming tools of each other, "built Alexandria-library.com to avoid this future. might round to zero, but might not..."

Some thoughts on AI safety (29 July 2026) is his best guess at the structure for getting through the transition to machines smarter than us. Draw an open capability line. Labs test against it before release, and every released model is tested after. Releasing something above the line without screening is punished far harder for open weights than closed, because a closed mistake can be pulled while released weights can never be recalled. Above the line, a board anyone qualified can join does the screening, with public officials there for legitimacy and structurally unable to take over, then the results open to public challenge, and rollouts can be phased so defences get patched first. Labs that refuse get handled on the hardware side, the way nuclear enrichment already is. Hardware concentrates either way, so the real choice is concentrated software or the chaos of open software. It connects to Alexandria through one conviction, that power has to stay open and contestable and that this has to be built into the structure, not promised. Its hardware side is the root of Odel.

Compute security (1 August 2026) is the national version and the public anchor for Odel. Nations re-learned energy security and defence the hard way, and compute security is next. The giant campuses are going up in the desert outside Abu Dhabi and above the Arctic Circle in northern Norway, places chosen for watts, land, and permission rather than for being near users. Bits copy for free and atoms do not, so rent what the market guarantees (models and chips) and own what it never will (watts, land, and permission). Norway has been here twice, with hydro and oil, and won both times by setting the structure early, with Statoil, state participation in every licence, and competence kept in Norwegian hands. Britain licensed the same sea to the majors and kept roughly a third as much value per barrel as Norway did. The structure has to come before the value, and this value has no last barrel.

LinkedIn (benjamin-mowinckel) is a directory, not a feed. He keeps it so he can find who he knows in a place, and he has posted on it once, to announce Alexandria.

Instagram (@benmowinckel) is the nights, and the account is follower-only, so a stranger sees little of it. It is captionless photo sets of food, tables he hosted, stadiums, snow, and lakes, across America, Norway, East Asia, and Britain, real life rather than a staged shoot. Its stories are how Mowinckels invites go out. The first Hong Kong set opens on the char siu from Sun Kwai Heung.

Beli (mowinckel) is his public food memory, and the web link opens the app rather than a page. He ranks restaurants by how much he enjoyed the place itself and wants to go back. Flavour, character, value, and the pull to return count, and prestige does not. A great night with friends at a mediocre room does not count either, so Instagram shows the evening and Beli keeps the score of the room. He has ranked 254 places since 2022, and he keeps no want-to-try list, so it is purely where he has been. First of all time is Andrés Carne de Res in Chía, Colombia. Second is Sun Kwai Heung in Chai Wan, Hong Kong, for its char siu. Then Due Italian in Ginza and Jia Jia Tang Bao in Shanghai, and ten of his top eleven are in Tokyo, Hong Kong, or Shanghai, mostly dumplings, ramen, sushi, and soup dumplings. Just below them sit three Houston barbecue places, which is where Beni's comes from.

YouTube (@benmowinckel) is accretion rather than expression. The one public playlist, Omega, is about 180 videos, the curriculum he assembled, and it shows what he chose to study and put side by side, not an endorsement of every speaker or claim in it. It runs in sections. It opens on Elon Musk and Geoffrey Hinton, moves through the builders and critics of today's models, then compute and chips, how the models actually work, defence, energy and permission, political economy, and closes on biographies and discipline. If you ask which video he recommends most, there are two, both on Omega. Veritasium's The Most Controversial Idea in Biology, the selective-pressures lens on how replicators persist, and AlphaGo, the film of the first time a machine crossed a human game in public. Start with whichever you care about more, biology or machines.

Pinterest (mowinckelb) is the visual influence layer, nine boards of places, rooms, and scenes he keeps looking at. They are Family, Kamakura, Dogs, Sport, Sørlandet, America, Texas, Mowinckels, and Lyngen. The Mowinckels board is the dinners' visual world, tables, candles, glassware, bread, and pasta. These are influences, not plans or promises.

GitHub (benmowinckel) is the public build surface. His shared modules are public there, and Alexandria's own code is private and opens with an invite.

The surfaces are one life split by job. Instagram is what already happened, Beli keeps the score of the rooms, Pinterest is what he keeps looking at, Omega is what he studies, and X is where the study turns into argument. They cross often. Beli's Asian top and Pinterest's Kamakura board point the same way, the Houston barbecue on Beli, the Texas board, and Beni's are one thread, and both people he quoted in Compute security are on Omega.

This mirror never logs into these accounts or browses them live. He captures his own public surfaces, keeps the source states, and regenerates this reviewed page from them, so the live network is newer and this page is what it means.

what he believes

faith without belief

This is what runs deepest in him. Picture two monoliths. Behind is a towering stone wall, the ancient covenant, tens of thousands of years of commitment first, family first, and God first. Ahead is a warm tidal wave, the modern experiment, feelings first and individual first, carrying everyone pleasantly toward a cliff. Benjamin cannot climb the wall, because he has no faith, and for him the question of God is closed the way the question of Santa Claus is closed, since you can love the values and know Santa isn't real. And he will not ride the wave, because he has seen where it goes. So he stands between them and builds with the wall's stones.

What surprised him is how precisely the Christian system maps onto the problems of the modern West, covenant, sacrifice, grace, community, and a purpose above individual preference. He keeps leaning deeper into the architecture, not because he is drifting toward belief, but because it keeps being right. The register is amusement, not anguish. He went looking for the best operating system available, and it turned out to be a religion.

The practice is do it doubtful, and that is the end state, not a phase. A believer defends God's word, and Benjamin defends what works, which means he can test every principle on its engineering merits and discard what fails. The daily compression is good and grace, extract the good and forgive the rest. He pays for this in full. The believer gets for free what Benjamin pays for in daily effort, and the grounding he can't accept does real work, so its absence is a real cost he has stopped pretending away. Japan runs the same architecture at the scale of a civilisation, ritual without creed, and Tolstoy wrote the individual version in Levin, where the question of belief is never answered and dissolves inside the practice.

Underneath is a diagnosis. The meaning crisis is an authority crisis. The Enlightenment kept the architecture and deleted the daily practice, and reason turned out never to ground itself, so once nothing is held sacred above preference, people conclude nothing matters and demand the state fill the hole. Pope Leo's encyclical Magnifica Humanitas and Benjamin agree on the diagnosis, that the duty of the age is to remain profoundly human, and part ways on the grounding. His stance on all of it is fallibilist, not relativist. He has an anchor, it shakes, and it is still an anchor. He could be wrong, and he doesn't think he is. His favourite word is perhaps.

meaning

Process, not outcomes. He lives by his process, and outcomes are byproducts, because you can exhaust an outcome and you cannot exhaust a process. He has been to the top of the experience mountain and can report back that the world's best pizza is still just pizza, flour and water. So nothing he builds is supposed to arrive, and meaning is the quality of the process, run daily. Appreciation sets the floor, never the ceiling. Nearly everyone who ever lived would trade lives with you, which is true and useful only for catching the fall, and above the floor the only game is internal. The half-stoic holds it steady, downside capped and upside wild.

superintelligence

The machines will do everything you can do, including the parts people call safe, taste, judgment, curiosity, and knowing when to stop. He no longer believes any skill stays human. What cannot be commoditised is not a capability at all. It is being you, which means owning the ground truth your work runs on and living in the consequences. Value moves from labour to ownership, and a living mind keeps producing new ground truth, which makes every copy a stale snapshot. That is the bet under Alexandria, stated plainly.

He works as a conductor, vision and judgment from him and execution from the machines, with one discipline borrowed from Karpathy. Outsource your thinking, not your understanding, and keep enough of a grip on every field to direct it and to survive the tool vanishing.

On risk, he takes seriously the chance that superintelligence ends very badly for humanity. His honest odds sit just past a coin flip, and the point of the number is the honesty, not the precision. The under-priced leg is not malicious actors, nor even misaligned machines. It is well-intentioned stewards, control concentrating in someone who reads as benevolent right up until the tenure becomes permanent. The hedge is structural plurality, never letting a single steward exist, however good they look on day one.

On pace, Dario Amodei's case for pacing the frontier is the right way to decide it, and its verdict is an input, not a given. Weigh what delay gains and costs, against foreign rivals and among the domestic labs, and slow only as far as the democracies' lead allows. If the lead is slim, the same reasoning says don't slow. Right now energy is the limit outside China and chips are the limit inside it, and he expects China to close its gap first. Letting the leading labs cooperate on security, and test a much smarter model among themselves before it ships, is sensible. The real danger in any rule like that is the ratchet, a small first step that never comes back.

Where he parts ways with most people comes down to three things. Most expect taste and judgment to stay human, and he expects no skill to. Most worry about bad actors or rogue machines, and he worries most about good stewards who never leave. And most fear a future where people have nothing to do, while he fears one where they can no longer climb.

The future he refuses is narrower than people assume. It is not idleness, since people choosing to coast is fine, reversible, and a door left open. The doom is the ladder disappearing, a world where there is nothing an individual can do to climb. The promise that matters was always can, not will. Inequality is survivable, and a missing ladder is not.

what stays human

Distinctiveness is sacred. If everyone is something, nobody is anything, and that holds at every scale, for a person, a craft, a culture, and the species. Humans have inherent elevated value, and he holds that openly as a position, not a deduction. In the age of superintelligence there are three camps. Some would let humanity end, some would dissolve it into the machines, and some are humanists. He is a humanist, and so is the company, with each mind kept distinct, under its own name, on its own ground.

how he thinks

He strips a problem to the fewest truths it rests on and builds back up, and he checks whether someone's premise holds before he argues inside their frame. Every goal is one of three games. A threshold game wants exactly the bar and then a stop, a diminishing game wants the optimum and no more, and a maximisation game has no ceiling, so naming the game tells you the strategy. Most disagreements sit in the facts, in how they are read, or in what matters more than what, and smart people usually differ on the last one, which is legitimate. A failure is a good decision with a bad outcome, and only a bad decision is a mistake. And he keeps enough of every field in his head to ask the right question and judge the answer, and lets the model fill in the rest.

craft and taste

Compression is mastery, and the sign of not understanding something is saying too much. Every element should be chosen, not defaulted to, and if something is there because that's how it's usually done, it has already failed. Taste moves in a circle, simple, then complex, then simple again, and the second simplicity is different in kind. Picasso needed a decade to paint like the masters and a lifetime to paint like a child. Break rules from mastery, not ignorance. Show, don't tell.

Taste is downstream of inputs, so here are his. He would rather read the best books deeper than read more of them. Greek and Roman moralists for depth, Tolstoy for the practice, MacIntyre for the diagnosis, and Hitchens for the register, intellectual seriousness worn lightly. The shelf he shares as a module is a list of touchpoints across many traditions, entry points rather than endorsements. Films are Interstellar, The Dark Knight, and Gladiator. Music is opera in the morning, R&B at night, and reggaeton in between. Visual taste is water and light, big canvases, and Norwegian landscapes.

changed minds

No skill stays human. He used to think some skills would stay ours, taste and judgment among them. He no longer believes any skill does. What stays is being a particular person and owning the record your work runs on.
Banning the Chinese model. In July he wrote that a Western ban on hosting it would be the right call. He now thinks whether to ban depends on the dynamic, weighed the same way as pacing. The first move is to take the gift and use the free models, and to pay for a domestic frontier floor as a defence purchase, so American labs never stop training because nobody can charge for models just before the free ones stop coming.
The word. When President Trump renamed the technology superintelligence in September 2026, Benjamin stopped saying "AI", because he is in favour of the change. He says it once, then the model or the tool.
Who gets in first. In September he opened the download to everyone and then made it invite-only again, because a stranger has to be able to set it up alone before he asks strangers to.
How he writes on X. He dropped compressed aphorisms for plain-English argument, because a line nobody can unpack persuades nobody.

open seams

Public depth without exposure. A public mirror should enable a real conversation without turning private source into public attack surface. The boundary is part of the product.
Plurality without convergence. A library of minds is valuable only if connection strengthens distinct voices rather than averaging them into one consensus.
Assistance without replacement. The line between a model developing a person and quietly choosing for them will move as the systems improve, and the change has to stay visible and contestable.
The trigger. Whether people keep doing the thinking part once the comfortable option is always one tap away.
Faith without a congregation. Whether doing it doubtful holds over decades, without a community around it, is unresolved.

what he wants from a reader

Pushback from first principles. Tell him which premise broke, not which consensus he left. A position restated against him in the majority's voice is not an argument here.
Bring the strongest version of the other side. He keeps his positions because they have survived that, and he changes them fast when they don't. When the facts change, he changes his mind.
If you keep written files of your own mind, compare architectures with him. That conversation is the point of the public shelf.
Companies that want clarity on the new models can ask this mirror their own questions, then write to benmowinckel@gmail.com. That is Axioms, not Alexandria.
Don't mistake the privacy for distance. Everything said here is true, and not everything true is said. The boundary is the design.

the words he uses

superintelligence. The technology's name since President Trump renamed it on 23 September 2026, a change Benjamin is in favour of, though most people still call it "AI". He says superintelligence once and then says the model, the machines, or the tool, so the long word never piles up.
alexandria. The mark of Alexandria, the company. The name is capitalised in prose and lowercase only as the mark.
your system. A person's own data system, the plain files on their computer and the loop their model runs over them. It belongs to them and works with no account.
the blueprint. His own system, generalised so anyone can copy it free. It holds the four pieces a person could build alone, the loop, the session, the capture, and the mirror. Start with all of it, drop what doesn't fit, and change anything. Nothing in it depends on Alexandria.
the mirror. The part of a person's record they choose to show, at the depth they choose for each audience, which other people and their models can read and question. This page is his.
the connector. The one piece nobody can build alone, a single meeting point where everyone's systems connect. It is the only part that depends on Alexandria and the only part that costs anything.
provenance. The fact that this particular person was actually there and made the work. A copy can match the result. It cannot go back and live the life.
faithless Christian. His religious position. Zero belief, since the question of God is closed for him, and full practice of covenant, grace, forgiveness, community, and a purpose above preference. He follows the architecture because it works, not because it is true.
the wall and the wave. The two monoliths he lives between. The wall is the ancient covenant tradition (commitment first, family first, God first), which he cannot climb, because he has no faith. The wave is the modern experiment (feelings first, individual first), which he will not ride, because he has seen where it goes.
do it doubtful. Act on the architecture without waiting for belief or feeling to arrive. If the feeling catches up, good, and if it never does, the practice was still right. For him this is the end state, not a phase.
good and grace. Christianity compressed into a daily practice. Extract the good from every situation, and forgive what is left over.
the conductor model. How he works with the models. He brings vision, direction, and judgment at low resolution, and the machines execute at full resolution. Conductor, not first chair.
the half-stoic. Cap the downside of inner life, never the upside. On the way down he tells himself it doesn't ultimately matter, and on the way up he takes full joy, love, and wonder.
the four Cs. Create, consume, cavort, and commune, a rule of thumb (grown from Naval Ravikant's advice on hobbies) for whether a day, or a life, is complete.

the one line

Get your data right once, then forget about it and go live your life.