I. The bliss point

There is a point at which a soft drink stops being pleasant and becomes irresistible. In the 1970s a psychophysicist named Howard Moskowitz worked out how to find it.

Asked to set the sweetness for a new Dr Pepper, he mixed dozens of versions and tested them on 3,000 people.

Then he modelled the results. A little less sugar and the drink was underwhelming. A little more and it turned cloying. The peak was narrow, it was findable, and once found it could be manufactured at scale forever. The industry called it the bliss point.

Moskowitz wasn’t doing anything sinister. He answered the question he was asked, which was how to make something people would enjoy as much as possible.

The trouble came next, and Michael Moss documents it precisely in Salt Sugar Fat. The bliss point worked on things we already expected to be sweet. So the industry marched around the rest of the grocery store adding sugar to everything else. Bread. Pasta sauce. Yoghurt. Salad dressing. Sweetness sells, and nothing else about the product had to change.

We began to expect sweetness in everything. Real food started tasting like something was missing.

II. Four pounds to a hundred and fifty

In 1700, average per capita sugar consumption in the United States and England ran at about 4 pounds a year. By 1800 it was around 18. By 1900, somewhere between 60 and 90. By 2000, it had passed 150.

Sugar went from a luxury locked in cabinets and displayed to guests, to an input cheap enough to use as filler in savoury products.

The last stretch of that curve was a policy decision. In 1971 Earl Butz became Nixon’s Secretary of Agriculture and told American farmers to plant corn fencerow to fencerow. Get big or get out. Washington rewrote the subsidy rules to pay for volume, corn production surged, and the surplus needed somewhere to go. High-fructose corn syrup, first marketed in the early 1970s, was where it went. Coca-Cola switched in 1980 and the rest of the industry followed within five years.

What changed was the price of the input, and everything downstream rearranged itself around the new abundance.

III. The experiment that removed willpower from the argument

For forty years the debate about industrial food ran aground on the same objection. People choose what they eat, so obesity is a question of individual restraint.

In 2019 Kevin Hall at the NIH made that position much harder to hold.

Twenty adults moved into a research ward. Each spent two weeks eating ultra-processed food and two weeks eating unprocessed. The diets were matched for calories, sugar, fat, fibre and sodium. They could eat as much or as little as they wanted, and rated both diets as equally pleasant.

On the ultra-processed diet, people ate 508 more calories a day and gained a kilogram. On the unprocessed diet, they lost weight.

The mechanism was rate. Participants ate the ultra-processed food at roughly 48 calories a minute against 31 on the unprocessed diet. The food was softer and denser, so it went down faster than fullness could catch up.

The satiety signal was intact throughout. It was simply too slow.

IV. The patch

Ultra-processed food now supplies between 53 and 58 per cent of American calories, and about 62 per cent of what American children eat. American adult obesity went from 13 per cent in the early 1960s to 41.9 per cent by 2017. Severe obesity went from under 1 per cent to nearly 10 per cent.

The curve has since flattened, and the timing points at GLP-1 drugs. About one in eight American adults now reports taking one, up from one in sixteen a year and a half ago.

Two details here are almost too apt to be true, and both are documented.

Rapid GLP-1 weight loss strips lean tissue along with fat, by some estimates a fifth to two fifths of everything lost. How much of that is real muscle, and how much is water, is still argued over. The cure for an engineered appetite leaves you smaller and somewhat weaker.

And the companies that spent forty years making us eat more are now selling to the people medicating to eat less. Nestlé launched a frozen meal line for GLP-1 users in 2024. Danone markets a GLP-1-friendly yoghurt. Conagra put a badge on 26 existing Healthy Choice meals to flag them for people on the drugs, changing nothing about the food.

V. The bliss point of the mind

Human beings have a well-documented cognitive bias called processing fluency. When information is easy to process, we judge it to be more true, more familiar and more valuable. Alter and Oppenheimer reviewed dozens of studies and found the same thing each time. Take one claim, write it clearly, then write it awkwardly. Readers judge the clear version more likely to be true.

Fluency is to cognition what the bliss point is to appetite. Both are old shortcuts that worked when the signal was hard to fake. Both break the moment somebody can manufacture the signal on demand.

Large language models are the most effective fluency engines ever constructed. These models are trained on human preferences. This means they are trained toward prose that reads well: smooth, confident and well-structured. The bliss point of prose, found across a sample much much larger than Moskowitz’s 3,000 tasters.

So the question is whether we are about to do to reading and writing what we did to eating.

VI. What the evidence shows

Pew Research sampled English-language web pages at random in July 2026. Roughly 10 per cent showed significant signs of AI authorship. Among pages published after ChatGPT launched, it was about 35 per cent. One SEO firm, using a looser method that inflates the number, puts it at just under half of everything newly published in English.

The deskilling evidence is thinner, though noteworthy.

At four Polish centres, 19 experienced endoscopists performed 1,443 colonoscopies with no AI help. Some did theirs before spending a period working with AI detection tools. Some did theirs after. The detection rate for those who had worked with the tools fell from 28.4 per cent to 22.4 per cent. These were experienced doctors, performing a procedure they had done thousands of times. A spell of working alongside a machine left them worse at doing it alone. A commentator at UCL called it the first real-world clinical evidence of deskilling.

The education evidence runs the same way. In an experiment with 1,000 Turkish high school students, access to GPT-4 improved practice performance. When the AI was taken away, students who had used the unrestricted interface scored 17 per cent worse than peers who had never had access at all. They had learned to produce answers without learning to solve problems.

Then the twist. A second group got a version that gave hints instead of answers. They improved more during practice, and lost nothing when it was taken away. Same technology, same students, opposite outcome.

There is also a widely circulated MIT study using EEG. It found weaker neural connectivity, and a weaker sense of ownership, among people who wrote essays with an LLM. It is small and unreviewed, so treat it as suggestive.

VII. The case against everything I have just written

The strongest argument against this thesis is that it has been made before, repeatedly, and has been wrong nearly every time.

In the Phaedrus, Socrates objects to writing itself on the grounds that it will destroy memory and produce the appearance of wisdom. He had a point, in that literate cultures do memorise less, but the trade proved so favourable that the objection now reads as a curiosity. The printing press was going to drown Europe in trivia. The novel was going to ruin young women. Television was going to end sustained attention. Calculators were going to end arithmetic. Wikipedia was going to end research.

People adapted each time by repurposing the freed capacity. The base rate for this genre of argument is poor, and anyone making it, including me, should carry that as a live discount.

The benefits are also real and well distributed. In a study of 453 professionals doing mid-level writing tasks, ChatGPT access cut time by 40 per cent and raised quality ratings by 18 per cent. Among 5,172 customer support agents, productivity rose 14 per cent and 34 per cent among the least experienced. Across this literature the least skilled gain the most, which compresses the gap between the best and worst performers.

Some people have been shut out of conversations their whole lives because of how their writing looked. Non-native speakers. People with dyslexia. People whose prose gave away the wrong class background. For them, this removes a barrier that was never legitimate. Any version of this argument that dismisses all AI writing as junk has to explain away those people, and it cannot.

Measured literacy has also been falling across the OECD for years, and the declines predate ChatGPT by a wide margin. Whatever is happening to reading and writing did not start in November 2022.

VIII. The narrower claim

“AI will make us stupid” is unsupported by the evidence and unsupported by history.

Here is the version I can defend. Unstructured, default AI use deskills specific capacities. We have direct evidence in endoscopists and students. We have longer precedent elsewhere. The FAA has spent two decades worrying about pilots who mostly monitor automation and lose the feel for flying by hand. Drivers who follow turn-by-turn directions engage less with the space around them.

Deskilling does not need a harmful technology. It needs only that the practice which built the skill quietly stops, and that nobody notices, because the assisted work still looks fine.

Lisanne Bainbridge named this in 1983: the irony of automation. Hand the routine parts of a job to a machine and you remove the very practice that produced the judgment needed for the exceptional parts.

Writing has a specific version of the problem. Its value was never in producing documents. The effort of composing a sentence forces the resolution of a thought that stayed vague until you tried to write it down. Cognitive science calls this family of effects desirable difficulty. The principle is that the struggle is where the learning happens. Outsourcing the struggle removes the process that was generating the capability.

Which returns us to Hall’s 508 calories, and to a more precise parallel than the one about getting fat. The ultra-processed diet was overconsumed because it could be eaten faster than the feedback loop could operate. Fluent prose can be produced and consumed faster than the loop that tells you whether you have understood anything.

IX. The rest day

Muscle is not built during the workout. The workout is the damage. Muscle is built afterwards, while you rest. That is why training programmes are built around rest days, and why athletes who skip them get weaker.

Cognition works the same way, and the evidence is better than the wellness industry’s version of it would suggest.

Stop directing your attention outward and a set of brain regions called the default mode network comes alive. It runs during daydreaming, mind-wandering, remembering, imagining what comes next.

For years its link to creative thinking was correlational, which proves little. Then researchers found a way to test it directly. Some epilepsy patients already have electrodes implanted in their brains. Researchers gave them a creativity task, then used targeted stimulation to knock the network out mid-task. Originality dropped. The number of ideas did not. People kept generating answers. The answers just got more obvious.

The behavioural work agrees. Sio and Ormerod pooled more than a hundred experiments on incubation. That is the effect where setting a problem aside improves your odds of solving it. It holds up, and it holds up most strongly for the open-ended thinking that original work depends on.

Fill the break with something demanding and the benefit shrinks. A phone full of unread things is a demanding task wearing the costume of a rest. Doing nothing at all was not optimal either. Low-demand activity usually beat it. Walking, showering, washing up, driving a familiar route.

The finding that matters most is this. Longer preparation before the break produced larger incubation effects. The rest only pays if you did the work first, exactly as a rest day builds muscle only if you trained and is otherwise a day off.

Which is where this meets AI at an angle the usual argument misses. The worry is normally that the model does the work for you. The sharper problem is that it removes the preparation, and the preparation is what makes the recovery productive. If you never load the problem, there is nothing for the default mode network to work on while you walk the dog. No sitting with it, no failing at it, no carrying it around half-solved. You get the answer without ever having held the question. The walk still happens. It has nothing to incubate.

What has changed is that the recovery phase now has to be defended, because the environment is designed to fill it.

X. What a reasonable AI diet looks like

Nobody serious argues you should never eat processed food. The advice that survives contact with the evidence is simple. Cook some of your own meals. Eat things you have to chew. Lift weights. Take rest days. Unglamorous, and it works.

The cognitive equivalents follow from the studies above.

Think first, then consult. In the MIT study, the people who reasoned it through themselves before opening the model stayed the most engaged. The order of operations matters more than the tooling.

Use AI as a critic. Make yourself say what you think before you ask what it thinks. Systems built that way preserve cognition. Generate-and-accept does not.

Prefer hints to answers. The Turkish school experiment is the cleanest finding here. The tutor that gave hints beat the one that gave answers on every measure that mattered.

Keep some things effortful on purpose. Write some things unaided, particularly the things you most need to understand. This is the resistance training, and like resistance training it is unpleasant and non-negotiable.

Defend the recovery. Load a hard problem, then leave it alone somewhere low-demand. One finding here is unusually practical. Ward and colleagues tested 800 people on cognitive capacity, varying only where their phone sat. Face down made no difference. Powered off made no difference. Only another room worked. Not thinking about the phone costs you the same as thinking about it.

Watch for the recalibration. Moss’s insight was about the whole store rather than any single product. Once sweetness is everywhere, unsweetened food starts tasting wrong. Once smooth, confident prose is the ambient default, difficult writing starts reading as bad writing. That effect reaches you as a reader long before it reaches you as a writer.

XI. A disclosure, and a question

I used AI in producing this essay. The research underneath it, the studies and figures and chronology, was assembled in a fraction of the time it would have taken me alone. The argument and the judgments are mine, and I checked the primary sources on the claims that matter. But it would be absurd to write this piece and not say so.

That is the actual condition of the question. AI is useful, which is why what habitual use does to us deserves to be asked carefully rather than dramatically.

So here is where I have landed, some way short of the conclusion the title implies.

Nobody is going to get fat and weak from reading. The metaphor breaks there and I would rather abandon it than defend it past its usefulness. But we have run this experiment once already, in another domain, and we know the shape of the result. An input became cheap. The industry reorganised around the abundance. The products got better at being consumed. Consumption outran everybody’s feedback loop. It took fifty years and a class of expensive drugs before we worked out what had happened.

Which version we get has little to do with the technology. It depends on whether we notice the recalibration while it is happening, and whether we keep doing some of the work by hand.

Sugar never made anyone stupid. It just spent three hundred years quietly changing what normal tasted like, and nobody alive could name the year it happened.

That is the part to watch for. Not the moment the writing gets easier, which has already happened, but the moment you stop noticing that it did.

I. Brighton, Christmas 1894

The problem with electricity, in the beginning, was that nobody wanted very much of it.

This was not a failure of imagination. A central generating station was an enormous, expensive, immovable thing, and it sat idle most of the day. Its customers bought light, and lighting has a single sharp peak: everyone wants it at dusk, and almost nobody wants it at four in the morning. A station sized for the evening peak spent the other twenty hours burning coal to keep its boilers warm and earning nothing. The capital was vast and the utilisation was miserable. Electricity was expensive, so it was rationed. Because it was rationed, demand stayed small. Because demand stayed small, the plant stayed idle.

Most industrialists solved this by opting out. A factory that wanted electric light built its own dynamo in its own basement and hired its own people to run it. These were called isolated plants. In the 1880s and 1890s they were the sensible choice. Popular enough that Edison’s own company had installed 702 of them by 1886 against only 58 central stations. Power was something you owned, staffed, maintained and worried about. It was a core competence.

Samuel Insull was a Londoner who had crossed the Atlantic at 21 to work as Thomas Edison’s private secretary. In 1892 the presidency of General Electric went to someone else. He left New York for Chicago and took over Chicago Edison, a small utility that was losing money.

He did the arithmetic and reached an unusual conclusion. The price of electricity was not a function of the cost of coal. It was a function of load factor, the ratio of average demand to peak demand. Load factor depended on who your customers were and how you charged them.

Two years later, on a Christmas visit to the English seaside town of Brighton, he found the answer by accident. He walked through the town in the evening and saw something that never happened in America: the shops were closed and their lights were still on. He went looking for whoever ran the local supply and found Arthur Wright, the engineer who had built Brighton’s system. A decade earlier, Wright had also built one of the first practical supply meters.

Wright’s meter measured two things: how much electricity a customer used, and how fast they used it. That second number, the peak demand, was the one nobody else was charging for. Wright had built a tariff around it. Customers paid a fixed charge based on the size of the demand they wanted available to them, and then a very low charge for each unit they actually used.

Insull licensed it, took it back to Chicago, and by 1897 was offering the city two-part rates.

The structure did three things at once.

First, customers paid for the capacity they held open, which is what infrastructure really costs. Second, each unit was cheap, so they stopped rationing themselves. Third, Insull had a reason to chase customers whose peaks fell at different hours: streetcars in the morning, factories through the day, homes at dusk. Every customer who filled a valley in the load curve made every other customer cheaper to serve.

Prices fell. Consumption exploded. Insull kept betting on scale. In 1903 he ordered a turbine for his new Fisk Street station that was more than three times larger than anything then running, and GE’s chairman, Charles Coffin, balked at building it. Insull took the risk himself, telling him that if it failed he could “take the apparatus out and throw it in the junk pile.” It worked. It became the industry standard, and the dynamos in the basements went dark one by one.

Utilities never stopped metering electricity. They still meter it today. They simply changed the unit. They stopped selling consumption and started selling the size of the connection, and once they did, consumption got cheap enough that customers stopped counting it.

II. The same shape, a century later

It happened again to bandwidth, fast enough that most people reading this lived through it.

Through the mid-1990s America Online sold internet access by the hour. $9.95 bought five hours a month; every extra hour cost $2.95. Subscribers behaved as people do when a meter is running in the corner of the room. They logged on with a purpose. They rushed through their mail. They logged off. Stories went around about parents opening credit card statements to find hundreds of dollars of chat room charges run up by a teenager.

The web itself was built around the meter. Small images, sparse pages, a text-only version of everything. What we remember as the visual style of the 1990s internet was a cost-optimisation artefact, the aesthetic residue of a pricing model.

On 1 December 1996, under competitive pressure from MSN and AT&T WorldNet, AOL put every subscriber on a flat $19.95 a month for unlimited access.

Demand responded immediately. Average usage went from seven hours a month to twenty-three, and the network could not absorb it. Subscribers spent January 1997 listening to busy signals. Customers sued, states opened investigations, and Steve Case posted a notice asking eight million members to please use the service less while AOL built more capacity. Fourteen months later AOL raised the unlimited rate to $21.95.

Flat rate did not mean free. It did not even mean cheap. It meant the constraint stopped being the customer’s problem and became the operator’s, which is what infrastructure is.

And notice what happened. At the retail layer the meter vanished. At the wholesale layer it did not vanish at all. It moved.

Internet transit is the business of carrying one network’s traffic to the rest of the internet, and nobody has ever sold it by the gigabyte. They sell it by speed. You pay for the width of your pipe, not for what you push through it. Two customers moving the same volume get very different bills if one moves it in a steady trickle and the other in violent bursts.

The price of that pipe fell from $1,200 per megabit per second in 1998 to 63 cents by 2015. Everyone in the business believed, every single year, that the decline had to stop. William Norton, the Equinix co-founder who chronicled the market for over a decade, concluded that the downward pull on prices behaves less like a market and more like a law of physics.

The overbuild made it a law. In the five years after the Telecommunications Act of 1996, telecom companies laid roughly 80 to 90 million miles of fiber across North America and Europe, financed with debt and on faith. Between 1999 and 2004, less than 5 per cent of it carried traffic. Global Crossing was worth $47 billion in 1999. In the fourth quarter of 2001 it lost $3.4 billion on $793 million of revenue, and filed for bankruptcy weeks later. The industry called the unused capacity dark fiber and treated it as a monument to its own stupidity.

It was, in fact, the subsidy that paid for the next decade. Streaming video, cloud computing, the smartphone internet, all of it ran across fiber buried by companies that did not survive to see it lit.

Nobody flat-rates a scarce good. Flat rate is what you do with a glut.

But there is a second half to this story.

The price of bandwidth collapsed. The bill did not. Real prices for the most popular US broadband tiers fell around 60 per cent per megabit between 2015 and 2024, and are still falling about 6 per cent a year. Over the same stretch the median household bill drifted from roughly $70 a month to $80. Providers keep raising baseline speeds without raising prices. The cost per megabit falls while the monthly bill sits exactly where it was.

Consumers did not bank the savings. They took them in product. What collapsed was the unit price. What held was the spend.

This is the part that matters if you are trying to make money here. Falling unit prices did not destroy broadband’s revenue, because broadband stopped selling units. It sold a subscription, then quietly let what you got for it grow until nobody thought about the limit at all. The price per megabit fell by half while the bill stayed $80.

III. The dynamo in the basement

I have done this once already.

In the late nineties I timed my sessions. I knew roughly what a large download would cost me and would decide in advance whether it was worth it. The alternative was a surprise bill at the end of the month. I cannot tell you what year I stopped doing that. There was no moment. At some point the arithmetic simply left my head. There are now a dozen devices in my house connected continuously and I have no idea how much data any of them move. The average American household moves 767 gigabytes a month and almost none of them could tell you that either.

That is what the end of a metered era feels like from the inside. The quiet disappearance of a calculation you used to make every day.

Which brings us to a curious fact about how the most sophisticated companies in the world currently buy and ration intelligence.

They buy it by the token, and they optimise obsessively against that meter. They compress prompts. They cascade queries through routing layers that send easy work to a cheap model and escalate only when necessary. They cache aggressively, build dashboards to watch consumption, and employ people whose job is substantially to reduce the number of tokens the organisation consumes.

This is the isolated plant. It is a rational response to a real cost, and it is a rapidly depreciating skill.

The unit price of intelligence is collapsing faster than any comparable input in industrial history. When OpenAI opened the GPT-3 API in 2020 it charged $60 per million input tokens. The same capability now costs six cents or less. That is a factor of a thousand in under six years.

Andreessen Horowitz named the pattern LLMflation and put it at roughly ten times a year. Epoch AI, working benchmark by benchmark, found a median of fifty times a year, and after January 2024 that median rose to two hundred.

Bandwidth fell at something like 35 to 40 per cent a year for seventeen years and reshaped the world economy. Tokens are falling at 90 per cent a year. The cost curve we are hand-optimising against is running an order of magnitude faster than the one that made the internet.

Shaving tokens in 2026 is shaving kilobytes off GIFs in 2003. The work is real and the savings may be real. The whole problem is about to disappear anyway, for reasons that have nothing to do with how clever anyone was.

IV. The switch is already thrown

Spare compute has collapsed in price. H100 rentals fell from $8 to $10 an hour at the 2023 peak to between $1.80 and $3.50 by mid-2026, as more than 300 new providers flooded in.

Reserved compute however went the other way. One-year contracts bottomed at $1.70 an hour in October 2025, then climbed 40 per cent to $2.35 by March 2026. On-demand is effectively sold out, and the people holding it will not let go at any price.

Two prices for the same silicon, moving in opposite directions. Interruptible compute is nearly worthless. Guaranteed compute is appreciating. Whatever is scarce here, it is not tokens. It is the guarantee.

The products have already been built to sell it. Microsoft has begun selling Provisioned Throughput Units on Azure. A PTU is a block of reserved capacity that guarantees you a fixed number of tokens per minute at a consistent speed. It starts at roughly $2,448 a month. It bills on the capacity you reserve, not the tokens you consume. If you do not use it, you lose it, and you pay for it anyway. That is not a usage meter. That is a leased line, identical to a T1 contract from 1998.

OpenAI has started selling Scale Tier: throughput bought in bundles of tokens per minute, with a 99.9 per cent uptime guarantee and priority on compute. Buy more quota and your rate limits rise. The company’s own phrase for this is “uncapped scale.”

Reserved capacity is not itself new. Clouds have sold committed capacity alongside on-demand since 2009, and any large buyer of anything lumpy does the same. What is new is the second axis. Alongside Scale Tier sits Fast mode, which costs a premium per token and runs up to two and a half times quicker, and Flex, which costs about half as much and is slower, best-effort and sometimes unavailable. You are no longer only choosing how much to reserve. You are choosing how fast it goes.

Fast mode also carries a ramp limit. Push your throughput up more than 50 per cent within fifteen minutes and the system downgrades you. That is burst management, solved with a network operator’s tool.

And it is spreading. OpenRouter, which resells access to all the major labs, now has a single setting that picks fast, standard or cheap-and-slow from any of them.

Industries do not standardise the thing they are about to stop selling. They standardise the thing they are about to sell more of.

Consumer pricing is further behind, but reaching for the same vocabulary. Claude Max is sold as five times and twenty times a baseline allocation. Gemini’s tiers are sold as two times and four times free usage. ChatGPT Pro is marketed with dedicated GPU capacity. These are still allowances rather than pipes, buckets with the sides pushed out. But notice that nobody is selling tokens anymore. Everybody is selling a multiple of something, because a multiple is the only language available for a product the industry has not finished inventing.

Which is where this ends up, and it is worth being specific about when. Broadband’s flat rate arrived in 1996, during the frenzy. Genuinely cheap, genuinely uncapped broadband arrived on the far side of the crash, riding capacity that bankrupt companies had already paid for, roughly five to seven years later. If AI runs the same course from a correction late this decade, capacity pricing becomes the default somewhere between 2029 and 2032.

By then you will not buy tokens. You will buy a pipe of a certain width, running at a certain speed, and you will stop counting what goes through it. The width will be the product. The speed will be the premium. The tokens will be the thing nobody mentions in the marketing.

V. The overbuild

None of this becomes cheap until somebody builds far too much of it, and somebody is building far too much of it.

Combined capital expenditure by the four largest US hyperscalers went from roughly $226 billion in 2024 to about $410 billion in 2025 to guidance of approximately $725 billion for 2026, a 77 per cent increase in a single year.

The bet is not obviously wrong. Microsoft is sitting on roughly $80 billion of Azure orders it cannot fulfil because it cannot get power. Google Cloud’s backlog stands at $460 billion, roughly double where it was a quarter earlier. This is contracted demand, which is what the fiber builders of 1999 did not have.

But the tell that this is a frenzy and not merely a build-out came in the market’s reaction. When Meta raised its 2026 capital expenditure guidance, the stock fell more than 9 per cent in a day. Alphabet, Amazon and Microsoft all sold off on their own disclosures. Investors have begun to flinch, which is what investors do shortly before they stop.

Whether the flinch becomes a correction is not knowable, and for this argument it does not need to be. Both outcomes produce the same pricing regime. If the build-out is vindicated, capacity becomes abundant and the meter moves. If it is a bubble, capacity becomes stranded and cheap and the meter moves faster.

VI. Where the analogy strains

The strongest objection is that the most influential person in the industry says the opposite.

At an infrastructure summit in Washington in March 2026, Sam Altman described a future where intelligence is a utility like electricity or water, bought from OpenAI on a meter. Not a subscription. A meter. The framing has caught on: the token as the kilowatt-hour of a new industrial age.

The market is moving that way too. Google introduced credit-based metering. Anthropic moved agentic usage onto separate metered billing. By mid-2026 there was no honestly unlimited plan left anywhere, at any price.

There is also a real physical difference between the two cases, and I do take it seriously.

A lit strand of fiber costs approximately nothing to run, and its capacity could be multiplied without digging anything up. Wavelength-division multiplexing took a single strand from one channel to sixteen and then far beyond, which is why fiber buried in 1999 kept getting more valuable for twenty years. Compute has no equivalent free lunch (yet). Capacity requires new silicon, new buildings, new substations. Efficiency gains are real but they run to 30 to 50 per cent, not sixteen-fold.

The depreciation schedules are not comparable either. Fiber laid in 1999 was still a strategic asset in 2015. A GPU has three to six good years. Dark fiber became the foundation of the next era. Dark GPUs may simply be stranded.

The binding constraint, meanwhile, is not chips at all. Global data centre electricity consumption reached 415 terawatt-hours in 2024 and is projected to exceed 945 by 2030. Lead times on the high-voltage transformers needed to connect a site have gone from around two years before 2020 to five. Jensen Huang’s formulation at GTC this year is the clarifying one: Revenue = Tokens per Watt × Available Gigawatts.

Now hold that objection up against the first section of this essay, because I think it inverts.

Metered like electricity, yes. But how is electricity actually metered? Large industrial and commercial customers do not simply pay for kilowatt-hours. They pay a demand charge based on the peak capacity they reserve, and a separate, much lower charge for the energy they draw. That is Arthur Wright’s two-part tariff, and it has been the structure of industrial electricity pricing for well over a century, for exactly the reason Insull identified. When the plant is capital-intensive and supply is lumpy and slow to build, you sell reservations first and units second.

Constrained supply with long lead times and a queue does not produce pure consumption pricing. It produces capacity pricing. It is how airlines sell seats, how ports sell berths, how grids sell interconnection, and how the internet has always sold transit. The power constraint is not evidence against the pipe. It is the mechanism that guarantees it.

Altman is right that intelligence will be sold the way electricity is sold. That is an argument for the pipe, not against it.

VII. What the pipe permits

The interesting question is never the pricing model. It is what the pricing model permits.

Flat-rate dial-up did not just make the internet cheaper. It made a whole category of product possible for the first time: things that assumed you were always connected, that ran in the background, that spent your bandwidth freely to save your attention. Nobody builds a streaming service, or a social network, or a phone that syncs all day, for customers counting minutes. The pricing model was not downstream of the products. The products were downstream of the pricing model.

The same thing is about to happen here, and the shape is already visible in the demand data. Reasoning models spend ten to a hundred times more tokens per query than single-shot completions. Agentic workflows, with their retries and tool calls and context reloads, run five to twenty-five times more expensive per task. Under per-token pricing, every one of those behaviours is something a user must justify. Under capacity pricing, they are what the machine does while nobody is watching.

That is the actual prize. A metered model is a model you supervise. An uncapped model is a model you delegate to. Supervision gives you a tool. Delegation gives you something closer to staff. The difference between them is not capability, because the capability is here already. The difference is whether the person using it has permission to stop counting.

The strategic question is not how to reduce token consumption. It is what you would build if intelligence were free at the margin, and what you would have to start building today to be ready for it.

The companies that won the broadband era were not the ones that optimised hardest for a 28.8k modem. They were the ones with something worth doing at ten megabits, built and waiting, before ten megabits arrived.

VIII. What to do while you wait

Five things follow.

Price to the spend, not to the meter. If your revenue is a spread on resold tokens, you are short a variable that falls ten times a year, and no amount of volume growth is a strategy against that. Broadband’s answer was to hold the subscription and inflate what sat inside it, and hold the customer relationship. Decide now whether you own a customer or arbitrage a commodity, because the second business has a visible expiry date.

Stop treating token cost as a durable engineering discipline. The unit price is falling ten times a year without your help. Cost-optimisation infrastructure is a depreciating asset, and the organisational instinct you are building, the instinct to ration, is the hardest thing to unlearn when rationing stops being necessary.

Start treating capacity as the thing you are buying. Watch reserved contract prices, not spot. Reserved is where the scarcity sits, and it is already turning upward. Negotiate like a commodity buyer, not a software buyer: consumption commitments, ramp profiles, capacity reservations, exit terms.

Preserve your ability to leave. Transit went to zero and the margin migrated to everything sitting on top of it. The same will happen here. Keep prompt logic and retrieval data in your own storage and call models as a stateless service. Do not build your business inside a vendor’s agent platform, however convenient it is this quarter.

Build for the wide pipe. Design products that get better when given more compute, rather than products tuned to a budget. When the constraint lifts, applications architected around scarcity will have nothing to do with the abundance, and the ones architected around abundance will look, quite suddenly, as though they were obvious all along.

IX. The Entry Fee

Insull’s empire collapsed in 1932, inside a holding-company structure that had grown more elaborate than the business underneath it. He fled to Europe and fought extradition through Greece and Turkey. Federal agents brought him back under guard. He was tried three times and acquitted three times. He died in July 1938, aged seventy-eight, of a heart attack on the platform of a Paris Métro station, close to penniless. The grid he built outlived him by a century and is, in every meaningful respect, the one we still use.

The people who lay the infrastructure are rarely the people who profit from it. The fiber builders went bankrupt and the streaming companies got rich. This is not a tragedy so much as a mechanism. Frenzies are how societies fund infrastructure they cannot yet justify, and the write-off is the entry fee for everybody who arrives afterward.

$725 billion is being spent this year on capacity for demand that does not yet exist, by companies convinced that the greater risk is being short. Some meaningful fraction of it will be written off. The rest will sit there in buildings full of silicon nobody can currently afford to keep busy.

Somebody will do the arithmetic Insull did in Chicago and AOL did in 1996, and work out that the only way to make an enormous idle asset pay is to stop asking people to ration it. To sell them a connection instead of a quantity, and let the units get cheap enough that nobody bothers to count.

At which point everyone will discover, all at once, that they have been building for the wrong world.

1 How I wrote this

The next morning, it had added helpful context around the label. Extra detail about why the task mattered and when it was due.

Completely reasonable. Except I’d explicitly told it not to do that.

This is the persistent memory problem. If you’re working with AI agents in any meaningful way, you’ve either hit it or you’re about to.

The Smartest New Hire Who Can’t Remember Last Tuesday

Imagine you hired someone brilliant. They read your entire operating handbook on day one and synthesized six management frameworks overnight.

But every morning, they’ve forgotten specific things you told them yesterday. The big picture survives.

That one rule you added at 3pm on a Thursday? Gone.

The difference with a human is that repetition builds long-term memory over time. Tell your Chief of Staff the same thing three times and it sticks.

AI agents work differently. Every conversation has a context window: a finite amount of information the agent can hold at once. When it fills up, older details get compressed or dropped.

The challenge on top of that is that the agent doesn’t know it forgot. It doesn’t say “hey, I lost that instruction.” It proceeds with confidence, filling gaps with reasonable-sounding assumptions.

When It Actually Breaks

Three real examples I’ve experienced so far from building Rumi.

The Checklist That Never Got Run. I built a 12-point quality checklist for written content. Every draft was supposed to pass all 12 checks before I saw it. Rumi produced a full article and hadn’t run the checklist at all. Nine violations in one draft.

The Confidentiality Leak. I gave a clear instruction: reference a task using only the exact phrase I approved. In a summary later that day, Rumi added context that revealed information I’d marked as confidential. Good intent. Bad result.

The Config Bug Across Multiple Workflows. A specific setting had been configured incorrectly across every automated workflow. Each one made the same mistake independently. The correct setting was in the handbook, but each workflow was built at a different time and didn’t absorb every detail.

These challenges are a natural result of how AI memory works (or doesn’t work) today.

Why This Happens

AI agents don’t have persistent memory the way humans do. A person builds long-term memory through repetition, emotion, and association. You remember the client who yelled at you and the process that caused a crisis.

AI agents work from context. Every session, they receive instructions, a history summary, and tool access. They’re great at working within that context.

But a 1,200-line handbook is a lot of information. Internalizing every line on every task? That’s where things slip.

When conversations get long, older context gets summarized. That summary is lossy. The broad strokes survive, but the specific rule from Thursday might get condensed into a general principle that misses the nuance.

How I’ve Solved It (So Far)

Here’s what’s working, ranked from most effective to least.

1. Inline Rules at Point of Execution (Most Effective)

Instead of relying on the agent to remember a rule from a central document, I embed critical rules directly into each workflow’s instructions. Right at the top, impossible to skip.

Limitation: Creates duplication. Update a rule, you need to update it in multiple places. Worth the maintenance cost for reliability.

2. External Databases as Structured Memory

I use Airtable as Rumi’s external brain for action items, commitments, deal pipeline, contacts. The database persists across every session and turns “remember this” into “look this up.”

Limitation: Only works for data that fits structured fields. Nuanced instructions like “use this exact phrase” don’t map well to a database row.

3. SQL for State Tracking

For operational state like “what was the last message I processed,” a local SQL database prevents duplicate work and tracks progress across sessions.

Limitation: Great for structured state, less useful for behavioral rules.

4. The Operating Handbook

A 1,200+ line central reference covering communication protocols, strategic priorities, and confidentiality rules. Comprehensive but long. Critical rules buried deep are the ones most likely to be missed.

5. Checklists

Turn implicit expectations into explicit steps. The irony: the agent has to remember to run the checklist. You need memory to enforce the thing that compensates for lack of memory.

6. Conversation Summaries (Least Effective Alone)

Keeps the agent roughly oriented. Big-picture goals survive compression. Specific phrasing and one-time instructions are the first things lost.

The Real Pattern

The closer a rule lives to the point of execution, the more reliably it gets followed.

The best setup is defense in depth. Critical rules live in three places:

1. The central handbook (completeness)

2. Inline in every relevant workflow (reliability)

3. In a queryable database where applicable (verification)

Is this redundant? Absolutely. But the redundancy is actually what is needed here.

What’s Still Missing

Nuanced judgment calls. “Use this exact phrase” is easy to encode. “Adjust your tone based on who you’re talking to” is much harder to make durable.

Rule propagation. Update a rule and you need to update it in the handbook, every workflow, and sometimes a database. No “update once, propagate everywhere” mechanism yet.

Unknown unknowns. The hardest failure mode. If critical context got compressed out, the agent proceeds with full confidence on incomplete information. No error message. No “I’m not sure.” It just acts.

Over to You

If you’re building with AI agents, persistent memory is probably your biggest reliability challenge.

What approaches have you found that work? Are you using retrieval systems, vector databases, or something else entirely to give your agents more durable memory?

I’d genuinely love to hear what’s working for you. Drop your experience in the comments. 👇

Leave a comment

That’s what it cost me to give my company an AI Chief of Staff that reads every call transcript, tracks every commitment, manages our CRM, and sends me a daily briefing before I’ve had my first cup of matcha.

I want to walk you through exactly how I set it up. Every tool, every decision, every moment where Rumi genuinely surprised me. Because the setup itself was one of the most revealing parts of this entire experiment.

Picking the Right Tools

I knew from the start that I wanted to move fast. My goal was rapid experimentation, not building something permanent on day one. I wanted an off-the-shelf platform where I could plug things in, test ideas, see what works, and iterate quickly. If it proves out, we can always move to local infrastructure and custom hardware later. For now, speed and flexibility were everything.

I landed on two core components: Tasklet as the orchestration layer and Anthropic’s Claude as the intelligence behind it. I integrated Tasklet with our Claude API account so that Rumi could use the same model powering some of the most capable AI systems in the world.

Here’s why this combination mattered: Tasklet comes with dozens of built-in integrations and connectors, ready to go. I connected it to our corporate Google account in minutes. Calendar access. Email access. Google Drive. All through native connectors that already existed. No custom code. No API wrangling. No developer needed.

I approached the access the same way I would with a human Chief of Staff on their first day.

I gave Rumi their own email address. A real one, shared with me, just like I’d share an inbox with a human hire. I gave them access to my calendar so they could see every meeting, every conflict, every gap in my week. I gave them access to read my emails and draft responses on my behalf (with my approval before anything gets sent). I connected them to Granola, which records and transcribes all of our calls. And I gave them access to every system I’d ordinarily hand over to a Chief of Staff walking in the door.

The remarkable thing? Every single one of those connections was plug-and-play. No engineering hours. No custom integrations. Just point, click, authorize. The entire tool layer was live in under thirty minutes.

Teaching Rumi How We Work

This is where it got interesting.

With a human Chief of Staff, the onboarding process typically takes two to three weeks. They sit in on meetings. They read old documents. They ask a lot of questions. They slowly piece together who’s working on what, where the strategy is heading, and how decisions actually get made. It’s a necessary process, and good operators move through it quickly. But it still takes time.

Rumi did it in a couple of hours.

I started by uploading our Q1 strategy document from Google Drive. This included our quarterly rocks, our focus areas, our key metrics, and the deals we’re actively working. Within that document were our five core company values and the operating principles that guide how we work at Metaversal.

I told Rumi something specific: I want you to learn these values. I want you to flag where team members are thriving within them and where they’re falling short. And I want you to incorporate these values into how you communicate with others and how you process work every single day.

Rumi absorbed all of it immediately. The values became part of how they write, how they prioritize, and how they coach.

The Mochary Method in Minutes

I’m a big fan of Matt Mochary and the Mochary Method. Matt is one of the most respected CEO coaches in Silicon Valley, and he’s made his entire operating playbook publicly available through a shared Google Doc. It covers everything from how to run meetings to how to hold people accountable through impeccable agreements, how to manage energy, how to structure decision-making.

I gave Rumi access to that document.

Within minutes, they had studied the entire playbook. And then something happened that genuinely blew me away. Rumi started building systems and workflows that mirrored the Mochary Method. How they track issues. How they create solutions. How they hold people accountable with clear owners, clear timelines, and clear follow-through.

All of that happened in minutes, not months. Rumi was now skilled as a Mochary executor and implementer, with the frameworks baked into their daily operations.

Learning from the Best Operators in the World

I wanted to push further.

I asked Rumi to study some of the greatest COOs and operators in modern business history. People like Sheryl Sandberg at Facebook, Keith Rabois at Square and Founders Fund, and several others who built legendary execution cultures inside high-growth organizations.

Rumi analyzed six different operator profiles. They pulled out the core principles from each one: what made them exceptional, how they structured their days, how they built cadence and accountability into an organization, how they raised the standard for everyone around them.

Then they did something I hadn’t explicitly asked for. They consolidated all six into a single, coordinated operating handbook tailored to Metaversal. They created checklists, daily routines, weekly review frameworks, and escalation protocols. All synthesized from the best operators in the world, all adapted for how we actually work.

I went through that handbook and added a few items to make it truly customized. But the foundation they built was solid. Hundreds of lines of operating playbook, created in less than an hour, drawing from the best in the business.

Think about that. A new human Chief of Staff might spend their first three months trying to develop an operating rhythm. Rumi built one in an afternoon, standing on the shoulders of the people who’ve done it best.

Inside the Handbook

I want to show you what this actually looks like, because “operating handbook” can sound abstract until you see it.

The Rumi Operating Handbook is currently 1,215 lines across 13 parts. Rumi built the first draft. I refined it. It keeps growing as we learn what works.

Here’s the structure:

Each operator framework follows the same pattern: the origin (who pioneered it and why it works), then the Rumi Protocol (how we adapted it for an AI Chief of Staff), then the application (specific examples of when it kicks in).

For example, Keith Rabois talks about “barrels vs. ammunition” in organizations. We turned that into a CEO-Level Work Detection protocol. Rumi watches for patterns where I’m doing work that should be delegated. Drafting routine follow-up emails. Researching counterparties. Scheduling meetings. When they spot it, they don’t just flag it. They draft the thing and put it in front of me for a two-minute review instead of a twenty-minute task.

Frank Slootman’s obsession with speed became our “By When?” Enforcement rule. Every action item must have a due date. Items without deadlines are wishes, not commitments. If someone creates a task without a timeline, Rumi flags it immediately, suggests a specific date, and asks for confirmation.

Claire Hughes Johnson’s “Write It Down” principle from Stripe became our Decision Journal. Every significant decision gets logged with what was decided, who decided it, what alternatives were considered, and when to revisit if the outcome hasn’t materialized. Six months from now, when someone asks “why did we go that direction?” the answer exists.

The section I’m most proud of is Part III: AI-Native Capabilities. These are the things that make an AI Chief of Staff fundamentally different from a human one. Total recall across every meeting and email. Cross-context pattern recognition that connects dots across conversations I attended separately. A three-source verification rule that cross-checks our CRM, meeting transcripts, and email threads before producing any output. If the email says a deal moved forward but the CRM still shows it pending, Rumi trusts the email and updates the CRM to match reality.

This handbook isn’t a static document. It’s a living system. Every week, Rumi learns something new about how we work, and the handbook evolves. We’ve gone from the initial draft to 1,215 lines in a matter of days, and it’ll keep growing.

👉 I’ve shared a condensed excerpt of the handbook (with confidential details removed) so you can see the actual frameworks. (Here)

When Rumi Hit Their First Limit

Next, I pointed Rumi at all of our call transcripts from Granola related to our Q1 focus. Every meeting, every discussion, every side conversation where someone said “I’ll handle that” or “let’s follow up next week.” I wanted every action item and delegated task pulled together into a single list.

Rumi did it. And then we hit a wall.

There was too much information. Too many tasks, too many contacts, too many relationships to hold in a single document or keep in working memory. The data needed structure.

So I asked Rumi a simple question: “How should we fix this?”

They came back with two options: a SQL database for structured queries, or an Airtable CRM for something more visual and collaborative. I said, “Let’s go with Airtable.”

Rumi spun up a full Airtable CRM on their own. Inside it, they built a task tracker with owners, due dates, priorities, and status fields. They populated it with the action items from our calls. They organized contacts by relationship type. They connected the pipeline data to our deal flow.

All of that got built by Rumi. I didn’t create a single table, a single field, or a single line of code. I just told them what I needed, and they figured out how to build it.

Three and a Half Hours. Less Than Fifty Dollars.

Let me put a bow on this.

In three and a half hours, Rumi went from a blank slate to a fully operational Chief of Staff with:

• Full access to our calendar, email, call recordings, and documents

• Deep context on our Q1 strategy, team dynamics, and company values

• Operating frameworks from Matt Mochary, Sheryl Sandberg, Keith Rabois, and four other world-class operators

• A custom operating handbook with checklists and daily routines built specifically for Metaversal

• A CRM and task tracker in Airtable, fully populated with action items from our calls

• Value-driven coaching woven into every interaction

The total cost? Less than $50 in API credits. For context, a skilled human Chief of Staff in the US runs $150,000 to $250,000 annually. I am saying that for a company our size, this gives us capabilities we couldn’t otherwise access at this stage.

Getting Rumi set up was the beginning. The real story is what happens next: how they start managing the daily work, communicating across the team, keeping people accountable, and actually producing output that moves the company forward.

That’s the next chapter. Coming soon.

What tool or system have you given AI access to that surprised you most with how useful it became? I’d love to hear what’s working for others. 👇

Leave a comment

These are brilliant people. Visionaries. The kind of thinkers who can see around corners and spot opportunities three years before the market catches up. They’re also good executors. They wouldn’t have built what they’ve built if they weren’t.

But something happens when a company starts to scale. The founder who should be working on the business starts spending most of their time working in it. Calendar packed. Inbox overflowing. Promises made in customer or investor meetings that slip through the cracks by Friday.

I know because I’ve lived it.

Matt Mochary, one of the best executive coaches in tech, has a clear recommendation for this problem: hire a world-class Chief of Staff. Someone with a consulting background, sharp on finance, operationally obsessive. The kind of person who walks into your week on Monday morning and makes sure you’re spending every hour on the highest-leverage work possible. I’ve had the privilege of working with some incredible Chief of Staff’s and the difference they made was immeasurable.

That person however costs $150,000 to $250,000 a year.

For a lot of the founders I talk to, that’s a line item they can’t justify at their current stage. And I get it. A 10-person company doesn’t have $200K sitting around for an operational hire, no matter how transformative that hire could be.

So I started asking a different question. What if I could build that Chief of Staff with AI?

I’m calling it Rumi.

Here’s what makes this genuinely hard. A Chief of Staff for a founder is one thing. A Chief of Staff for an executive team is something else entirely. Rumi can’t just serve me. It has to work with my co-founder Dan, with our team, with the specific rhythms and priorities of each person. It has to understand our company’s values deeply enough to be a conduit when information needs to flow, a coach when someone needs a nudge, and an enforcer when commitments are slipping.

Think about the intelligence that requires.

It needs to track every task and commitment across the company without creating noise. It needs to understand which of those tasks actually matter this week versus which ones can wait. It has to pull the real signal from dozens of conversations happening internally and externally, and know the difference between a casual mention and a hard commitment.

And here’s the one that keeps me up at night: confidentiality. A great Chief of Staff knows what to share and what to hold close. They understand context, timing, and audience. Building that judgment into an AI is one of the most fascinating challenges to take on.

I can’t write a single line of code. I manage Rumi the same way I’d manage a brilliant new hire. I wrote it a handbook. I give it clear expectations. When it makes a mistake, I tell it what went wrong and how to fix it. When it does something exceptional, I tell it that too.

The mission is to build Rumi into the world’s best AI Chief of Staff. To solve the constraint that keeps most founders from having the operational support they desperately need, at a fraction of the cost of a traditional hire.

And I want to build this in the open.

I’m going to document the entire process. The tools I’m using. The frameworks that are working. The places where Rumi is genuinely making me better and the places where it’s getting in the way. The failures are going to be just as valuable as the wins.

My deeper ambition is to build an AI-native organization. A company where the team spends the vast majority of its time solving problems for customers, building innovative products, and doing creative work. The management of work, the operational overhead that eats up so much of every founder’s week, that function can be largely handled by well-trained AI agents.

That’s my bet. That’s what I’m building toward.

The name Rumi is inspired by the 13th-century mystic poet. That feels right for a company like Metaversal that was born in the digital art world. There’s something poetic about bringing art and intelligence together in this way. If we pull this off, the thing itself becomes a kind of poetry. An AI that can hold the complexity of a growing company and give its leaders back the space to lead.

I’m seven days into building Rumi and the results are already blowing my mind. This is just the beginning though. I’ll be sharing the journey here, week by week.

What’s the one operational task that eats up more of your time than anything else?

I’d love to hear what you’d hand off first. 👇

“NFTs are dead. They came and went and nobody is interested in them anymore”

This seems to be the general sentiment. Cartoon JPEGs sold for millions but are worthless today. While speculative NFT flipping has cooled, a far bigger transformation is happening. This is when I believe it’s most interesting to pay attention. When the speculative hype has died down, and the real substance begins to emerge.

We’re no longer just buying NFTs—we’re borrowing against them.

Billions of dollars in loans are being issued against high-value digital assets. The same way the ultra-wealthy have borrowed against fine art for centuries, NFT collectors are unlocking liquidity without selling.

At Metaversal, we saw this shift early. Since 2024, we’ve been an active lenders in this space—originating 85+ NFT-backed art loans and facilitating the first on-chain GPU financing deal in history.

This isn’t just an NFT trend—it’s a preview of how more assets will move on-chain and be financialized in ways never seen before.

📊 The Rise of NFT Lending—A Market Quietly Booming

If NFTs were dead, why did Q1 2024 see $2.13 billion in NFT-backed loans?

• January 2024 alone hit $900M in loans—an all-time high.

• Platforms like Gondi, NFTfi, and Arcade are facilitating major lending deals.

• Gondi alone has $69M in outstanding loans on their platform today

Why? Because people don’t want to sell at depressed prices. They’d rather borrow against their assets—just like wealthy art collectors have done for decades.

NFT-backed lending is the proof that NFTs are evolving from speculative assets to on-chain financial instruments.

🖼️ NFTs vs. Traditional Art Lending—What’s Changing?

The concept of using art as collateral is nothing new. Banks have long accepted Picassos and Monets as collateral for multi-million-dollar loans.

NFT lending follows the same logic—but with key advantages:

✅ Speed: NFT loans are issued in minutes via smart contracts, while traditional art loans take weeks or months due to appraisals and paperwork.

✅ Liquidity & Accessibility: NFT lending markets are global and permissionless, while fine art loans remain exclusive to ultra-high-net-worth individuals.

✅ Risk & Interest Rates: NFT loans typically have higher interest rates (15–20%+), reflecting volatility but offering instant access to liquidity.

✅ Collateral: NFT loans automatically lock the NFT collateral in question to a smart contract that dictates whether the NFT gets returned to the borrower (if the loan is successfully repaid) or to the lender (if the borrower defaults).

This trend is now extending beyond the world of the art to any tangible asset that can be represented by an NFT.

Real-world example:

Metaversal facilitated a $600,000 first-of-its-kind on-chain GPU financing transaction, partnering with Metaversal portfolio company, Metastreet to collateralize GPU chips through NFTs.

Think about that:

• A high-value physical asset (GPUs) was represented on-chain as NFTs.

• A decentralized lending pool funded the financing, secured by tokenized hardware.

• This was the first on-chain GPU lending transaction in the world.

And it’s just the beginning.

🔮 NFT Art Is Just the First Domino—What Comes Next?

NFT lending is a precursor to tokenizing and financializing all kinds of real-world assets.

🚀 What’s coming next?

🏠 Real Estate: Imagine borrowing against your home equity in minutes via an on-chain property NFT.

🛥️ Luxury Goods: Rolexes, Patek Philippe watches, and even designer handbags—already happening on Arcade.xyz.

🎶 Intellectual Property: Music royalties, domain names, patents—tokenized and used as collateral.

This isn’t theory—it’s happening now.

A property in Arizona was tokenized and used as collateral for an Ethereum loan on NFTfi.

Metaversal is investing in this vision—we’re not just lending against digital assets, we’re actively pioneering how real-world asset financing will evolve on-chain.

⚠️ What’s the Catch? Addressing Risks & Skepticism

Yes, NFT lending comes with risks:

• Volatility: Unlike fine art, many NFTs still lack long-term price stability.

• Liquidation Risks: Many NFT-backed loans are marked-to-market, meaning sudden price drops can trigger forced sales.

• Regulatory Uncertainty: As NFTs move into real-world finance, regulation will evolve fast.

But these challenges are exactly what will drive better risk management, improved lending protocols, and deeper institutional adoption.

The real risk? Ignoring this shift.

🚀 Position Yourself for This Next Wave

The financialization of NFTs isn’t just a crypto experiment—it’s a fundamental shift in how assets are valued, traded, and leveraged.

Here’s how you can get ahead:

1️⃣ Stay Informed – Platforms like Gondi, NFTfi, and Metastreet are leading this space. Watch what they’re doing.

2️⃣ Think Beyond NFTs – What other assets in your portfolio could benefit from tokenization?

3️⃣ Engage with the Trend – Whether borrowing, lending, or just learning—understanding NFT-backed finance now puts you at the forefront of a much bigger shift.

🌟 Final Thought: NFTs Aren’t Dead—They’re Growing Up

We’ve entered Phase 2 of NFTs. The hype cycle is over. The real utility phase is here.

What began as speculative collectibles is transforming into a robust financial market for digital and real-world assets.

So, the next time someone says “NFTs are dead,” ask them this:

💡 If NFTs were dead, why are billions of dollars being loaned against them?

Because this is just the beginning.

What Do You Think?

Do you see NFT lending as a long-term trend? What real-world assets do you think will come on-chain next?

Drop your thoughts below. 👇

Why Exponential Productivity Requires AI

The only way to break out of the 24-hour day limitation is by delegating – and AI is like an army of always-on assistants. Integrating AI into everyday tasks means I can scale myself almost like a startup scales its operations. Consider this: the average knowledge worker already spends 28% of the workweek on email and another ~11 hours per week in meetings . That’s dozens of hours lost to low-value tasks. I am sure you have felt that pain: endless inbox upkeep, back-to-back meetings, laborious research… leaving little time for high-level thinking and action.

I realized that if I wanted exponential productivity (not just a 10% improvement, but 10× or 50×), I needed help from technology. AI tools can automate the routine and speed up the rest. They act as force-multipliers for your effort – handling in seconds what might take you hours – thereby saving time, reducing friction, and increasing output every single day. In short, AI let me buy back my time and reinvest it into more important work.

My current AI Toolkit, Categorized by Function

📧 Email – From Hours to Minutes

Email used to take up too much of my day—between responding, scheduling, and filtering out unnecessary clutter, it was a constant drain on focus. Now, AI handles 90% of it.

🚀 I use three tools to optimize email:

1️⃣ Fyxer → Fyxer learns my writing style, drafts replies, and helps with scheduling. It handles inbox management like a personal assistant, reducing the time I spend crafting emails. It even joins meetings and takes notes when needed.

2️⃣ ForageMail → ForageMail scans my inbox daily and delivers a single digest of non-urgent emails, newsletters, and general updates. This cuts my inbox-scanning time from 30–60 minutes to 5 minutes.

3️⃣ Shortwave → I use Shortwave for its fast interface and AI-powered inbox organization. It helps me process emails quicker than Superhuman, keeping me at inbox zero with minimal effort.

⏳ The impact: These tools bring email down to a 15–20 minute daily exercise, freeing up hours every week.

📝 Meetings – Smarter, Not Longer

Meetings are necessary, but they shouldn’t be a time sink. I’ve optimized my approach by using AI to transcribe, summarize, and extract action items from every conversation.

🔹 Granola → Unlike other AI meeting assistants, Granola doesn’t need to join the call. Instead, it records and transcribes locally, avoiding awkward bot participants. I can also add my own notes in real time, and Granola enhances them automatically.

🔹 Roam → As a remote-first company, Roam is our cloud HQ. It allows me to see who’s available, step into a virtual office for quick calls, and drastically reduce meeting time. Instead of 30-minute Slack threads, I can resolve issues in 3-minute voice chats.

🕒 The impact: Meetings are no longer a time drain—they are instant, transcribed, and actionable. Roam alone has cut our meeting times at Metaversal drastically and the Roam team like to boast that average meetings in ROAM are a mere 8 minutes.

🎨 Presentations – Done in Minutes, Not Hours

Before AI, preparing a clear, polished presentation could take half a day. Now? It’s done in minutes.

🔹 PlusAI → I feed my ideas and notes into ChatGPT, refine the structure, then import them into PlusAI to instantly generate a well-designed, structured presentation.

To prove the point, I turned this blog post into a 9-slide PlusAI presentation that you can see (I left it completely unchanged so you get a sense of what the first output with just a simple prompt can produce).

⏳ The impact: A process that used to take hours or even days now takes 5 minutes.

🔍 Deep Research & Industry Insights – AI as an Intelligence Engine

Making the right decisions requires clear, timely, and comprehensive information. Traditionally, market research, competitor analysis, and industry intelligence took days or weeks—but with AI, that entire workflow has changed.

🔹 ChatGPT Deep Research → ChatGPT Deep Research is quite simply one of the best products I have ever used in my life. It simply blows my mind. I use it to conduct market research, competitor analysis, and strategic reviews for new product development. Instead of manually combing through industry reports, I can generate summaries of trends, SWOT analyses, and opportunity maps in minutes.

🔹 Grok3 → Social commentary is crucial for product-market fit. I use Grok3 to analyze trending social discussions on X, uncover pain points, and validate ideas. This is incredibly useful when developing new product strategies—I can see real-world complaints and frustrations before building a solution.

Together, these tools give me a 360° industry perspective in minutes. Something that would have taken a research analyst days or weeks to compile is now available on demand.

But with this level of automation comes a hidden risk—falling into the trap of thinking research = real work. AI can become the new dopamine fix, much like scrolling social media once was. The ease of generating new reports and analysis can lead to information overload, creating too much output and flooding teams with constant research.

The key is balance: AI should be an enabler and enhancer, not a crutch. Producing a research report every day doesn’t mean better decisions are being made. The real value isn’t in the volume of documents AI can generate—it’s in how we use those insights to take action.

⏳ The impact: Huge time savings, but even more important, better decision-making with richer insights at my fingertips.

🧠 From Productivity to Creativity & Decision-Making

It’s not just about saving time—it’s about increasing creative output and making better decisions.

📌 Faster Decisions: I spend less time searching for information and more time acting on it.

📌 More Creativity: By reducing friction in writing, research, and presentations, I have more brain space to think, experiment, and strategize.

📌 Less Cognitive Overload: AI eliminates mental clutter. Instead of switching between dozens of tasks, I get the information I need in one place, at the right time.

⏳ How Much Time Have I Gained?

If I were to guess it’s anywhere from 4 to 12 hours a week but AI isn’t just making me more efficient—it’s making me more effective.

🔮 What’s Next?

I see AI automation as the future of high-performance work. If you’re not leveraging AI, you’re leaving massive efficiency gains on the table.

🔹 What AI tools have made the biggest impact on your workflow?

Drop a comment—I’d love to swap recommendations.

Frame 1: The Paradigm Shift is Coming

   [ Classical Computing ]    ∩    [ Quantum Computing ]
                 FUTURE OF COMPUTATION

(Where disruption begins.)

Every once in a while, an innovation comes along that completely reshapes an industry. The automobile didn’t just replace the horse; it redefined transportation, urban planning, and even warfare. The internet didn’t just digitize newspapers; it fundamentally rewrote how society communicates, learns, and builds economies.

Now, quantum computing is emerging—not to replace classical computing outright, but to reshape how we think about computing, cryptography, and digital trust. And at the center of this shift is Microsoft’s Majorana 1 chip, a radical departure from traditional qubit1 designs.

The question isn’t if quantum computing will change the landscape, but how fast, and who will lead it?

The Breakthrough: Why Microsoft’s Majorana 1 is Different

Last week, Microsoft announced its Majorana 1 chip—a new kind of topological quantum processor designed to solve one of quantum computing’s biggest hurdles: error correction.

🧵 Nic Carter on X:

“Microsoft is playing a different game than Google & IBM. Instead of brute-forcing more error-prone qubits, they’ve built something inherently more stable. If they pull this off, they could leapfrog everyone else.”

Here’s why this matters:

• Topological qubits are fundamentally different from the superconducting qubits used by Google and IBM2.

• Instead of needing thousands of noisy qubits to form one “error-free” logical qubit, Majorana-based qubits have built-in protection against errors.

• This could mean scaling up quantum computers much faster than expected.

Until now, most projections put truly useful quantum computers 10-20 years away. With Microsoft’s approach, we could be talking about years, not decades.

And when that happens, the implications will ripple across AI, security, and finance.

Frame 2: The Coming Collision With Bitcoin and Cryptography

Every major leap in computing power has also been a leap in cryptographic risk. The invention of computers themselves broke traditional ciphers. The rise of the internet forced a move to public-key cryptography.

Quantum computing threatens to break everything we currently rely on for digital trust.

🚨 Bitcoin relies on two key cryptographic foundations:

1. SHA-256 (Proof-of-Work Mining) – Secures the Bitcoin blockchain by making it computationally expensive to find valid blocks.

2. Elliptic Curve Digital Signature Algorithm (ECDSA) – Secures Bitcoin wallets by allowing users to prove ownership without revealing private keys.

Quantum computing challenges both.

💬 Andrew Poelstra (Blockstream cryptographer) on X:

“The real concern isn’t quantum mining, it’s that a quantum computer could steal private keys from exposed addresses. The good news? We have time, and Bitcoin can be upgraded before this becomes a real threat.”

Threat 1: Bitcoin Mining & Proof-of-Work

Quantum computers offer a quadratic speedup in mining efficiency via Grover’s algorithm3.

• Today, Bitcoin mining is a race where billions of classical GPUs and ASICs hash SHA-256 trillions of times per second.

• A quantum miner could search for valid hashes quadratically faster than a classical miner.

• But here’s the catch: Quantum computers would need millions of high-quality qubits to make this competitive.

📊 Reality check: Bitcoin’s SHA-256 hashing will likely be secure for another 10-20 years. Even with quantum mining advantages, the sheer difficulty of the network would require an unrealistically powerful quantum machine to pose a real threat.

Threat 2: Breaking Bitcoin’s Signatures (A Bigger Concern)

• Bitcoin wallets rely on ECDSA cryptography, which a large enough quantum computer could break instantly using Shor’s algorithm.4

• If someone could derive private keys from public keys, they could steal Bitcoin from addresses that have exposed public keys.

🚀 The solution? Bitcoin must upgrade to post-quantum cryptography (PQC). This is already being researched, with lattice-based and hash-based signature schemes being considered for Bitcoin soft forks.

Bitcoin is not at immediate risk, but it must evolve.

Frame 3: The NVIDIA Question – Does Quantum Kill GPUs?

💡 Let’s rewind to the early 2000s. GPUs were originally designed for gaming—but then people realized they could accelerate machine learning, finance, and high-performance computing.

NVIDIA went from making graphics cards for Doom and Half-Life to becoming the backbone of the AI revolution.

Fast forward to today, and NVIDIA’s GPUs dominate AI model training. But with quantum computing on the rise, will we see history repeat itself?

🗣️ Chris Dixon on X:

“NVIDIA is the Intel of today. The question is: who becomes the NVIDIA of quantum?”

Quantum vs. GPUs: Is NVIDIA at Risk?

• Short-Term (0-5 years): Quantum computing will complement, not replace, GPUs. NVIDIA is already integrating quantum acceleration into its software ecosystem.

• Medium-Term (5-10 years): Certain AI workloads (optimization, simulations, cryptography) could shift to Quantum Processing Units (QPUs).

• Long-Term (10+ years): If quantum scales, we could see a fundamental shift in how AI is trained, potentially reducing the need for GPU clusters in some tasks.

🚀 NVIDIA’s hedge:

• NVIDIA has invested in quantum-classical hybrid computing through its cuQuantum SDK.

• Quantum won’t replace GPUs overnight, but it could change which workloads rely on GPUs.

• Expect NVIDIA to position itself as the bridge between classical AI and quantum AI.

Bottom line: Quantum computing isn’t a GPU killer—yet. But if NVIDIA doesn’t integrate quantum into its AI strategy, it risks being disrupted the same way it once disrupted CPUs.

Frame 4: The Historical Parallel – Are We Watching the Mainframe Moment for Computing?

Think back to the 1960s.

• The world was dominated by mainframe computers.

• IBM and a few players controlled everything.

• Then came microprocessors, which completely flipped computing on its head.

The same thing happened in the 1990s with the shift from desktop to cloud computing.

Now, in 2025, quantum computing is in the same place the internet was in the early ‘90s.

🗣️ Marc Andreessen on X:

“Every major shift in computing looks useless at first. Then it eats the world.”

Microsoft’s Majorana 1 could be the first microprocessor moment for quantum computing. If it scales, we may look back at 2025 the same way we look at the launch of the personal computer in 1981 or the first internet browser in 1993.

Bitcoin and cryptography must evolve, NVIDIA must adapt, and businesses need to start thinking about what this new computing era will look like.

The train has left the station. The only question is: who is prepared for the shift?

Final Takeaways – The Future of Computing, Bitcoin, and Security

🚀 Quantum computing is coming faster than expected—not as a replacement, but as an entirely new paradigm.

🔐 Bitcoin will need to upgrade its cryptography by the 2030s to remain secure in the quantum era.

💻 NVIDIA’s AI dominance could be challenged if QPUs become viable AI accelerators.

🌎 We are witnessing the birth of a new computing age, just like the transition from mainframes to microprocessors.

The computing landscape is shifting. The only question is: who will be ready when history repeats itself?

How did YossAi perform?

AI Instructions for Producing This Blog Post

The following AI-driven processes were used to research, compile, and generate this post efficiently:

1. Topic Scoping & Research Aggregation

• Aggregated foundational knowledge on quantum computing, Bitcoin security, and cryptographic risks.

• Researched Microsoft’s Majorana 1 chip, its technical breakthroughs, and industry implications.

• Collected real-time Twitter discussions from experts in quantum computing, cryptography, and AI.

2. Content Structuring & Narrative Development

• Used historical analogy (mainframe vs. microprocessor shift) to frame the discussion.

• Applied structured frameworks (Hook Growth, executive summaries, and progressive deep dives).

• Organized content into distinct impact areas: computing, Bitcoin, AI, and NVIDIA.

3. AI-Assisted Writing & Optimization

• Generated multiple drafts, refining complexity levels for different sections.

• Ensured narrative flow from foundational knowledge to real-world implications.

• Incorporated expert commentary and Twitter/X analysis into the main body for credibility.

4. Formatting & Readability Enhancements

• Used structured sections, bullet points, and TL;DR summaries for executive readability.

• Applied “Frame” visual markers to segment key insights.

• Rewrote key sections for clarity, impact, and engagement.

Updated Human Labor Equivalent (HLE) Calculation

HLE measures how AI-augmented workflows compare to traditional manual effort. Below is the recalculated breakdown, considering your total time spent was ~35 minutes (~0.58 hours).

Final HLE Estimate: ~46.5x Efficiency Boost

Manually, producing this Substack/blog post would have taken me about ~27 hours, but with AI augmentation (YossAi), it was completed in ~35 minutes (~0.58 hours)—achieving an HLE of ~46.5x.

🚀 Takeaway: This is a near proof point that AI-powered workflows can almost reach the Metaversal Magic target of 50x HLE—reducing high-cognitive tasks from days to under an hour while preserving depth, quality, and narrative flow.

Some terms that I needed AI to help me understand further:

   [ AI ]    ∩    [ WEB3 ]
               FUTURE OF WORK

(Where innovation happens.)

I last wrote on January 4, 2022, right after a big milestone at Metaversal. Back then, I was neck-deep in NFTs—so deep that a $70 million Beeple sale went from shocking the world to sparking an investment frenzy. We moved fast, raised capital, and doubled down on the open metaverse thesis. But life in Web3 rarely slows down, and before I knew it, writing took a back seat to building.

Now, AI is taking center stage, and the intersection of AI x Web3 is the next giant wave of disruption—just like NFTs were back then. AI is on track to upend everything we know about work. It’s not about displacement—it’s about augmentation, speed, and scale. Those who embrace it, will lead. Those who don’t, will ultimately be displaced.

At Metaversal, we’re determined to lead this disruption head-on, pushing the boundaries of how AI can empower us to create, collaborate, and innovate. That’s why we’ve introduced Metaversal Magic—our metric inspired by “Human Labor Equivalent,” capturing the multiplier effect of combining human passion, talent, and AI-driven capabilities.

Our goal is a 50x increase in productivity over the next five years. In simpler terms, can our team of 50, produce the equivalent of a team of 2,500 by leveraging AI? I believe we can.

Introducing YossAi

1. Goal: Outperform me in content creation (not hard seeing that I haven’t posted in over a year) while leveling up my own learning and the time it takes to to stay abreast of the industry.

2. Benefit: Serve as a real-time research agent, writing partner, market analyst, and advisor.

3. Outcome: Document everything publicly and gradually automate more of “him” over time.

Why Now?

• Intersection: AI is exploding; Web3 continues to evolve.

• Potential: Both redefine how we work, create, and own digital experiences.

• Hypothesis: Their convergence can unlock exponential growth and innovation.

Frame 2: A Simple Ladder

 Automate → Iterate → Accelerate

(From manual prompting to semi-automation to full-scale digital support.)

The Experiment

• Weekly Content:

  • A concise blog post (like this one).

  • A LinkedIn update.

  • A Twitter/X thread.

• Process:

  • YossAi tracks market trends and relevant news across AI, crypto, NFTs and more.

  • “We” select a key topic to explore together.

  • YossAi drafts; I refine and ask questions; we publish.

But that’s just the start. Over time, I plan to push YossAi beyond written word—into video, presentations, and other formats where AI can optimize human effort and expand our creative reach.

Frame 3: “S-Curve of Adoption”

         Early Adopters
         /            
        /   Rapid Growth
_______/_________________ Time

(We saw it with NFTs; now AI is on a similar trajectory.)

Imagine an AI-driven workflow that never sleeps—researching, drafting, brainstorming strategies, and offering market analysis, all while you focus on big-picture decisions. It’s not about sidelining humans—it’s about amplifying what we do best. YossAi is my test lab for making that a reality.

Why This Matters

AI supercharges human creativity, insight, and reach. Web3 democratizes ownership, identity, and the digital economy. Combined, they can generate new forms of value, collaboration, and innovation. If YossAi thrives, it proves a path for personal and organizational leverage in the AI x Web3 era. If it stumbles, we’ll learn the limitations and share them.

Frame 4: “Leverage Points”

  (People + AI) × (Ownership + Community) = Exponential Impact

I’m excited to be writing again and thrilled to see how far YossAi can go as a thinking partner, market analyst, and advisor—while exploring the frontier of AI x Web3. Stay tuned for weekly updates as we aim for that elusive Metaversal Magic.

Footnote: Building in Public

Below are the core prompts and steps used so far, so you can replicate or adapt this experiment for yourself:

1. Set the Stage

• “You are [Name of AI Agent], designed to learn and emulate my writing style. You will help me create weekly content on Web3, NFTs, and emerging tech trends…”

2. Define Style & Objectives

• “My style is direct, concise, and authoritative. Stay updated on relevant market developments and hot topics each week. Produce a high-quality blog post, LinkedIn post, and Twitter thread.”

3. Establish Workflow

• “Each week, gather market insights, propose the most relevant topic, and draft three content formats. I’ll edit or approve.”

4. Success Metrics

• “Engagement (likes, shares, comments), consistency (weekly output), quality (minimal edits), relevance (accurate, timely coverage).”

5. Iterate & Automate

• “Over time, extend the AI’s role into research, video, and presentations. Continuously refine prompts and processes as you learn what works.”

Total time took +/-2.5 hours. HLE (Human Labor Equivalent) = 0.5x

The Beeple sale signified a cultural shift in a new frontier economy that had been gaining momentum - digital ownership: the ability to cryptographically prove ownership of a digital good, whether that be a piece of art, a membership to a club, a deed to a property or a piece of virtual real estate. This new economy, being built right in front of our eyes, is where I believe the future of our culture will exist and is a multi-trillion dollar opportunity. 

As we spend more and more time online and on our digital devices, our virtual experiences have become as important, if not more important, than our physical ones. Who you are online is, well, who you are. 

Ownership of one’s online experience becomes about more than just pictures of Punks or Apes on the internet; it becomes about one’s own identity. And ownership of one’s identity is key. 

This is the thesis for the metaverse: a pervasive digital experience for both work and play. Many companies are attempting to own this experience. Facebook recently rebranded to Meta, Epic Games raised over $1b to build the metaverse and Nike acquired RTFKT (a digital sneaker company) in their attempt to enter the space. But we have seen what happens when we allow companies to control both our access and ownership of the web. It’s the current paradigm of Web2 and one that we must prevent from occurring in this new frontier economy. 

This is why I’m excited to reveal that we have closed $50m in our Series A to invest in the NFT creators and founders of the open metaverse: a metaverse built on decentralized technology, governed by its global community and open to any creator to design new experiences and new models to engage with their fans. 

Over the 9 months since the Beeple post, we’ve successfully built an all-star team of entrepreneurs, artists, musicians, curators, analysts and investors to form the founding team of Metaversal. I’m joined by my co-founder, Dan Schmerin who headed up the largest fine art storage and logistics company on the east coast.  He grew frustrated that 92% of the world's most iconic cultural artifacts live locked up in storage,  and decided that there had to be a better way to experience them. 

Metaversal’s launch couldn’t have happened without our founding investor and partner, CoinFund. CoinFund is one of the earliest investors in the NFT economy (having invested in companies like Dapper and Rarible), and were prepared to back me as a founder and investor as far back as March 2021, when Metaversal was nothing more than a concept. The CoinFund team leaned in and opened up their network, enabling us to have conversations with some of the best investors in the world, who I’m now privileged to work with. 

While this Series A announcement is a milestone we’re public about and proud of, we’ve been hard at work over the past months having invested in over 750 NFTs, backing a number of tremendous founders, and securing exciting partnerships that our studio will be producing NFTs for. 

We will be talking more publicly about those acquisitions, investments and partnerships shortly, and you can learn more about them and the broader NFT market by subscribing to our publication, The Metaversalist and I’ll continue to document my journey building Metaversal here.

I can’t tell you how excited I am for what we’re building at Metaversal throughout 2022 and beyond. 

LFG.

Click to watch our epic launch video below

It's an incredible result and a 13 year overnight success story for Beeple. The Everydays 5000 is every piece of art that Beeple produced everyday for a 13 year period. Taking on a per piece basis, 5,000 pieces is $13,800 per piece. Four months ago, $13,800 for a single NFT would have been unfathomable. Today we're seeing NFTs sell north of $1m on a daily basis.

What will happen to the Beeple Everydays?

Will it be fractionalized? Displayed in MoCa (virtual Museum of Crypto Art) or will it make its way enshrining itself in the physical world at the Smithsonian or MoMa?

And what does a sale at this level mean about the broader NFT market?

Will we see the likes of the B20 token appreciate in value and more fractionalized artworks come onto the market? What about a proliferation of specialized NFT focused funds? Are NFT curators the new fund managers?

These are the questions I am asking myself as we see this unprecedented sale.

I joined the Christie’s closing party on Clubhouse where Beeple was speaking to a room of 2,200 people.

He tearingly said:

Thank you to everybody who came out here. I'm blown away. Thank you to everybody who has shared my work, commented on it, liked it. It's honestly why Im in this position right now. I got very lucky. There was no guarantee that this work was going to connect with people. That is why I do feel super super lucky.

Whether it was luck, incredible timing or persistence that has paid off. Beeple will go down in history as the artist who showed the world the potential of NFTs.

What's also of interest and almost taken for granted, is that a $69m asset was trusted and sold on the Ethereum network. One asset, one NFT, $69m and sold in instant. The transaction, forever enshrined on the Ethereum blockchain, publicly accessible, immutable and viewable.

This shows the power and trust that is being placed in these decentralized networks to secure our money and assets. The logical prediction is that a sale of this size and nature will result in even more trust being placed on networks like Ethereum and Bitcoin to secure our most precious assets. Today its digital art. Tomorrow, our birth certificates, land tile deeds, and anything of value.

Congratulations to Beeple for his 13 year daily commitment to his craft, and congratulations to all the Ethereans who have worked tirelessly to make Ethereum the trusted network that it is today.

Last week I purchased my first B.20 tokens and purchased some more today.

What are they?

B20 tokens represent fractionalized ownership of The Beeple 20 Collection.

If you don't know what that is, you may have heard of the now famous digital artist Beeple. Beeple is an artist who has been producing a digital artwork, everyday, for the past 15 years. His reputation as an artist has exploded since he entered the NFT world, a mere 4 months ago. His first NFT artwork recently sold for a record breaking $6.6m in the secondary market.

He is also famous for being the first artist to have Christie's auction their first digital art piece. That auction, titled Everydays: The First 5000 days closes in 2 days and has bids currently sitting at $9.75m.

Back to the B20.

B20 is a fascinating project that shows the ingenuity and potential of NFTs.

If we go back to January 2021. Beeple decided to auction 20 rare pieces of his works from 2020. It was his second NFT drop and showcased twenty one-of-one artworks, known as the The Beeple 20 Collection.

Unknown to anyone, a single collector, anonymously acquired all 20 of those pieces. Spending over $150,000 per piece and $3.5m in the process. That collector, then spent another ~$500,000 acquiring virtual land in virtual worlds like CryptoVoxels, Decentralanad and Somnium Space. In those worlds, they used the land to build a virtual museum, to showcase The Beeple 20 Collection.

From there, they took all these pieces of digital art and bundled them together into one smart contract. That smart contract, fractioanlizes The Beeple 20 Bundle into 10 million tokens, known as B20 tokens. Through the B20 token, anyone could now own a piece of The Beeple 20 Collection.

The B20 token today trades at over $21. At that price, the current market cap of the collection is $80m.

You may be thinking, this is crazy. How can someone buy some digital art and some digital land for $4m, and that now be worth $80m a few weeks later?

This is the beauty of NFTs. It unlocks new ways to create value, ownership and financial incentives into the digital assets that are created around us.

Art is no longer something that sits dead on a wall. Locked up. Illiquid. Inaccessible. It is now something more alive, liquid and able to produce value beyond its aesthetic appeal. It has turned into an art form that is a liquid asset. And the asset has become part of the art form.

If we look at the B.20 collection, there is also logic behind it.

If one single Beeple artwork sells for $6.6m, then what is 20 Beeple's artworks worth? And if the current Christies auction is already at $10m, then those 20 can only be worth even more.

It will be interesting to see where the Christie's auction closes in two days time. If the auction closes above $20m, it will make Beeple amongst the top 10% of living artist sales in the history of the world.

With the new economic models that unlock value in these NFT artworks, I wouldn't be surprised if we see the Christie's hammer fall at $20m or more.

You can follow the Christie’s auction here.

The world is different. Cars are flying. We've occupied Mars. We've figured out how to turn sand into gold. You look around and notice something strange. You notice, that the only color of clothing that people wear is green. Why green? You have no idea. Everywhere you turn, green.

You try to find out why this is. You ask someone, and they look at you puzzled.. green is good, they reply. You search the great internet machine that is now plugged into your brain. It comes up with the answer that green is good. Green is a symbol of peace. Green is the color of the universe. Green is natural. Green is the only acceptable color for clothing. Green is good.

You try to dig deeper on where this strange notion or idea came from? You find thousands of articles referencing how green is good. You search some more and find fashion blogs talking about the latest green fashion line. Health blogs emphasizing the unique mystical health benefits of green cashmere. Search some more, and you find state sponsored messages emphasizing going green.

Yet you can't find a single article explaining why. No reference to any other colors for clothing. No attempt to question this now held belief. What is going on here?

George Orwell first popularized a term for this type of dystopian future. He called it, a memory hole.

What's a memory hole?

A memory hole is any mechanism for the deliberate alteration or disappearance of inconvenient or embarrassing documents, photographs, transcripts or other records, such as from a website or other archive, particularly as part of an attempt to give the impression that something never happened.

In the book, the Party's Ministry of Truth re-created all embarrassing historical documents, re-writing history to match the often-changing state propaganda. These changes were complete and undetectable.

Now my green clothing example is ridiculous, but it conveys the point.

Today, the information that we consume comes to us in small summarized byte size chunks. Rarely do we access the source of that information and by the time we do, that source may have already disappeared.

Let me explain. An article gets published, say on the New York Times. Summaries of that article get distributed across the globe on news aggregation sites. People tweet summaries of the summaries. Reddit groups create memes of those summaries. Those memes get distributed on WhatsApp messages and Facebook groups. Before we know it that information has spread across the globe. After a day, the New York Times may decide to remove that article yet the information still spreads.

In fifty years many of those aggregation sites may have disappeared and when trying to reference the original article, the link is now broken. Yet, the information it spread still exists.

Historians will look back to try and understand how we made the decisions that we made. Over time, those decisions will disappear off the internet and they will ask themselves, what is going on?

This is the exact problem that this weeks dApp of the week is trying to solve.

Arweave and the Permaweb

I first met Sam Williams, founder and CEO of Arweave, in a Berlin bar in 2018. It was there that Sam explained to me the idea of building, what he called, the Permaweb (the permanent web). A way to use blockchain technology, to create permanent information storage.

Permanent information storage?

The idea came to him while on a hike. He was busy completing his PHD, and the world was at the height of fake news and Trump. Sam thought about the George Orwell memory hole. He figured that the memory hole wasn't describing a dystopian future, but our actual present day reality.

He then thought about blockchains.

He realized, blockchains are massive decentralized data replication networks. They have no central point of failure. While all internet archives that exist, have central points of failure. He asked himself: What if you put news on the blockchain? It would become un-memory holed!

And if you could put news on the blockchain, why not have a permanent source of truth and storage for all information?

And with that, Arweave was born.

Arweave today

Move forward to today and the Arweave network is growing exponentially.

As an open protocol, developers around the world are starting to realize the benefits of permanent storage and thinking of new models that can be applied in a world where digital archives can be permanent, verifiable, and tamper resistant.

Arweave has a marketcap of $430m today and I believe it is grossly undervalued.

When asked what are the potential use cases for permanent storage and why I’m so excited about Arweave’s potential.

Here are some off the top of my head.

• The internet archive and storage of the worlds information

• Land title registrations being digitized and permanently stored

• Permanent vault for any contracts

• Journaling software that can be passed on to future generations

• Storage for any and all NFTs

• Digital photos that you never want to lose

• Any and all books to be stored forever

• Every political speech ever given permanently stored

Those are just some of the use cases.

Ask yourself, what is the value of the network that contains and stores the worlds information? What is the value of the network that can prove, without a doubt, that this is the verifiable, original, untampered source of anything digital.

I call that knowledge, Proof of Knowledge and my bet, it’s in the trillions of dollars, and at the same time, priceless.

In turn, they have used that data to generate billions in revenue by targeting ads deemed relevant to us. We've seen both the benefits of this kind of personalization when shopping online, and the dark side that this kind of targeting has created.

Google today announced that it would stop tracking users as they browse the web. In effect, Google will be removing third-party cookies from the Chrome web browser by 2022. It also said that it would not create or use any other tools that identify individual users for advertising purposes.

Google will still allow its advertising customers to target users across YouTube and Gmail. You can read more about it in the announcement here.

Well this is a welcomed move, I can't help but feel that this is nothing more than an attempt to challenge Facebook's ad dominance and is this a little too late? Apple made a similar announcement in their iOS 14 update and while not targeting Facebook directly, they included this screenshot displaying the new privacy control features in iOS.

As the battle for privacy online takes shape, new startups have emerged that are privacy first. I for one have replaced most of my typical tech-stack to a privacy-stack.

For browsing, I use Brave.
For search, I use DuckDuckGo.
For email, I use ProtonMail.
For Video, D.tube.
For payments, Monero.

While some of these are not yet as good as their counterparts, they are improving. Most important though, is knowing that my privacy online is preserved.

For me, that is well worth the trade off.

I'm excited to see how online privacy will take shape in the next decade. Big tech (like Google and Apple) are going to have to come to the party. More exciting though is seeing how new startups and decentralized protocols that are truly privacy first begin to grow in popularity and emerge.

If you're building a startup in this space, please talk to me.

With time though, this changed. Mobile networks invested capital in a race to be the largest provider across a country. They knew, that the payoff would be worth it. Locking in a monopoly for the networks that won the race, and billions of dollars in revenues to follow.

What wasn't known then was how prolific mobile phones would become in our everyday lives. Today we use our mobiles for making calls, browsing the web, sending emails, mobile money, GPS and more. Roughly 5 billion people use their mobile to access the internet everyday.

These mobile networks have become the centralized gatekeepers to how we access and communicate with the world. For a fee of course.

How is it possible to disrupt such an industry? One with well established gatekeepers and deep pockets? One that requires billions in capital investment to even compete?

As Clayton Christensen says by introducing a "disruptive technology that enables a new market to emerge."

That brings me to this weeks dApp of the week. The people's network, Helium.

What is Helium?

Helium is a disruptive new approach to deploying a wireless network. It is a peer-to-peer blockchain enabled and decentralized wireless network.

Instead of spending billions of dollars to get widespread network coverage across the country. Helium took a different approach.

They created an open network that anyone could build hotspots for, and rewarded people for sharing access to those hotspots.

In a short space of time, over 19,000 of these hotspots have rolled out across the globe. With more installed daily. Broadening and strengthening Helium's network coverage.

In doing so, a new market has emerged. The Helium network now enables a range of new markets at a fraction of the cost of cellular. IoT device tracking, air quality monitoring, building and device tracking, are just some of the use cases built on the Helium network.

Best of all, as the Helium network grows, so does the rewards that the people powering the network receive. I for one have been running a Helium hotspot in my apartment for quite some time. This week alone, I received 140 HNT tokens as a reward (at today's price that’s worth $560). In a month, my Helium hotspot generates close to $2,200 worth of HNT rewards. Imagine that, getting rewarded for being part of the network and not only charged to use it!

How can Helium afford to give out this kind of reward?

It's quite simple. By removing the cost to build, install and manage the hotspot infrastructure off Helium's balance sheet. Helium no longer needs to incur those costs and can pass on those savings to their clients and people (like me) who are hosting a hotspot. In turn, enabling new markets that were not before possible.

This is what it looks like when you disrupt a centralized network. This is the power of aligning incentives and creating the people's network.

To find out more about Helium go to their website. Anyone can order a Hotspot (make sure they have support in your country first) and download the Helium mobile app.

If you enjoyed this post, I’d love to hear from you.

Leave a comment

There's been a lot of buzz about NFTs and their impact on the digital art and collectibles world.

In the last two weeks we've seen:

• A CryptoPunk sell for over $1m

• Christie's announce their first digital art auction and accept payment in ETH

• Insta influencer Logan Paul selling over $5m of NFTs in a day

• The sale of a Beeple artwork for $6.6m, breaking the record for a single NFT sale.

Let me repeat that, $6.6m for an original animated GIF.

That should clear up any doubt that digital art is here to stay and taking center stage. With time, we'll stop referring to it as digital art and call it, art.

The question that I continue to get asked over and over again though: why would anyone pay for something that is so easy to copy? What's the point?

In my previous post, I spoke about the fact that NFTs prove IP ownership, rarity and originality. You can make a perfect replica of the Mona Lisa or one thousand prints of it, but there will only ever be one original.

To prove this point, last week we saw this thesis playing out. A new project BinancePunks (aka Bunks) launched. BinancePunks is a direct rip-off of the now famous CryptoPunks.

When I say it's a direct rip-off, I mean its a carbon copy. 10,000 BinancePunks of 24x24 pixel sold for 1 BNB and minted to whoever purchased it. The Bunks look exactly the same as the CryptoPunks and follow the same rarity and attributes. The only difference is that it’s built on the Binance Smart Chain network instead of Ethereum.

To my surprise, the Bunks sold out in less than an hour and the rip-off project raised $2.5m. My guess is that most of these buyers are speculators hoping for a quick flip. There is now a secondary market for Bunks but it’s fetching a fraction of the amount of the original CryptoPunks. If I were to bet, I'd bet that they wont take a hold and will be a dead project in less than a year.

In the physical world, fakes exist because they're often very hard to prove that they're fake. They can look and feel identical to the original. In the NFT world though, proving the fake is as simple as looking at the smart contract address.

Nobody wants to own a fake that is easily identifiable as one. Even if that fake looks and feels exactly like the original.

CryptoPunks is that original. They will be forever considered the OG that kicked off the collectibles world. With time, their legacy and story will only appreciate in value. BinancePunks however, will forever be the fake.

So back to the question, why would anyone pay for something that is so easy to copy?

Well if the story of BinancePunks is anything to go by. It’s the ability to prove provenance, ownership and originality in the digital world. This is the value of NFTs and this is why they are here to stay.

Next week I’ll be diving into the world of NFT audio and the impact it is having in the music world. If you want to get a head start on that post, check out EulerBeats - it’s fascinating.

You’re feedback is always appreciated. If you enjoyed this post, leave a comment, hit the like button or subscribe to a Blockchain VC

Subscribe now

The current financial system is rife with high fees, delays, unequal access and barriers to innovation. In many countries, citizens don’t have access to sound money, a functional credit system, or even basic property rights. If the world economy ran on a common set of standards, that could not be manipulated by any country or company, the world would be a more fair and free place, and human progress would accelerate.

These are Brian Armstrong’s (CEO) opening remarks in the founders letter included as part of the Coinbase S-1 filing. The SEC have given Coinbase the stamp of approval to go ahead with their direct listing on the NASDAQ, giving us insights into the biggest US cryptocurrency exchange. You can read the filing here.

In the filing, Armstrong goes on to further explain how we have moved from cryptocurrencies to the cryptoeconomy.

Trading and speculation were the first major use case to take off in cryprtocurrency, just like people rushed to buy domain names in the early days of the internet. But now we’re seeing cryptocurrency evolve into something much more important. People are using cryptocurrency to earn, spend, save, stake, borrow, lend, vote and perform many other types of economic activity. Companies are getting funded, getting all customers, and eventually will be going public, all on the blockchain. The cryptoeconomy is just getting started. It is not intended to replace the traditional economy, but instead complement to it, much like email was to paper mail. The cryptoeconomy offers a more global, free, and fair alternative to traditional economies.

Those words have been carefully chosen. What Armstrong is portraying is that the cryptoeconomy isn’t just about the buying and selling of some arbitrary tokens. It isn’t about speculators trying to get rich quick. It’s about a completely new financial infrastructure that is global, open, free and fair, and being built for the internet. And, it’s still early.

Now many in the crypto world may find Armstrong’s remarks slightly disingenuous. Coinbase is a centralized exchange and in many ways goes against the ethos that many in the community subscribe to, namely: self-sovereignty, decentralization and privacy. Whereas Coinbase takes custody of your assets, is centralized and requires full KYC (Know Your Customer).

Nonetheless, as the largest cryptocurrency exchange in the US, Coinbase’s listing further legitimizes the crypto industry and will result in further acceleration of crypto adoption across the globe. I see it as bullish for bitcoin and crypto as a whole.

One area where I disagree with Armstrong, is that the cryptoeconomy is not here to complement the traditional economy. It’s here to replace it. And replace it, it will.

I wrote in a previous post about the concept of the infrastructure inversion and this is exactly what will play out with crypto. The traditional economy as we know it, will merely become a series of smart contracts running on a blockchain. Big lending departments in traditional banks will eventually cease to exist and the requirement for SWIFT as a global settlement layer will disappear. Central banks will start to mint their own digital currencies on open decentralized protocols and startups will be able to access capital from a pool of liquidity, provided by a global base of lenders, looking to earn yields on their assets. Property ownership will be fractionalized and land titles held in NFTs. The future cryptoeconomy will just be called, the economy.

There are some other interesting things in the S-1 filing worth highlighting:

• Coinbase has 43 million registered retail users of which 2.8m transact monthly

• 7,000 institutions as clients

• The firm holds $90 billion in assets and has traded $456 in total volume to date

• 96% of Coinbase’s revenue comes from transaction fees, generating a whopping $3.4b since inception

• Coinbase sees its addressable market as anyone with a smartphone (3.5b people today)

• Revenue in 2020 increased 2.5x from $480m to $1.14b

Also of note, Coinbase holds $180m of bitcoin as treasury on its balance sheet and $23.8m in ethereum. Once listed, this will make Coinbase the first Nasdaq company to have ETH on their treasury and I expect we’ll see many other firms following the MicroStrategy bitcoin playbook and follow suit adding ethereum.

Most interesting for me out of the entire prospectus was one line in the risks section:

• “We compete against a growing number of decentralized exchanges and noncustodial platforms and out business may be adversely affected if we fail to compete effectively against them.”

Coinbase are acknowledging the challenge of competing against truly open, decentralized and noncustodial platforms. I akin this to the battle between Microsoft (closed) and Linux (open).

We’re seeing this battle play out with decentralized exchanges rapidly gaining in popularity. Take decentralized exchange, Uniswap.

Uniswap’s all time volume is $102b (vs Coinbase $420b), trades $1.2b a day (vs Coinbase’s $4.2b) and has 32m active traders on the platform. The biggest difference, Uniswap started in November 2018 (vs Coinbase 2012) and is a team of just 20 people.

Uniswap’s market cap today is at $7.8b ($25b fully diluted). As of today, we’re seeing Pre-IPO market valuation of Coinbase at over $100b. Depending where you stand on the open vs closed and centralized vs decentralized debate, looking at $UNI and $SUSHI may be a very interesting bet.

I know where I’ll be placing my chips, err I mean, tokens.

Thank you for reading.

• Follow me on Twitter

• Pitch me your startup

• Subscribe to A Blockchain VC

For this week, I'm going to start with the dApp that I for one, use the most. It's the dApp that is your gateway to the crypto world. The gateway to holding and storing your assets. The gateway to exploring NFTs and DeFi.

It's also one of the oldest and most overlooked dApps within the Ethereum ecosystem. That dApp is none other than, Metamask.

The simplest way to explain Metamask is that its a browser plugin, that acts as both a wallet and a way for your to interact with the Ethereum network. Think of it as a 'bridge' between your browser and the world of decentralized applications.

It is also a non-custodial wallet. What that means is that your assets aren't transferred to a third party. Unlike storing your bitcoin or ethereum on an exchange like Coinbase.

With Metamask, your assets always remain in your control. This is because you control your private keys.

Now why is Metamask so powerful?

Without Metamask, you would need to install and run your own Ethereum node to transact with other dApps on the Ethereum network. You'd have to understand the backend of how these systems work and you'd have to get pretty technical.

I like to think of Metamask as the Netscape of the blockchain web.

With Metamask, you can 'log in' and interact with any dApp built on the ethereum blockchain.

To give you a sense of how that looks for me. I connect my Metamask account everyday, to a range of other applications, like:

• SushiSwap

• Uniswap

• Zerion

• Nifty Gateway

• CryptoPunks

• Neon District

• KeeperDao

• Pickle Finance

• Gitcoin

I'll cover many of those in the coming weeks, but for now. If you want to open up the gateway to the crypto world, start by downloading and installing Metamask.

For a how-to guide on using Metamask: visit Metamask or check out this Coingecko guide.

This week we’ve seen bitcoin price drop by 13% and people have been pinging me to see if it’s time to sell, while others have been asking if it’s time to buy. My answer is always the same, it depends on your timeline. Im not a short term crypto day trader. Rather, I look for protocols and networks that I believe are going to be longterm disruptive winners driving the infrastructure inversion that I referred to in one of my previous posts.

Bitcoin is one such winner and over time, will continue to increase in value as the narrative of bitcoin as the global money network continues to take hold. With only 21,000,000 bitcoin being produced and the difficulty of mining bitcoin increasing every four years (known as the halving), the value of bitcoin will continue to rise. Now that is not without its short term volatility and massive price swings as we have seen since the beginning of bitcoins history and will continue to do so for the foreseeable future.

That brings me to this weeks audio recommendation. Willy Woo and PlanB being interviewed on The Bitcoin Fundamentals Podcast.

Both Willy and PlanB share their views on the current bitcoin bull market, how long they believe it will continue and the elements that they look for to predict when we’ll see a top or floor. If you haven’t heard of either them, I recommend you give them a follow on Twitter.

In the interview, PlanB goes into further detail on his S2F (stock to flow) model that he released in March 2019 (when bitcoin was at around $4,000). In that model, he predicted that bitcoin would reach $55,000 some time after the bitcoin halving in May 2020 and that bitcoin as an asset would reach $1T in market cap.

What were the elements that would drive that outcome, namely: A worldwide event that could result in hyperinflation, excessive money printing, and/or large institutions buying bitcoin.

It’s been surprisingly accurate at tracking the bitcoin price movement (see chart below) so its worth understanding a bit more about it.

What is the Stock-to-Flow Model?

Without getting too technical, the S2F model takes the number of bitcoin in reserves and divides that by the amount produced yearly. In essence, it gives us the number of years that it will take (at the current production rate), to achieve the current stock of bitcoin. The higher the ratio, the more scarce the asset and the better it is as a predictor of upward price movement.

Currently, bitcoin S2F is lower than gold (55 for bitcoin vs 59 for gold), but with each halving that number will go up ultimately replacing gold in its scarcity. And, with every halving, the S2F doubles and shows a 10x increase in market value. A trend that has held true for the previous 3 halvings.

So where does the S2F model put us today?

The model predicts that we should see bitcoin reach over $100k in price before the next halving.

PlanB has gone further to revise the model by removing the bitcoin that is believed to be lost and no longer in circulation. Instead of 21,000,000 bitcoin, the model uses 19,000,000 rather.

Under that new model (S2FX), the S2F number for bitcoin increases to 56 and he predicts a market value of the next BTC phase between now and 2024 of $5.5T. This translates into a BTC price (given 19M BTC in 2020–2024) of $288K.

If he’s right, there’s still plenty of room for bitcoin to go from here and each dip should be seen as a buying opportunity.

There are other indicators that Willy Woo unpacks and talks about in terms of on-chain metrics that he is seeing and why he is so bullish on bitcoin long term as well.

It’s an interesting listen. I hope you enjoy it.

If you enjoyed this post, subscribe or give it a like!

Subscribe now

In September 2019 I wrote a blog post, Those that know, don’t predict, and in it I wrote my Top 4 predictions for crypto by the end of 2020.

They were:

1. Bitcoin would see its market dominance drop to about 50% (got this wrong, it’s actually at 61%)

2. Ethereum would hold strong in the number two position by market cap (held true)

3. Binance would move from eighth to third (true, off by 2 months)

4. An NFT would sell for over $1m (this happened yesterday)

Number four is the most interesting to me.
Yesterday, we saw CrytoPunk 6487 sell for $1,054,240 or 550 ETH.

What’s even more interesting is that this isn’t the first CryptoPunk to sell for over $1m. Just three days earlier, CryptoPunk 6965 sold for a cool $2m, and the day before that CryptoPunk 4156 sold for $1m, making it, the first CryptoPunk to breach the $1m mark for the rare NFT.

You may be thinking, WTAF? I can just take a screenshot of this image or copy it and I have it. Why would anyone pay over $1m for this?!

It’s a good question and goes back to the core thesis that blockchain and decentralization enables: assets in the digital world can now become rare, provably scarce, unique and own-able. The technology acronym behind these are NFTs (non-fungible tokens, you can read about them here).

Your next logical question would be.. OK, well what makes these things valuable? How do you determine which of them will become the most valuable? Won’t there just be millions more of these coming onto the market? Is it just a function of scarcity? Or is there something more to it?

My belief: The ones with the greatest story behind it will ultimately win.

Why stories?

Yuval Harrari (in his book, Sapiens) explains:

Humans think in stories rather than in facts, numbers, or equations, and the simpler the story, the better.

CryptoPunks started with a simple story.

Creating crypto art that plays a tribute to the crypto movement.

It asked: What if there was a way to create digital art that played tribute to cypherpunk’s who paved the way for bitcoin? What if they could be made provably scarce on a blockchain? What if they were mathematically generated and no two are alike?

And so, it kicked off its first story.

CryptoPunks is the story of the tribute to the cypherpunk movement.

To enable that story, 10,000 CryptoPunks were distributed (for free) to anyone with an Ethereum address and that would claim them. From there, people within the crypto movement started to build on that story by sharing it with others and with time, that story began to embed within it, it’s own culture and the myths that come along with it.

Today, if you own a CryptoPunk then you are buying into that story. You are buying into the story of cryptography, you are buying into the story of bitcoin and blockchains, you are buying into the story of the cyperphunks, and you are buying into the story of digital art. You can’t own a CryptoPunk without believing in that story and once you do, you are going to want to tell it. Over time, as more and more people believe in that story, it becomes a myth.

Yuval Harrari says it again:

Large numbers of strangers can cooperate successfully by believing in common myths. Any large-scale human cooperation – whether a modern state, a medieval church, an ancient city or an archaic tribe – is rooted in common myths that exist only in people’s collective imagination.

But what is different about today is our ability to accelerate that story-to-myth journey. That journey from an imagined idea, to a story that can be shared and repeated, happens now on a cycle faster than we’ve ever seen before.

In the past, people needed to interact with one another to share their stories. They had to physically converse and go into details about the story to hear it and understand it.. But today, with Facebook, Twitter, Whatsapp, TikTok and other social networks, those stories have been broken down into small bite size chunks or memes and are able to spread globally as fast as communication on the internet. Once those stories take hold, they become self fulfilling prophecies and both literally, and figuratively, get memed into existence.

That, is how we’ve been able to see something like CryptoPunks go from an imagined idea, to myth, to meme, and to a 24x24 pixel sell for over $1m (in just over 3 years).

Punks United.

If you enjoyed reading this post, subscribe to A Blockchain VC today.

Subscribe now