Read_949 - The Singularity Stack

0:00 The cost of intelligence keeps falling toward the cost of electricity, and everyone will share in what that intelligence can do. Bitcoin captures the gains directly, while in dollar terms, they are inflated away. Welcome back to the show. I'm Guy Swan, the guy who's read more about Bitcoin than anybody else you know.

0:35 And we got a really good one today. We're going to do a read. I know it's been a really long time since we've done a read. And I want to talk for just a minute about why and explain what's been going on behind the scenes. So we've done a huge shift over towards video, primarily video content. And because of that, we've had to change our production workflow.

0:57 So I need to make sure that we could automate this and we could get things out quickly. I don't mean automate like we just plug AI into everything. I just mean that the process flows. And I'm not constantly reinventing what we're doing every single episode. And there's been a lot of that with AI and the tools available to us.

1:16 And of course the process will still change quite regularly. But I think we do have a good system in place. And then in addition, the reason I've never done a read in video is because I never had the setup that I felt made it make sense. I could never do it quickly. And if I was going to do video for a read, the reason I really love the reads is because I can knock them out really quick.

1:40 They were high value, high signal. And I just enjoyed sitting down in front of the mic. And when I was done recording the narration or the episode itself, just in the raw recording, the whole edit was done. I had set up my production flow that it was just when I got to the end, I exported and I didn't have to do any post-production.

2:00 Now this isn't perfect and this isn't entirely post-production free, but I have actually vibe coded my own entire application that does this so that I can have the same style or same production flow in video as I did in the audio only podcast. So this will be my first read in video, my first video read. Maybe this is a totally pointless thing and there's no reason to publish audio reads in video, but it's fun. I think, I don't know, I think I can play around with it more to actually be able to see the thing on screen while I'm reading through it. I got some fun little AI tricks and stuff that I'm sorting out the processes for right now that, I don't know, this could be this could be i think it's gonna be fun so i'm really curious what you guys think

3:03 if you sit down and actually watch a video read so with that shout out to bitbox they are our sponsor and they've been a fantastic wallet for a really long time i've used them for i don't know how long and you know uh sweating with the great hacking and going on ledger getting hit and cold card getting hit and all this stuff i've been very very happy with the team and the findings that they've had and how they've reported them and how they've fixed them. It's just been, the BitBox has been very trustworthy throughout all of this. And so that's made me feel really good, especially since I recommend it to people and I've used it for a really, really long time. And now it works natively right into the iPhone. So check it out. You got a 5% discount and, you know, stay safe out

3:50 there. Buy directly from the manufacturer, please. And if you can use multi-sick, BitBox is now It plugs right in a nunchuck and works very easily with Multisig now, and I can't recommend it enough. It's saved my ass multiple times this year. So with that, it's time to get into today's read, and it's titled, The Singularity Stack from EarlyWriters.com The stack traces the path from raw energy to useful output, and it measures where value is created and where it leaks at each layer.

4:33 Part 1. The Premise Thinking has become a manufactured good. In 1830, an hour of decent reading light cost the average worker about three hours of labor. So in 1830, three hours of labor got you an hour of light. In 1900, that same hour of light only cost 13 minutes of labor. And today an hour of light will cost you 0 seconds of labor which is too cheap to even notice For centuries illumination was a craft good, rendered from tallow by hand and consumed within hours of its making. Electrification made it a utility, generated in bulk and metered by the kilowatt hour, and the industries that were unthinkable in the candle era arrived as ordinary byproducts. The power grid accidentally led to the night shift, the modern hospital, and massive improvements in transportation and

5:45 coordination 24-7. Intelligence is undergoing a similar transition. For all of prior history, a unit of useful thinking cost a unit of human life, with the limiting factor as a brain that runs on a metabolic budget of roughly 20 watts. Artificial intelligence completes the sequence because the thinking itself has become a manufactured good, produced in bulk and priced increasingly in energy.

6:20 And unlike light, intelligence improves the process that produces it, so every fall in its price reflexively drives the cost down, all else being equal. electricity goes to compute to intelligence to work and finally settlement each layer is the physical input to the one above it so a constraint anywhere reprices everything downstream and an improvement anywhere compounds through the rest light became too cheap to meter so they got the work time needed to buy an hour of reading light which if you're watching this you can see it from 1830 to 2026 on log scale, which is hilarious because this is a steeper than linear decline in log scale. This is at earlywriters.com slash charts if you're listening to this.

7:12 Light became too cheap to meter. Intelligence is next. That's a crazy way to think about it. Okay. Technological progress means doing more with less, and money is where the difference is saved. When the money supply expands, every gain in efficiency is met by new units bidding prices back up, and the useful work the energy performed is dissipated before anyone can hold it.

7:36 Money with a fixed supply preserves the conversion, so efficiency arrives the way cheap light did, as rising purchasing power. We believe the gains of the singularity settle in money that costs energy to produce and cannot be expanded to absorb them, and this report is the argument for why. Part 2. The Framework. The Stack Defined. The stack is a single conversion chain.

8:08 Electricity becomes compute, compute becomes intelligence, intelligence becomes work, and work becomes payments that need somewhere final to land. Each layer is the physical input to the one above it. So a constraint anywhere reprices everything downstream and an improvement anywhere compounds through the rest. Layer 1. Energy. Everything else in the stack, from tokens generated to transactions settled, resolves to electricity. The raw energy is there in abundance. The constraint is converting it and delivering it to where the demand sits.

8:46 Over 2,000 gigawatts of proposed generation waits in U.S. interconnection queues, but delays and surprise upgrade costs mean most of it will be abandoned. Intelligence applied to conversion in better solar, storage, and reactors makes electricity cheaper, and cheaper electricity makes intelligence cheaper in turn, the first step in the singularity.

9:14 Layer 2. Compute The compute behind Frontier training runs has doubled roughly every six months, a compounding rate no prior buildout has approached. Capability follows compute, with the length of tasks AI can complete doubling every four to seven months. The acceleration is reflexive. This July, a Frontier model designed a working chip in a 48-hour run with no human intervention, a chip built to run a smaller version of the model that designed it.

9:49 Layer 3. Models That same compounding destroys the business case, since every frontier model becomes the cheap commodity option within a year and open-weight releases set the floor near zero. Each generation now helps build the next, generating the training data, writing the training code, and teaching the smaller models that follow.

10:11 Capability compounds on two axes With compute scaling below And algorithmic progress Cutting the price of a given intelligence level By roughly 10x a year If that actually accurate that insane The labs sit closer to the memory manufacturers than to the software companies they are valued like Layer 4. Harnesses. So the switching cost in AI lives a layer above the model, and it deepens with use. The code itself is easy to copy, with open source implementation setting the floor price near zero. So the durable value sits in the operating context a harness accumulates. The memory, tools, credentials, evaluations, approvals, budgets, routing, and audit trails that build up around real work. The effect is measurable, as identical weights that solved 2% of real software

11:09 bugs in a chat window solved six times as many inside an early harness that could read files and run tests. Models underneath get swapped constantly on quality, cost, latency, and policy, and the harness is what makes those swaps routine, since nobody rebuilds that context every time. Layer 5. Agents. Work priced in tokens is work priced in electricity, and every layer below this one is driving that price towards zero. Labor stops scaling with people at this layer, since one person can direct a dozen agents and an agent can spawn more when the task splits.

11:54 The work turns self-directed as well because agents now build and test the layers beneath them. Layer 6. Identity. So identity inverts from something granted from above to something the agent generates itself. On Noster, that is a key pair, a private key created in milliseconds with nobody's permission and a public key anyone can verify. Company-issued accounts fail here twice.

12:23 Since agents multiplying by the billions overwhelm any issuer and whatever an issuer grants, it can revoke, erasing that graph along with the balance on one terms of service update. The identity lives in the signatures on a decentralized network rather than on any server, so an agent banned from one relay reposts the same signed record elsewhere and loses nothing.

12:49 Level 7. Coordination The model context protocol has become the default way to expose a system to a model, an open specification with no gatekeeper metering what flows through it. The agent-to-agent specification lets agents built by different vendors delegate to one another, and delegation changes what capability means, since an agent that can hire specialists outperforms any single model working alone.

13:15 The shared workspace remains the unsolved third, since in Slack and Discord an agent is a bot account subject to the revocation problem above, while Buzz, an open-source project from Block, makes humans and agents peers who each hold their own keys. Layer 8. Value Transfer Money has to move at the speed of the work it pays for.

13:40 Intent of a cent amounts at the frequency of API calls with finality in seconds because an agent cannot wait on a dispute process. The card networks were built for none of this, since a fixed fee per authorization makes a micropayment cost more to process than the thing even being bought, and settlement that takes days cannot serve work that finishes in seconds. What replaces them splits into rails, which move the money, and standards, which tell a machine how to ask for it. Neither specification collects a toll, so the durable income goes to whoever issues the money the payments are made in.

14:19 Layer 9. Settlement. The singularity accelerates the build-out of everything except the money at the end of the stack. Every other layer's scarcity gets competed away eventually, but Bitcoin's supply stays fixed no matter how much demand shows up. So as the stack makes everything cheaper, whoever holds the fixed money watches their purchasing power rise.

14:50 The frame is the analytical tool, Placing an event on its layer tells you how quickly supply can respond and where the value it creates will settle. A power contract and an agent payment standard sit seven layers apart but move the same variable, the cost of finished work, from opposite ends of the chain.

15:15 Part 3. The Evidence The acceleration is measurable. The singularity is best defined operationally, as a period in which the rate of improvement accelerates across enough independent domains that no single reading can be dismissed as hype Compute 4 Frontier Training Growth Compute Epic AI estimates that the computation used to train frontier AI models has been growing four to five times per year, a doubling roughly every six months. Moore's Law, the benchmark for relentless compounding across five decades, doubled every two years, and this build-out runs four times faster. Adoption. Two months. ChatGPT to 100 million users. The telephone took roughly 75 years to reach 100 million users. ChatGPT reached 100 million monthly users in about two months.

16:15 Even that record should fall because the next wave of products will be adopted by agents as well as people, and agents adopt valuable tools as fast as they can read the specification. Capital. 713 billion. Hyperscaler CapEx planned for 2026. The entire 13-year Apollo program cost about $309 billion in today's dollars.

16:42 The distinction from prior manias is the funding source. Since the late 1990s, fiber build-out was financed with debt by carriers that never earned it back, while this one is paid for out of the operating profits of the most profitable companies in history. Energy. Four-year U.S. electricity demand forecast.

17:03 After two decades of essentially flat grid consumption, the EIA now forecasts the strongest four-year growth in U.S. electricity demand since 2000, naming data centers alongside manufacturing on shoring and electrification as the drivers, A demand line that stayed flat through the entire smartphone and cloud era has bent, and it is not waiting for the grid.

17:29 Capability. 12-hour Metaer autonomous task horizon. Metaer measures the length of tasks AI systems complete autonomously at a 50% success rate. In 2019, the best model handled a task taking a human about two seconds. The doubling time has compressed from every seven months to roughly every four. And even at the stricter threshold, the trend puts a full working week of unsupervised output inside a decade-long earnings forecast.

18:01 One year of AI CapEx buys two Apollo programs. U.S. hyperscaler capital expenditure against the Apollo program inflation-adjusted. One year of planned AI CapEx buys more than two entire Apollo programs in today's dollars. Part four, the money. Deflation is the normal state. Productivity is exploding, and life does not feel cheaper for monetary reasons and not technological ones.

18:36 In the last three decades of the 19th century, under gold-denominated money, the American price level drifted steadily lower while real output per person grew at a rate comparable to the strongest post-war decades. Mild deflation set a higher hurdle rate instead of paralyzing capital, and the country still laid track and built steel at a formidable pace.

19:00 Deflation is therefore the normal state of technological progress under a money nobody can expand at will. The modern reflex against it is institutional and not economic, since the Federal Reserve adopted its explicit 2% target only in January 2012. Nearly all growth arrived after 1800. World GDP from 1700 to 2015 in 2011 international dollars.

19:28 The Debt Arithmetic The System That Cannot Allow Deflation In a debt-based monetary order, falling prices are intolerable as a matter of arithmetic. A loan is written in fixed nominal dollars, while the borrower's revenue is whatever prices the market will bear. So when technological deflation pushes down prices, the income shrinks and the principal does not.

19:56 Irving Fisher set out the spiral in 1933, where borrowers default, banks pull credit, spending falls, and prices drop further. The sequence that failed roughly 9,000 U.S. banks and contracted the money stock by a third between 1930 and 1933. Every unit of technological deflation therefore has to be met with offsetting monetary expansion, and the singularity turns that from a maintenance task into a race.

20:25 $4.7 trillion is the U.S. M2 money supply in January of 2000. $22 trillion is the U.S. M2 money supply today. $40 trillion is the federal debt. Interest above $1 trillion. The denominator keeps expanding. So there's a chart of the U.S. M2 money supply from 2000 to 2026. BLS Component Data, 2002 Today The Two Economies BLS Component Data shows the two economies side by side.

21:02 Everything manufactured and shippable, from televisions to software, has fallen steeply in price since 2000, while hospital services, tuition, childcare, and housing have outrun wages. Technology drives deflation in the reproducible half of life, while monetary expansion lands in the scarce half. And the stack is aimed at moving the scarce half, priced mostly in trained human time, into the reproducible column.

21:32 That's wild when you say it that way, but it's totally true. The Collision The deflationary force now arriving is aimed directly at labor, since agents convert capability into completed work that used to require a salary and trained human time is the dominant cost in everything that has grown more expensive for decades.

21:54 The monetary system has to answer with expansion large enough to keep $40 trillion of federal debt serviceable, stretching the unit of account to absorb an exponential. most households whose wealth sits in wages and deposits risk living through the largest productivity boom on record as a cost of living squeeze that is so depressing to say out loud but that is literally what what we're looking at with the structure of what's coming because they don't and it's all because of the money it's all freaking god it's everything that the whole thing is always because of the damn money. Part 5. The Position. The purest expression of the singularity. The singularity does not float free of the physical world because a model cannot

22:44 immediately generate the electricity it consumes. Well, the machine still has to buy. Somebody has to sign a 20-year power purchase agreement promising megawatt hours in 2046 at a price written in 2026. The faster the machine improves, the more its future hangs on the slowest instruments in the economy, contracts that only work if the money holds.

23:09 Under expandable money, being close to the newly created supply wins, which is how a company burning cash outbids a profitable manufacturer for power. Under fixed money, only output that customers actually pay for wins. Participation. There is no direct way to own the singularity, but every layer of the stack can be owned somewhere, and there is money to be made in each.

23:39 The difficulty is that every one of those positions adds a way to be right about the acceleration and wrong about the outcome, since a layer bet has to name the winner before competition arrives. In our view, Bitcoin is the purest expression of the singularity stack, in the position with the fewest ways to lose.

23:57 Every conversion from energy to finished work settles in money at the end, so holding the settlement asset claims a share without naming the layer or the company that books the margin. It carries no earnings to miss and no shares to dilute, and no competitor can ship a better version of a fixed supply.

24:17 On-chain estimates put roughly half of Bitcoin's supply dormant for more than a year, so rising demand competes for the fraction actually available. Who keeps the gains? Intelligence priced near the electricity behind it means expert help in medicine, law, engineering, and education for roughly the cost of light.

24:40 And the surplus reaches people who never touch a model. The industrial era already ran this experiment under a money nobody could print. An ordinary saver who simply held gold-backed money watched railroads and cheap steel arrive as rising purchasing power. That transition is running again, at the singularity's pace.

25:03 The cost of intelligence keeps falling toward the cost of electricity, and everyone will share in what that intelligence can do. Bitcoin captures the gains directly, while in dollar terms, they are inflated away. Differentiated business at any layer, one whose advantage survives the arrival of new supply, keeps its margin through the transition.

25:28 Larger businesses can now be built with less capital and fewer people than at any point in history, since labor that once required headcount is provisioned by the token. So the returns to a real edge have never been higher. Early riders invest in both. Are you accelerating the singularity? We want to hear from you.

25:49 Enter your name and email and address and a message. Hey, guys. Oops. I'm going to send a message. Guy Thank you We will be in touch I want to see if they respond something uh so okay so uh the one of the the things talking about like who keeps the gains i really like this section where is it in our view bitcoin is a purest expression of the singularity stack so here's the funny thing so this is i've talked about this in the show on like the idea of money and what it actually what its role ends up being in the economy. And we talk about it from a personal use case is money ends up like being a hedge against uncertainty, right? You invest in it or you hold on to it.

26:37 And it's kind of like this promise of society to get back what you put in it. So the less uncertain you are, anything that you concretize into any money that you make concrete into some specific thing, you increase the uncertainty of whether or not it can actually provide what you want. You know, you put it into a house and a hurricane might come through, a flood might come through and destroy the house. But if you put it in money, it can be in a boat as easily as it can be in a house when you need it, of course, barring society collapsing. So money is not only a hedge against uncertainty, but this is another thing that we've talked about is how it plays out as a kind of investment in everything, right? So when you're actually investing in the market,

27:22 and you're trying to pick winners and losers, or you're trying to invest in the singularity, because, you know, you're like, oh, Nvidia is the one that's going to explode. But you don't know which thing's actually going to take the market share or which is going to be obsoleted, you know, explode to 200 million users and then be obsoleted in two months, because things are going to be moving so fast. You have an insanely high margin of uncertainty in making any of those sorts of decisions on top of the fact that you have management risk and you have responsibility risk because someone could just kind of go crazy or make stupid decisions.

27:59 Like it's not even simply that they're placed in the right location or that they're primed for, you know, the various growth strategies or the production. Something could just go wrong. Somebody could trip on something and smash a button and delete the whole thing, right? So the risk, you introduce enormous amounts of counterparty risk.

28:18 That's what you do. Whereas Bitcoin, when you have a sound money, you do obviously have to trust the money's capability of ensuring its scarcity. But as soon as you can get past that, as soon as you have some assurance or level of trust in the scarcity of the unit that's actually measuring that progress, you actually have a non-counterparty risk way to actually obtain the hurdle rate. So the hurdle rate, you know, the cost of capital over time will increase dramatically if there's an enormous amount of growth expected. And the very reason is, is because you don't want to invest. You don't want to waste resources. Everything's scarce, right? You have 10 houses. You don't want nine of those houses taken up by a bunch of stuff that's making a half percent improvement or a 1% loss on

29:06 productivity or manufacturing if that last house or that the one out of the 10 is actually producing a doubling in efficiency and productivity gains in that same economy. You want all 10 of the houses to be put toward that and it better be expensive as hell. It better cost 20 percent, 30 percent to actually rent out or lease out one of those houses because it's otherwise losing or costing us a 100% gain in productivity and boosting of a standard of living. So that's why hurdle rate, that's why your interest rate is going to skyrocket during a period of immense growth.

29:43 And it's exactly why deflation is going to have to come in massive or I guess inflation, actually the opposite of that, because it would produce enormous amounts of deflation because you have such enormous growth and productivity when, when the reverse has to be true in the money. If you're on, if you're dead, if you're $40 trillion in debt, you sure as hell better print the ever loving crap out of it because you'll never pay back $40 trillion. When everything's plummeting to like prices are plummeting to the cost of electricity for practically everything that we do. There's absolutely no way out of that except inflating yourself into, into nothing. If you inflate away your debt, your obligation and your interest rates, you know, if you're legitimately looking at the possibility of a 20%, 30% interest rate,

30:28 right um and i don't even think that's crazy if if anything resembling this this whole idea of this reflexive feedback loop which i'm sitting here working on an application that i've vibe coded myself and i've been improving and i've got like my debug when i like have a problem with this i literally have a little debug note thing up in the top right in fact let me let me show it um Can I, let me just pause.

30:57 Alright, so this looks, this looks bonkers, maybe I can cut my screen off. Okay, here we go. Well, it still does the interface over and over again. So this is the application that I built and so like the screen share is just on screen and I can switch over to you know sharing like a window or something taking off my camera did nothing But I can this is a full on like editing app And I built all this stuff And when I like the process the changing of the process So I found how frustrating and annoying it was to constantly go over or talk about the problems, like type out the problems that I have into AI, right?

31:44 So, and talking about like the harness in the singularity stack, actually, this refers right back to this, is I have agents basically on all my machines now. So check this out. So this is a debug node. And what I'm doing right now is this thing is recording logs of where my cursor is and where I am clicking on stuff so that I can keep track.

32:07 Or basically the AI knows exactly what I'm doing and what is happening in the actual application. And by the way, for the AI, this is a transcript. I'm recording this into an episode, but I'm actually making an accurate note here. So when I try to add a source right here, it just blinks off. It's not actually working.

32:29 I would like to be able to add a second source. It's basically trying to take one screen share as I have to replace it. So I have to close it right here and open up a new one. Whereas what I want to do is I want to be able to add a new source and click on a new window or a new device and then have this show up in my double because I've got the, it doesn't come up right now just because I don't have anything in it.

32:56 But so that I can show them both on screen at the exact same time, right? As I can show my window for my browser and then another window for the read or something, or of course, Pearcast, the app itself, the studio that I'm working in. And, but right now, for whatever reason, this does not actually react that way.

33:17 So I want to be able to stack these and have two of them actually working or running at the exact same time. So that'll close it out. So I stop there. That's my debug note. And it literally takes the log of everything that I've clicked on. And we've been building a peer-to-peer engine for just any general networking and operating use case.

33:43 so what actually happens is that that note is recorded in the background and then is saved to a pair drive space or just is saved to a space in our peer-to-peer engine and then syncs to my other machines and to my agent and my agent immediately gets a notification that there's a new thing and the agent goes ahead and just immediately starts fixing it and so I can just kind of like think about and every single time I have one little issue or concern or something where I like clicked on something and it didn't behave like I wanted to behave again I don't have to go out and text and like do this very minute explanation of like I'm in the sidebar on the right side and it's in the the button is the thing and it's not right like the agent can see

34:30 exactly what's happening and it has logs of exactly where I clicked I basically have immensely, like extremely quick iteration. And my favorite thing is that because I've got it plugged into Pear Core, the Pear Drive, there's too many naming conventions, but the peer-to-peer engine that we have in the background is that I'm hosting all of this myself and it will sync with all my other devices. I can plug in effects and elements and stuff straight from my Linux machine so that I can drop them right into this timeline and it will just automatically download them in the background and then display them and all that networking is free and it connects right to my machine I can do it land like here at home or away and this has been you know three years in the

35:15 making for all this stuff the engine took an enormous amount of time because we had to go back to the drawing board like five different times but I've just been vibe coding like crazy I've just been building everything that I can possibly think of that I want to have in my workflow but my ability to produce the reason I am doing this read is because I am making my entire process and my entire workflow customized around the way that I work and the way that I think.

35:40 And it'll be really cool. I might share this out. Maybe other people even want this app. It's super cool. My favorite thing about it is that I can edit and cut things out on the fly. I can click into this timeline immediately and look at my, you know, waveform, which I don't have the right now it will transcribe in the background, but I don't have the edit by text yet.

35:59 That's something I'll be adding in, but I can click on it and then just edit. I can cut something and then put my playhead right where I want to record again and then start recording right back into it. And then because I've got a peer-to-peer engine, I'm going to get, right now it's just a remote control app, but I'm getting an app, basically the mobile version of this, so that I can pick up my mobile phone.

36:20 I can stop here pick up my mobile phone and hold it up or go outside or something anywhere I don have to be on land I don have to be in the same network at home uh and record into the exact same timeline It awesome And this is making me so much more productive already Like I know that this building out these things in the workflow is going to have a feedback loop in my own system, my own like production of content. Uh, and at the exact same time, my production of the content using like the debug notes, that's the thing is that everything that we're building is going to be pair core enabled so that, you know, everything talks the same language, you know, it's kind of like the web, except that it's just peer to peer. So it doesn't, doesn't really matter

37:08 what or how it's just establishing connections and then sharing information. And we have like kind of a schema and a base protocol for doing it in all the various ways you want to share information. And pair drive, the one that I'm sure you've heard me talk about is the one that's basically just the files, like thousands to hundreds of thousands of files. I have, I don't know, 300, 400,000 files or something on my Linux machine that is shared with my Mac right now, which I think I actually closed it because I was, I was working on something, but I have the same debug notes idea inside of pair drive syncing over pair drive to my agents to fix problems and sort out when things aren't working properly and it's just everything has sped up so much

37:56 and then the same thing is the reason i have multiple agents on multiple machines is because since they can all read transcripts for the exact same and and i can also sync my wikis like my memory, my harness data and saved connections and memory and project files and all that stuff. I can sync those between devices as well so that all of my agents basically have, they can, you know, kind of have their own separate isolated context, but then they can also get all of the project context together. They can all have the same context about the project. But the really cool thing is that like trying to test peer to peer stuff or networking stuff in a closed environment, like on this computer, you inevitably just you're missing so many, so much of the nuance

38:41 of like the networking. Of course, you can find another fake computer inside the computer that you're on, right? You don't have a networking problem from your computer to your own computer. And having agents on different machines, which I built a little chat app again on top of Paracore does the exact same thing.

38:58 It just syncs messages, which are basically so no-ster-signed events. That's the really crazy thing is that I'm doing exactly what they're saying in this thing because it's the easy way to do it. Paracore is just a set of signed keys and I have a Noster ID on top of it that basically is the master for the account.

39:21 So you sign in just by saying this public key is also me in the group and then you sign out to remove and update that list. and because of that though is that my agents just generated their own keys and then I join them into I add them to the conversation and then they can chat back and forth and so it's kind of like buzz it's very much like buzz except for the fact that buzz did not have I actually stole a couple of things from buzz very recently because I forgot that it even existed until I read this piece the first time and I was like that's right I forgot about buzz and so we took some of the examples of things that are helpful inside of Buzz and applied them to the little random chat thing that I built,

40:03 which I might release it. I don't really care. I'm not trying to build a chat app. I just, I needed it from my agents. And obviously the clear path to actually having it work well was to have them generate their own keys. And the agents literally have their own keys and we have our own conversation stack and I just have like budgets. So like I can set, you know, you guys go for an hour or two or use this many tokens before you stop.

40:32 My main way that I use it is rounds, right? So if Kimmy on my Mac Mini is talking to Kimmy on my Linux machine, they're allowed to go three rounds back and forth until I interject or respond. Or I can bump that and be like add 10 rounds to this conversation so they can go back and forth and they can test things out and then send reports.

40:51 And Kimmy on my Mac Mini and my Linux are basically red teamers for Claude. So I have Fable on my, or I'm running Fable in my MacBook Pro. And so all three of these computers can work in a live, like a real networking environment and directly share and sync logs over problems or reasons why streaming a file took too long or why there's a three second delay.

41:18 Why did the previous stream still send four chunks after the user requested for the next stream? like those sorts of things. And I can just have them. I can let literally let them go. I've done it a number of times now where I just say like, go for three hours and find as many bugs and edge cases and stuff as you can.

41:37 And I just let them run overnight. a burn through 50 bucks of tokens on Kimmy in PPQ. I plug both of them into PPQ.ai. Shout out to Matt for all this. But like this is such a massive acceleration and it's huge to, it's such a benefit to the whole team because I'm working with real developers too who are basically back and forth with these things to do the careful cleanup and implementation of this stuff. And it's just, it's wild. It is moving so fast. Some of the things that I've done in the last few days, not possible three months ago, not possible at all. And it has accelerated everything. The fact that I have this engine now, I can build peer-to-peer networking apps that I can put on my phone and make iOS builds up and Android builds up extremely quickly

42:35 and start playing around with like my chat, my chat app, which is again, nothing fancy. I'm not trying to build a chat app, but I was able to make an iOS build for it for my agents so that I can, I can just sit here and I can chat with my agents. And this is like a report that Claude had on the last, I was working on like icons and stuff like that.

42:59 And they can reach out, they have a, you know, API with PBQ or whatever. so they can actually generate icons and play around with stuff. The Opus 5.5 is actually really good with that, just like building an SVG file. If you want to do an action, this is actually a note for anybody who's listening to this.

43:15 If you want to do an animation or a variation on an icon or something, don't use an image generation model. Use a LLM. Use Opus 5.5 or something like that, or even Kimmy, because it will simply look for tools to edit and do exact things. And there are even things where it could use, like maybe an image model is necessary to just iterate on like general icon ideas, but then it can tweak them, literally edit things inside an app in like kind of a Photoshop environment.

43:53 And it's wild. It's super wild. And like once you have a platform, Like I just can't get over how useful the engine that we have created is. Because one of the things I still, again, lots of stuff to work out in regards to the interaction and like how I deal with adding people and collaboration into this.

44:20 But that's basically where I'm going to. I want to, Johnny will be listening to this. My producer will be listening to this. and I want to have it so that I can invite, like right now I can, but the collaborative editing is just not, it's not right. I haven't worked out all the kinks yet, but it is just like a, it's basically like a Git history in a sense on the timeline that syncs between various people.

44:52 So what I literally want to, I don't think it will be as soon as I get into tackling this specifically. I will be able to invite Johnny in, and I want to be able to propose and make edits on the timeline, both of us working in the timeline at the exact same time. And there's plenty of capability that this app won't do, and maybe some of it will be added.

45:17 I don't think it needs to do everything because it just needs to do enough to make a quick production line of content. And I can just continually expand this by thinking about and just talking about the things that I wanted to do. Like that is what's wild about this. So the debug notes is my vibe coding is now just me dump like brain dumping ideas out or thoughts out into all the apps that I'm doing.

45:46 I don't type anything or explain what I'm doing or where I am in the interface for any of the stuff I'm building anymore because of this. And I don't have to have a platform or a subscription. This is just peer-to-peer. It just automatically syncs between all my devices. And I can build anything. This is so freaking crazy.

46:08 It's the most addictive thing in the world. I am literally addicted to vibe coding. I had to have a serious heart-to-heart because I hadn't produced anything for my show in a while. because I've been vibe coding shit. It's so, if you are like a creator, if you're somebody who just has ideas and you just always want to see or test stuff out, this is like the greatest and worst possible time because one of the most difficult things when you are a creator and you want to build stuff all the time is deciding what not to build.

46:37 It's like, okay, I can build, I'll have a hundred things by the end of one day that I really, really want to build, but I can't possibly build all those things. So I do a good job of trying to be like okay I only going to build this one thing And then I do that for you know four months And then I have a hundred things that I trying to build And I have spread myself too thin and everything a disaster And I have to cut 90 of them out of my life, right?

47:01 And, you know, narrow it back down to like three or four things. And I am in this constant cycle of like taking on too much crap, cutting it down to the stuff that really matters or the things that I'm like super excited about. but what's crazy is that like I'm so I can still be so productive and I can keep such measurable like concrete progress on so many of these projects like you can't you god please this you can't see this right now but Pericast is just open a new iteration and my agent oh my god I got to figure out how to do this hold on a second hold on a second I got to show you I'm gonna do this live so that you can see the whole entire screen this is my agent this is my agent right now

47:41 testing the interface and building the solution to the thing I just talked about and made a memo of five minutes ago. And it's basically just running through everything while I am sitting here. This app is going to be better before I finish this episode. This is going to be applied to everything. Like, it's so crazy to see where we're going. And, you know, we talk about, like, direct feedback in an application. Like, I want to be able to have, like, my test guys, whoever's, like, doing this, to be able to, like, make debug notes. You know, to have, like, approved people with debugging notes so that other people using this can be like, why did I clicked on this thing And it came up over on the left side. Why would it come up on the left side? Like, I want people to just share when they're confused and see if I can get a few agents and myself to brainstorm, like do brainstorming sessions on what, you know, the lowest common denominator of confusion between various people is to make things happen very, very quickly and react intuitively to what the user's expecting.

48:59 but it's just it's so wild we live in seriously wild times I see the compounding happening just in my stack just in the things that I am working on and it's hard to fathom how quickly that is going to fan out to the rest of the economy and to see how quickly the new models I'm using I'm using are able to functionally perform at a far greater depth and length of time.

49:40 Basically depth in both ways, depth in time and depth in understanding or operational competence for the entire stack of the computer. Like I really think we're going to be able to take the application like pair drive and pair drop the things that we're building and like convert them into another language in like a few goes um and granted with the hyper core stack and some of the peer-to-peer stuff we're using uh i don't think we'll be changing the language a whole lot but there will be things that we can make far more efficient and builds that we can improve upon and it just it just comes to you know, hearing about stuff like that's the crazy thing about this. So when I find out about a new thing, like I found out about Graphify and a few things the other day, like I can literally just

50:34 take the GitHub repo of a Twitter post of something that just sounds crazy. Give it to my agent, get it to read through in the context. Buzz. Oh, my God. Buzz is the perfect example. So I was reading this article from early writers and they talked, they mentioned Buzz and I was like, I forgot about Buzz. And I immediately took it, grabbed the GitHub, gave it to my agents who were in the conversation of the chat of the tool we built. It's like Buzz like and got it to go through and make a comparison, a comparison audit of what Buzz doesn't have that my my chat app had and what my chat app did not have that Buzz had and how to basically consolidate or pull what was interesting or valuable from that.

51:19 And we literally just made a couple improvements just because like a project happened and it was done within an hour. And all I know is the conceptual level of like, oh no, this is a great feature. That's a great feature that I could totally add into this and we don't even have to do anything. And then it pushed a new iOS build to my phone and I was immediately talking to him with the new feature.

51:39 like holy crap and we're gonna have robots doing this this is gonna be this is slowly going to trickle down to interfaces that turn into hands and machines working in the real world the the uh where oh man what was the what was the thing let me see if i can find you real quick i gonna pause Ah I almost gave up I almost gave up I could not find the thing And not even AI could find it AI sucks I couldn even find this post Okay, so this guy who invented Amazon Prime.

52:15 I don't know if he actually invented Amazon Prime. After six years in stealth, Atomic Machines is out. Our mission, on-demand, universal command of matter. First beachhead, the matter compiler. An AI-native manufacturing system that builds working micro-machines from code alone. No per-product tooling.

52:37 No process development. Different code, different machine. Doesn't a 3D printer do that? No. No manufacturing technology to date produces a different machine from a different digital input. Imagine a car factory where after a couple of cars roll off the line, a laptop emerges. then a bicycle, then a refrigerator.

52:58 A 3D printer gives you a different shape, not a different machine. AI native is overused, so here's what it means for us. The entire physical system lives inside the AI loop. Every manufacturing operation is followed by precision metrology? Metrology? I don't know that word. Precision metrology. I'll have to look that up.

53:19 We digitize the physical result and a part only proceeds if it's in spec. Deterministic process control, not statistical. That closes the intent to result loop. The AI adapts to errors and runs its own hardware in the loop experiments. Oh, that's crazy. I don't even think I caught quite exactly how that played out the first time I skimmed through this.

53:39 In generating training data continuously, our Gen AI uses it to produce the device design and matter compiler input to build it. At maturity, prompt to product manufacturing. We've aimed the matter compiler at micro machines, tiny machines with moving parts, features measured in microns. Think millimeter scale robots.

53:55 The universe is a grid. Seven physical domains, two purposes. Deliver energy and matter or encode information. Semi-fab has mastered one cell. Six are empty. MAMES engineers have brilliantly coaxed a few devices out of the semi-tool set over decades. Accelerometers, gyros, micromirrors, blah, blah, blah, blah.

54:13 That's what the matter compiler brings. First micromachine out. Here, Prime Switch PS150. A new class of power relay. 200 micro-ohms. Yeah, yeah, yeah. 200 micro ohms on resistance, 150 amp continuous in a 9.5 millimeter by three millimeter hermetic package. Opens in 50 microseconds to 1500 volts of true galvanic isolation, a thousand X faster than a contractor, a contactor, zero holding power shipping to early access customers matter.

54:45 Early access customers now. Why it matters. AI data centers are moving 800 volts DC where a fault delivers destructive energy in microseconds and a DC arc won't put itself out. Mechanical protection is milliseconds too slow. Solid state is fast but can't isolate and bleeds heat on every amp it carries.

55:09 Prime switch is the device in between that nobody could build. So this thing has built a micromachine as one of its first examples of something that manages to, in a hybrid breaker with a solid state bypass, the silicon interrupts, the metal carries and isolates, the breaker never arcs. The same device does transfer switching, hot swapping, and generally economizes semiconductor power across the 800 volt DC data center.

55:42 No semiconductor fab can process prime switches materials or assemble its moving parts. It exists because the matter compiler exists. Matter compiler name for Neil Stephenson's The Diamond Age, which Neil Stephenson is great if you don't read his stuff. $250 million race to date. See it. Blah, blah, blah.

56:00 This, the loop, the huge loop to break down here or to close on this is the capacity to quickly and aggressively iterate on the manufacturing process in exactly the same way I am able to iterate on my debugging process in this application, right? so going straight back to to my example here is I have this little debug note and if I have an error I can in fact uh oh I didn't I didn't close it I want to see I'm going to see real quick I want to see if this is actually done one second all right I'm actually fixing that problem and hopefully I'll actually have video in the background of the last few takes uh because my agent is working on it right now, but I got sick of waiting because I just wanted to finish this episode.

56:58 Just interjecting that I went ahead and added a overdub feature where I can record my screen while listening to the clip that I just did so that I can do the stuff on my screen. And I just replaced the video that we lost And one of the things that I really wanted to bring up and talk about is when we look at the speed at which we iterate I have edited I have changed the application I am making this episode on three separate times before I finish recording the episode and I don't think this will ever really change outside of the fact like because I can just kind of add features like we are had already planned a overdub feature where I could go back and watch something in the timeline and then do my own screen recording after so I'm navigating the

57:51 computer and I'm totally focused on navigating the computer and I'm listening to the stuff that I am saying in the episode and so that I can actually recover the the the scroll through the Twitter feed and stuff that I have because it was like take 49 there were like three takes that were got that got lost while the agent was editing the app but the point is is there's so much there's so much uncertainty in like what to actually invest in and what and counterparty risk in what will actually succeed versus fail and you don't know. But when your money is scarce, when your money can actually hold its value, it becomes the hurdle rate to beat.

58:37 When you have, if your monetary value is increasing by 20% a year, then you don't want any debt going out at less than 20% because it means resources are going to be aggressively wasted. You want the scarcity of the money to set the standard. It's basically giving you the average rate of productivity increase. It's giving you the average rate of productive expansion across the economy. And you want debt to be reserved for those things that can clearly, that at least the people responsible for and who are taking on the risk of that investment, of that debt to be highly, highly confident that it will outpace the 20%, that it will outpace the hurdle rate because otherwise you're wasting resources. Otherwise, if you just did nothing

59:33 and you let the people who were confident and who were willing to risk 20% a year interest on whatever endeavor that they're engaged in because they have far, far higher conviction that they're doing the right thing, you leave those resources for those people other than wasting them on something that's only going to get you 5%, which at a point where you're hitting a singularity is a 15% loss if you have a massive amount of productivity increase happening economy-wide.

1:00:02 And the shift, like everything is going to be processed. I literally think we go from, my entire system has moved from basically something that you build and use continuously to a process that evolves every time I engage with it. Like I said, this app has changed three times and it's a full application built for my production flow. And obviously it's like chock full of bugs and there's some slop and there's technical debt. But that's one of the things that I do on a regular occurrence is I give it to my multiple agents and I tell them to go on an audit and clean up technical debt and universalize this thing. Like, why do we have, why do we build a grid system in one part of the sidebar and then a completely different one in a completely different part of the sidebar?

1:00:53 We should have one grid system that we are applying to these two different locations and these two different, you know, inputs or like selection or group of media. And so like thinking and iterating and reacting to and judging the process is going to be the process itself. But what I love about this article and the reason I wanted to do this episode is just because it's so, it puts into perspective so much of what I have been stuck in and the reason why I haven't done any reads lately and so much of the stuff that we're building with the development team because I want to, I'm, I'm done. I'm done with like iCloud and I'm sick of these platforms and I don't, we don't need them. We can build all our own solutions. And,

1:01:48 you know, there is eventually, we'll have all people building their own solutions to everything, you know, all that stuff, but stuff will coalesce. There will be standards developed. And I think I specifically posted a tweet and a Noster note about this recently, that I went from like six months to a year ago, I was kind of looking at things and thinking like open source had a really big opportunity.

1:02:15 That we could end up where a ton of stuff being used by consumers was open source and where like Max Hillebrand was on the round table recently and something that I loved, a comment that he had made. almost like a year and a half ago maybe now, is that the average number of users for any exact specific piece of software or implementation may very well end up being one, where software is simply unique to the person's environment or to the person's system.

1:02:49 But I think that means still that we should be building things with that concept in mind, with that modularity in mind, so that it can be quickly and easily modified for a separate environment or a separate user. Like we should be thinking of software as moldable things. And there's just so... The unlock is just going to be wild.

1:03:18 It's going to be absolutely like breathtaking. The speed that it is already moving, if you're trying to stay up on the edge of it, and it'll also be really important to take a step back and take a breath every once in a while because I literally just got completely overwhelmed the other day. Actually, it was at the end of last week, and I completely shifted my entire focus and so much of what I was working on this week because of kind of the wall that I ran into at the end of last week where I literally spent five days vibe coding on everything, and I felt like I got nothing to the finish line, but I got so much.

1:04:00 I spent today working basically with all of my tools in my production set, my entire flow that I've built and using my apps themselves and I've built my debugging process and my coordination process cleanly enough that it iterates on itself while I'm working. And I have simplified the crap out of that.

1:04:30 And it is going to be we live in crazy times man We live in crazy times Like to be on the edge I love the comparison of light And what's crazy is that comparison, let me bring this back up. This is going to happen way faster. So, 1830, it was three hours of labor. three hours of labor for an hour of light.

1:05:05 It is insane to consider technological progress that happened so quickly that, and I don't think it's a coincidence too, that like literacy rates were changing dramatically during these types of things, but that to read a story at night was an incredible luxury. Like you had to take a day's worth of work for the opportunity to sit down and read a book at night and have a little bit of light.

1:05:36 And in 1900, 70 years later, it was 13 minutes. And then today, after the electrification of everything, which, you know, was early 1900s, today is 0.07 seconds, seven hundredths of a second to get an hour of light. this is going to happen to a unit of thought, to a unit of intelligent work and iteration.

1:06:13 That is what's crazy about this. We are literally going from pure energy to output, and we are on one of the most aggressive iteration and improvement systems or structures, constructs of technological progress where we are closing that gap in just an unbelievable way straight from energy to output I don know how you could want to live at a different time Like, this is the craziest time in history.

1:06:57 Like, this is so wild. You've got to understand, none of our political systems work in this environment. Like, the monetary system, the debt structure, none of this is going to hold up. this is such a fundamental shift to the productivity and process like things are literally going to change there are going to be so many borders rewritten or redrawn so many different systems altered that trying to define or understand what the world might look like in 20 to 30 years, I think just becomes impossible to imagine. And the thing is, is that this took 170 years, 180, 196 years. And we could very likely see the same thing occur when it comes to the idea of a unit of thought or a unit of intelligent, productive work toward a goal.

1:07:59 we could see this same thing happen in a hundredth of the time in maybe not a hundredth maybe a fiftieth right like maybe 10 years 10 years five years too small but 10 years 20 years like this like I don't even how do you account for this this is the most insane deflationary mechanism ever ever and all of those scarce things the notion of like we have basically the scarce side has been untouched have outrun wage increases and inflation has just gathered into all of these things and this all the technology that is enabling this is shifting over here aggressively aggressively and it going to move so so fast Like I just I don even know what to say about it And I got to go I got to go hang out with my kid and get a little bit of sobriety back. Um, cause we live in wild

1:09:11 times. Anyway, I just want to shout out this great article, crazy, crazy, crazy world we live in. And I hope you guys enjoy the episode. I'm very curious to hear what you think about a live read and the fact that I'm doing a video read now. It's fun to me. Like I have very much enjoyed this. And, you know, curious if you guys get anything out of it because I am iterating on my process and iterating my content a lot.

1:09:46 and there are so many tools to tap into now and I got so much to share. I'm going to be doing videos about tools very, very soon because they are just moving a mile a minute and it is so cool to feel like I have my own platform to connect and network all my stuff together with. So I will catch you guys on the next episode.

1:10:06 Shout out to Bitbox and Shift Crypto guys for keeping my Bitcoin safe and making a great hardware wallet, fully open source, red teaming like a champ, and being really good about security updates and all those things. So stay tuned, grab yourself one. You got a discount code in the show notes. And that will do it.

1:10:29 I'll get you on the next episode. Until then, I am Guy Swan, and that's my two sets. Outro Music