28:32Simon Scrapes · Tutorial
14 GENIUS Ways to Give Claude Code SUPERPOWERS
A 28-minute field guide to the setup decisions that separate Claude Code power users from people still using it like a chatbot.
June 12thThe argument in one line.
The AI industry's subsidization wars are artificially distorting the market, and Chinese open-source models are the only force that might compel frontier labs to reveal what their products actually cost.
Read if. Skip if.
READ IF YOU ARE…
- A developer or founder tracking which AI models are worth switching to right now.
- Someone who follows the business dynamics between frontier labs and the open-source models undercutting them.
- A creator or author with information products who is worried about AI eroding book or course sales.
- Anyone following US AI policy and the specific regulatory pressure on Anthropic.
SKIP IF…
- You want deep technical benchmarks rather than commentary and hot takes on them.
- You have no interest in the US export control story or AI industry business dynamics.
The full version, fast.
The US Bureau of Industry and Security has issued an export license requirement for Claude Fable and Mythos, locking Anthropic out of its own models via government bureaucracy it helped invite. SpaceX has acquired Cursor in an all-stock deal, giving Elon Musk a direct incentive to cut Anthropic off from one of its largest revenue sources. Meanwhile GLM-5.2, a Chinese open-source model, ranks second on the Code Arena Frontend benchmark behind only Claude Fable -- a model that currently does not exist -- at 82% lower cost than Opus 4.8. Microsoft is already exploring DeepSeek over OpenAI and Anthropic for Copilot as usage-based pricing makes frontier models uneconomical. The host argues the real story is that subsidies are masking the true cost of AI, and only when they end will the market find out what everything is actually worth.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:55
01 · Cold open (skip this part)
Satirical hump-day intro; host signals the comedy writing has suffered since Fable went offline.
00:55 – 02:42
02 · Why Fable is stuck
BIS export control letter to Dario Amodei; license required for Claude Mythos 5 and Fable for all foreign destinations; host mocks the SNAP-R application process.
02:42 – 05:16
03 · SpaceX acquires Cursor
SpaceX exercises option to acquire Cursor in all-stock deal at $2.5T valuation; Cursor metrics (1M+ users, 60% enterprise, 67% Fortune 500); Anthropic revenue dependency; Elon's incentive to cut Anthropic off.
05:16 – 07:53
04 · Cursor wants to kill code
Cursor CEO Michael Truell's vision: pseudocode representing software logic replaces formal languages. Mo's counter: every attempt to escape code recreates code because precision demands it.
07:53 – 11:19
05 · Chinese open-source model beats out Fable
GLM-5.2 ranks second on Code Arena Frontend behind only Claude Fable (currently locked); 82% cheaper than Opus 4.8; Microsoft exploring DeepSeek for Copilot; subsidization war thesis and open-source as the circuit breaker.
11:19 – 13:18
06 · AI is killing Tim Ferriss
Tim Ferriss blog post: nonfiction sales down 57% YoY, on pace for 80% fewer copies vs. 2022. His books are lookup tables -- chatbots are the better interface now. What survives: comedy, fiction, irreducible experience.
Lines worth screenshotting.
- The US export control letter to Anthropic requires a license for Claude Fable and Mythos -- Dario Amodei has to apply through the same broken government website as everyone else.
- SpaceX acquiring Cursor gives Elon Musk a direct financial incentive to migrate Cursor away from Anthropic models and choke off a major revenue stream.
- GLM-5.2 ranks second on the Code Arena Frontend benchmark -- behind only Claude Fable, which is currently unavailable -- making it the de facto best accessible model for front-end coding.
- GLM-5.2 costs 82% less than Claude Opus 4.8: $3,000 buys 618 million output tokens from GLM-5.2 vs. 60 million from Claude Fable.
- Microsoft exploring DeepSeek over OpenAI and Anthropic for Copilot is the clearest signal that usage-based pricing is forcing real cost conversations the subsidization era delayed.
- The $200/month plans from Anthropic and OpenAI deliver $8,000-$14,000 of actual usage -- a deliberate loss designed to build workflow lock-in before prices normalize.
- Tim Ferriss's nonfiction catalog is on pace to sell 80% fewer print copies in 2026 than in 2022, with sales down 57% year-over-year in 2026 alone.
- The Four-Hour Body is a lookup table -- in 2026 a chatbot that has read every book is the better interface to those answers than the book itself.
- Cursor CEO's pseudocode vision keeps failing because precision requirements inevitably recreate code -- nondeterminism is the fatal flaw of any abstraction above formal languages.
- The endgame of the AI pricing war is the end of subsidies: whoever can sustain losses longest wins, and SpaceX is probably the most capitalized player in that race.
- What survives AI commoditization of information: experience that is not solely informational -- comedy, fiction, live entertainment, performance.
- GLM-5.1 had impressive benchmarks but failed to penetrate developer culture because subsidized frontier models made price feel irrelevant -- GLM-5.2 is the same bet at a moment when that is changing.
Open-source models are ending the AI subsidy era.
WHAT TO LEARN
When a Chinese open-source model outranks everything except a model that does not exist, the frontier labs' pricing advantage dissolves -- and every workflow built on cheap subsidized tokens is sitting on borrowed time.
- The $200/month plans from frontier labs deliver $8,000-$14,000 of actual usage -- a deliberate loss designed to create switching costs before prices normalize.
- GLM-5.2 ranks above every currently available model on Code Arena Frontend and costs 82% less than Opus 4.8, making the cost argument for frontier labs harder to sustain as subsidies fade.
- Microsoft pivoting Copilot toward open-source models is the first major enterprise signal that usage-based pricing has made frontier AI costs untenable at scale.
- Any information product that functions as a lookup table is structurally vulnerable to AI substitution -- Tim Ferriss's 57% year-over-year sales drop is the data point.
- What survives AI commoditization of information is irreducible experience -- comedy, fiction, live performance, and anything that cannot be summarized into a chatbot answer.
- Cursor's ambition to replace code with pseudocode keeps failing because precision requirements inevitably recreate code -- the problem is nondeterminism, not abstraction.
- The Fable/Mythos export license situation is a preview: when governments treat AI models as weapons exports, even the labs that lobbied for regulation get caught in the machinery they helped build.
Terms worth knowing.
- BIS
- Bureau of Industry and Security -- the US Commerce Department agency that administers export controls on advanced technology, including AI models deemed sensitive.
- Export Administration Regulations (EAR)
- US rules governing the export of dual-use goods and technologies. A license under EAR is now required to export Claude Fable and Mythos to any foreign destination.
- GLM-5.2
- A large open-source language model released by Zhipu AI (China) in mid-2026, featuring a 1M-token context window and benchmark performance competitive with top frontier models.
- Code Arena
- A public benchmark leaderboard that ranks AI models on coding tasks including frontend web development, used here to compare GLM-5.2 against Claude and GPT models.
- Subsidization war
- The strategy of selling AI subscriptions well below cost to build workflow lock-in before raising prices -- the $200/month plans effectively deliver $8,000-$14,000 of usage.
- Usage-based pricing
- Charging per token consumed rather than a flat subscription, which removes the subsidy and exposes the true cost of AI inference, making open-source alternatives more competitive.
- SNAP-R
- Simplified Network Application Process Redesign -- the US government web portal for submitting export license applications under BIS, which Anthropic must now use to get a license for Fable and Mythos.
Things they pointed at.
05:55linkMichael Truell on Lenny's Podcast
Lines you could clip.
01:46
“The reason we can't access Mythos and Fable right now is because Dario can't access the website, man. The website is broken. He hits the submit button. Nothing happens.”
07:53
“Anytime you try to venture away from code, you get into the land of randomness and nondeterminism -- and that's just not what software is.”
10:52
“The endgame is the end of the subsidies. That's when we find out what all of this is worth.”
12:04
“The Four-Hour Body is a lookup table. In 2026, the best interface to those answers is a chatbot who has stolen all his books.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Good morning. It is June seventeenth, Wednesday, also known as hump day in The United States.
00:07Why is it called hump day? Well, in case you're from Europe where nothing ever happens, it's because some dude in the eighteen hundreds was just went around humping went around humping everyone.
00:24Comedy writing here has been ever since Fable left us. The comedy writing here has been awful, but, yeah, some some dude was humping people.
00:32So, you know, go out, hump hump your wife, hump your your husband, your wife's boyfriend.
00:39She has a boyfriend. Let her stay out a little late tonight. Buy her flowers.
00:43Tell her you love her. Tell her thank you. Slap her ass.
00:46I I hear this is great for their confidence. Anyway, there is a lot to talk about today.
00:51A lot.
00:55Still no word on fable. A a little bit has come out. And here's this letter from the secretary of commerce, Howard Lutnick.
01:05Dear mister Amade, some serious serious verbiage here. I'm informing you that a license is required for Claude Mythos five.
01:13Now here's the part that was yeah. If I was diarying this man, I would be a little insulted. He's given them instructions.
01:21Once submitting a license application through the simplified network application process redesign, snaprsnapr.biz.gov, you must indicate in the additional information box that the application is submitted based on this and attach a copy of the letter.
01:35Dude, Dario can't even it doesn't even have a direct line. He's gotta go through the website like everyone else, and the website probably sucks ass.
01:42He's gonna forget his password. He's gonna need to verify his ID. It the reason we can't access Mythos and Fable right now is because Dario can't access the website, man.
01:51. He hits the submit button. Nothing happens.
01:55Is it this is what he asked for, dude? This is what he this is what he wanted? I mean, what did he think it was gonna look like?
02:00This is what you have to wonder. Dario's, like, begging to be regulated. You know, he thought the government would be some neutral, arbitrary, but, you know, this is politics.
02:09This is left versus right. OpenAI, Google, you might consider them somewhat I would consider them centrist if anything, but Anthropic is just so damn lefty.
02:18Just typical San Francisco stuff. It's just so awkward for them. Like, I've seen reporting on this whole saga that, know, the language just isn't there.
02:28There's, like, a language barrier between the administration and Anthropic. They just speak different languages. And you've heard Dario talk.
02:35I mean, the guy the guy speaks inner dude and who can understand that?
02:41Now the big news, of course, is that SpaceX has agreed to acquire Cursor to exercise the option to acquire Cursor in an all stock transaction, which is insane.
02:52As of now, it is the fifth or sixth most most valuable company in the world in the in The US stock market, uh, $2,500,000,000,000 valuation.
03:03These are the most valuable companies revenue versus profit. Now NVIDIA's 215,000,000,000.
03:10Its profit, 117, so about half. Now look at SpaceX.
03:14This tiny little thing here. This is not like me crying bubble or anything like that or or inflation or or hype. Look.
03:21The market loves a good story. That's always been the case. But, you know, you look at a lot of these companies, you're like, where where do they have left to grow?
03:28You look at Apple and Microsoft and Amazon. They've already blossomed. The SpaceX story is is is wonderful.
03:34Is there a a path to them reaching $200,000,000,000 in revenue in in ten years?
03:39I I could I could see a story there. It's really the super nerdy ones who are using Claude Coder Codex. A lot of the super casual people and the people who are not coders at all are using Cursor.
03:50They announced yesterday at their event that they are releasing their own homegrown model from scratch. 1,500,000,000,000 parameters.
03:58Absolutely insane. Same size as Claude Opus GPT 5.5, pretrained on 100 k GPUs.
04:05You you gotta wonder if this isn't something that was made possible by Elon's access to all these GPUs. Cursor grew from 100,000,000 to 1,000,000,000 to 2,000,000,000 in eighteen months, which is the fastest SaaS to $2,000,000,000 ever.
04:201,000,000 plus paying users, 60% enterprise, and 67% of the Fortune 500.
04:27Almost 50% of Anthropix revenue enterprise revenue came from Cursor. This was back in the early days. And so you gotta imagine that a that a substantial size of Anthropic's revenue comes still comes from Cursor.
04:39Uh, I imagine Elon's first order priorities will be to to move that slider all the way down to zero, just completely choke out Anthropic. Will it hurt them? I mean, a little bit, but it's not gonna be a death blow by any means.
04:50So look. I mean, this this is a a really good move, think, for SpaceX and for Cursor.
04:56It all depends on SpaceX's valuation. If the Cursor CEO wants to get paid, he just has to make sure Elon does not tweet for for the next four years. Should be okay.
05:05Do not live live tweet, uh, any of his Ambien Insomnia episodes should be okay.
05:12Uh, speaking of Cursor, do wanna show this tweet from yesterday. This will be interesting to you if you are a developer, Cursor CEO Michael Truell.
05:21Of course, I'm butchering the pronunciation. I never hear these guys pronounce their names out loud. So Truell.
05:26Yeah. Let me just assume the cooler way of pronouncing it. Michael Truell.
05:30I mean, yeah. Who it's impossible. It's impossible for something to be pronounced like that.
05:34Michael Truell. Let's do this. Let's let's look up the pronunciation.
05:39My pronouns no.
05:42Autocomplete. You had one job.
05:45One job, which is to autocomplete this. Truel. Truel.
05:48Our goal with Kershur is to invent sort of a a new type of programming, a very different way to build software that's kind of just distilled down into you describing the intent to the computer for what you want in the most concise way possible and really distilled down to you just defining how you think the software should work and how you think it should look.
06:08You know, there's a group of people who think that, you know, software building in future is gonna look very much like it it does today, which mostly means text editing, formal programming languages like TypeScript and Go and C and Rust. And then there's another group that kinda thinks, you know, you're just gonna type into a a bot, and you're gonna ask it to build you something, and then you're gonna ask it to to change something about what you're building.
06:28And it's kind of like this, you know, chatbot, Slackbot style where you're talking to your engineering department. We think that there are problems with with both of those visions. The problem with the the chatbot style end of things is that it lacks a lot of precision.
06:39You know, the the version of the world where kind of nothing changes, we think we think is is wrong because we think that the the technology is gonna get much, much, much better. And so a world, you know, kind of after, you know, after code, I think it looks like a world where you have a representation of the logic of your software that does look more like English.
06:55Right? You have kind of written down. You can imagine in document form.
06:58You can imagine in kind of an evolution of programming language towards pseudo code. You can you can edit that at a high level, and you can point at that. And it won't be kind of the the impenetrable millions of lines of code.
07:08Um, it'll instead be something that's, like, much terse or easier to understand and easier to navigate. What's interesting is that he's he says this is our goal. But to me, that sounds easy to to to prototype.
07:17Like, how hard could it really be to invent this abstraction? What I think is going on I actually tweeted about this. And I said it it would be funny if this pseudocode representation doesn't just end up in one of these AI labs rereleasing
07:30programming languages in a couple years as, like, a novel new discovery. Like, the fact that it's their goal and they haven't been able to do it just tells me that every time they've tried, they've reached they've just recreated code. Like, you want something precise.
07:42You want something that is predictable. Well, that's just code.
07:45And anytime you try to venture away from code, you get into the land of randomness and nondeterminism, and that's just not what software is. Moving on to a new open source model from China, GLM 5.2.
07:58I mean, I don't know what's going on, but the benchmarks on this look absolutely insane. Uh, you know, almost unbelievable, like, literally.
08:06Like, I I just don't the benchmarks have become meaningless or something because this is an open source model that if you were to take out Claude Fable, if Claude Fable didn't exist, which is the world we live in, then GLM 5.2 beats out OPUS 4.8 and GPT 5.5 on, uh, front end coding.
08:27An open source model that you could sell for I mean, not you probably couldn't afford to, but it's super cheap. It's 82% cheaper than OPUS 4.8. This is a useful chart.
08:37How many how how many output tokens is $3,000 by? If you were to use DeepSea v four Pro, that's 3,450,000,000 tokens with GLM $5.26182000000 tokens.
08:48And with Claude Fable five, that's 60,000,000 tokens. So you could see how insanely of, uh, economical these are.
08:55I mean, this this this single handedly built China's single handedly, like, solving the AI bubble here. Microsoft is exploring deepseq over OpenAI Anthropic as Copilot co work moves to usage based pricing, which Copilot being usage based and being built on top of OpenAI Anthropic, there's zero money to be made there.
09:14I mean, it's exorbitantly expensive, and they're looking to the open source models. As we talked about a couple times already, the $200 plan on Anthropic and OpenAI is getting you $8,000 and $14,000 respectively of usage, highly subsidized to build addiction to and to build the workflows in your company so that once so that once the price goes up, there's nothing you could do.
09:37Satya has published an article on X talking about how we cannot put every everything in the hands of OpenNear Anthropic. We need to make these models sort of replaceable and build the overlying, uh, infrastructure around these models so that we could easily swap one model for the other.
09:55If it wasn't for these open source models, what I would expect to happen is is to continue to see, uh, subsidization wars between SpaceX, OpenAI, and Anthropic.
10:05And whoever runs whoever can can keep the subsidies going longest wins. And if that were the the race, I would imagine that SpaceX would be the most capitalized and they would just sort of compete to the bottom. GLM 5.1 had the same sort of impressive aura to it, but usage wise, it didn't really permeate through the developer culture.
10:26Most developers still preferred, uh, Opus and GPT 5.5, probably because it was so heavily subsidized. Once we see the true price man, if we can only see the true price of these things, if we can get to that phase of this whole episode, we could start having much more better conversations.
10:42But OpenAI and Anthropic are just with these highly subsidized plans that do not allow us to arrive at the endgame. The endgame is the end of the subsidies. That's when we find out what all of this is worth, and we're just having such a hard time making progress here.
10:58And so I cannot wait for these subsidies to be over. And, actually, it's it's already sort of happening very slowly. Everything is moving over to usage based pricing, which you love to see because it's starting to force people's hands, and we're starting to see big actors like Microsoft moving to open source models.
11:13This is really good stuff. If if you're a fan of the bubble exploding, it's sort of happening very slowly.
11:18Uh, blog post by Tim Ferriss, the author of four hour work week. He talks about how AI has killed, uh, sales of his non fiction books.
11:29He says before we dive into my dirty laundry, let's state the obvious. Millions of people have a vague sense that AI is changing things, and LLMs sure are convenient for getting answers quickly. And by 2026, his sales are down 57 over versus 2025.
11:43And he's saying that these books were on the bestseller list for for years and years and were a very reliable, consistent, and predictable source of income for him. He says if the run rate hold, my catalog will sell roughly 80 percent fewer print copies in 2026 than it did in 2022.
11:58And he says what's actually going on is that on some level, the four hour body is a . And in 2019, the best interface to those answers was a book, but in 2026, the best interface to those answers is a chatbot who has read all his who stolen all his books.
12:14Right? I mean, how cruel is that? If how to books are getting crushed because LLMs seem to provide faster, cheaper, and more personalized advice, then what's next on the chopping block?
12:24He says, how to YouTube videos. Why scrub through a twenty four minute video to find the forty seconds you need when an AI can watch it for you? And so he says, will anything survive in its current form?
12:34He says, probably. Experience that isn't solely information. Comedy, entertainment, storytelling, fiction, etcetera.
12:42Yeah. Look, man. The only jobs right now are are stripper or or comedy.
12:46My comedy career is tanking here, so I gotta I gotta stop eating cake, man. Tim Ferriss wrote, of course, the four hour work week, and now the AI can summarize that whole thing in four seconds.
12:58So it it it outdid him, the AI. I mean, it it four seconded him, dude. This is serious stuff, man.
13:05AI is definitely shifting things around. It stole all these people's books, now you can just talk to the to the AI. So hang in there, and I will see you next time.
13:16Thanks for watching.
The Hook
The bait, then the rug-pull.
The title buries the real provocation: Claude Fable -- the model this open-source challenger is supposedly measured against -- does not currently exist, because the US government locked it behind an export license. Mo Bitar opens with a comedy bit, then unspools five stories that together describe a market where the most important AI model is unavailable, the second most important is being acquired by Elon Musk, and the Chinese replacement is 82% cheaper and nearly as good.
Named ideas worth stealing.
10:00model
The Subsidization War thesis
- Frontier labs sell at massive losses
- Build workflow lock-in before price normalization
- Whoever sustains losses longest wins
- Open-source is the only circuit breaker
Mo's explanatory model for current AI pricing: the $200/month plans that deliver $8,000-$14,000 of usage are deliberate loss-leaders designed to create switching costs before prices normalize.
Steal forAny analysis of SaaS pricing strategy or AI vendor selection decision
12:04concept
Lookup table theory of information products
Tim Ferriss's insight that his books function as lookup tables -- and that in 2026, a chatbot is a faster and cheaper interface to the same lookup table than the book itself.
Steal forEvaluating whether a content product is defensible against AI substitution
How they asked for the click.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
