27:33Riley Brown · Tutorial
Claude Mythos Is Coming: How to Prepare Before Everyone Else
A 27-minute briefing on Anthropic's unreleased frontier model and the five-step preparation playbook for using it before your competitors do.
June 7thThe argument in one line.
Claude Fable 5 is capable enough that the human bottleneck is no longer intelligence but account setup and payment authorization — making reusable, agent-accessible building blocks the real competitive moat going forward.
Read if. Skip if.
READ IF YOU ARE…
- A developer or builder already using Claude Code or agent platforms who wants to understand what Fable 5 changes about what is feasible today.
- Someone thinking about what software products to build in an AI-first market and looking for a framework on where value will accumulate.
- A vibe-coder using Sonnet or Opus who wants to know if the Fable 5 cost spike is worth it before the June 22 plan-inclusion window closes.
- An entrepreneur trying to understand why some companies (Supabase, Vercel, Convex) get defaulted to by AI agents and others do not.
SKIP IF…
- You want a rigorous technical benchmark comparison — this is a practitioner demo, not an evaluation paper.
- You are doing frontier LLM research — Fable 5 auto-throttles those topics back to Opus 4.8.
The full version, fast.
Claude Fable 5 is Anthropic's first Mythos-class public model — above Opus, notably stronger at tool use, spatial reasoning, and following visual examples. Three live demos make the capability case: a Lovable clone built in two prompts, a 23-page McKinsey-style research deck generated from a single PDF upload, and a persistent computer-use agent texted over iMessage. The more durable argument is structural: agents are good enough now that the remaining bottleneck is account setup and payment authorization. Agentic payments will remove that bottleneck, making well-documented, agent-accessible building blocks the infrastructure moat of the next economy. Use Fable 5 now while it is still included in Max and Pro plans; after June 22 it shifts to API pricing at twice the cost of Opus.
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 – 02:47
01 · Introduction — What is Fable 5?
Announces Fable 5 as a Mythos-class model released June 9, 2026. Explains model hierarchy (Haiku/Sonnet/Opus to Mythos/Fable). Context: 3.5 Sonnet birthed vibe coding in 2024; this is a new tier entirely.
02:47 – 05:17
02 · Building Lovable in 2 Prompts
Side-by-side live demo: Rylable vs actual Lovable both build a Notion dark-mode notes app. His version produces inline tables and editable headers; Lovable does not.
05:17 – 07:10
03 · Open-Sourced Rylable
Shows the full 8-prompt chat thread, demos Minecraft game built via voice, explains Daytona + Convex as the core building blocks, announces open-source release.
07:10 – 08:38
04 · Recreating a McKinsey Research Report
Uploads a McKinsey PDF, prompts Fable 5 to make a matching AI outlook for H2 2026. Result: 23-page Meridian Global Institute deck with matching chart styles in one shot.
08:38 – 11:37
05 · Presentation Tip — Finding and Using Examples
Tip: ask Claude to find and download 20 high-quality comparable reports, drag them in as style references. Demos max-effort mode and saving the result as a reusable chart-report skill.
11:37 – 12:47
06 · Texting Claude Fable 5 via Chorus
Shows Chorus.com agent platform — texts a computer-use agent over iMessage, switches to Fable 5 via /model, gets a native Swift iOS app built and delivered to his home screen.
12:47 – 14:26
07 · Games Built with Fable 5
Community showcase: Pat Simmons's Age-of-Empires-quality RTS, a city block simulator with multi-agent traffic and night/day switching.
14:26 – 15:37
08 · Limitations of Fable 5
Frontier LLM research silently auto-throttles to Opus 4.8 via prompt modification and steering vectors. Security/hacking also restricted. Anthropic staff have unrestricted Mythos internally.
15:37 – 21:30
09 · The Building Block Economy
Core thesis via Mitchell Hashimoto: agents should use reusable building blocks, not reinvent infrastructure. Your SOP/skill/plugin IS the moat. Agentic payments will be the next unlock.
21:30 – 22:24
10 · Pricing + Fable Window Closes June 22
Fable included in Max/Pro plans until June 22, then API pricing at 2x Opus. The Lovable clone cost approx 200 dollars in API credits. Use it now, be ambitious.
Lines worth screenshotting.
- Claude Fable 5 is a Mythos-class model — a whole new tier above Opus, not just an incremental Sonnet-style update.
- Fable 5 can clone a production-grade app like Lovable in two prompts by following screenshot examples directly.
- One-shotting a 23-page McKinsey-style research deck costs roughly 200 dollars in API credits and takes under ten minutes.
- Fable 5 auto-throttles to Opus 4.8 on frontier LLM research topics, silently, via prompt modification and steering vectors.
- Anthropic employees have unrestricted access to the full Mythos model; the public Fable 5 is a safeguarded subset.
- AI agents are now better than humans at using building blocks — the bottleneck is no longer intelligence, it is account setup and payment.
- You almost need to build brand recognition with AI agents: Supabase became a 10B company partly because agents default to it.
- Agentic payments will let agents autonomously sign up for and pay for building blocks — making the /budget 100 clone lovable command paradigm real.
- The most valuable thing you can build right now is a reusable, agent-accessible skill or plugin that agents can invoke repeatedly.
- If you write really good instructions for AI agents, they can use your building block as effectively as any employee.
- AI agents hiring humans on Fiverr when they need help is an emerging workflow that agentic payments will unlock.
- Fable 5 is included in Claude Max and Pro plans only until June 22, 2026 — after that it switches to API pricing at 2x Opus cost.
- The building block economy argument: agents should not reinvent infrastructure any more than you would reinvent highways just for your commute.
- Max effort mode in Claude is often overkill — useful for smoke-testing a skill, wasteful for most production tasks.
Building blocks beat model capability as the real moat.
WHAT TO LEARN
Fable 5 is powerful enough that the constraint on what agents can do is no longer intelligence — it is access to well-documented, agent-accessible building blocks.
- Giving an AI agent a concrete visual example — a screenshot, a PDF, a URL — is more reliable than describing the output in words; the McKinsey deck demo works because the model has a direct style reference, not just a verbal description.
- Reusable building blocks (databases, sandboxes, hosting, auth) are the infrastructure layer of agent-native workflows; the agent that defaults to your tool because your documentation is clear becomes your most scalable distribution channel.
- The human bottleneck in current agentic workflows is account setup and payment authorization, not reasoning — which means the near-term unlock is agentic payments, not smarter models.
- Writing down your SOPs and workflows as agent-accessible skills is not an automation project; it is the equivalent of hiring staff who can execute your processes without retraining on every run.
- The most expensive model is not always the right choice — max-effort mode on a 23-page report is overkill for most tasks; understanding when to drop back to a cheaper model is itself a skill worth developing.
Terms worth knowing.
- Mythos-class model
- A new tier in Anthropic's model lineup above Opus, representing a qualitative capability jump rather than an incremental update. Fable 5 is the first Mythos-class model released for general public use.
- Building block economy
- A framework coined by Mitchell Hashimoto arguing that the most valuable assets in an agent-first world are reusable, well-documented software primitives that agents can invoke — not the agents themselves.
- Agentic payments
- An emerging capability that would allow AI agents to autonomously sign up for and pay for third-party services without human intervention, removing the last major bottleneck in fully autonomous workflows.
- Daytona
- A cloud sandbox service that spins up isolated virtual machines for code execution. Used as the sandbox building block in Riley's Lovable clone so every app runs in its own cloud environment.
- Convex
- A database-as-a-service provider that AI agents can integrate with via API key. Frequently chosen as the default database building block in AI-generated apps.
- PEFT
- Parameter-Efficient Fine-Tuning — a technique used by Anthropic to steer Fable 5 away from frontier LLM research topics without retraining the full model.
- Steering vectors
- Activation-space interventions applied at inference time to shift model behavior in a particular direction — one mechanism Anthropic uses to limit Fable 5 on restricted topic areas.
- Chorus
- Riley Brown's agent platform (chorus.com) that makes Claude Code-style computer-use agents accessible via iMessage, letting users text commands to an agent running on a full desktop environment.
Things they pointed at.
Lines you could clip.
01:03
“I'm telling you for 99.9% of tasks, it is the best model in the world.”
05:13
“This ad is better than Lovable. It's actually better than Lovable.”
18:07
“You almost need to build a brand to AI agents.”
20:15
“AI agents hiring humans when they need some help — that is going to unlock so many different workflows.”
21:54
“That lovable clone was 200 dollars in API credits. So it was like eight prompts, 200 dollars.”
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
00:00This was the biggest week in the history of AI agents. Anthropic released a new AI model called QuadMythos, and they're calling it Fable five.
00:10And if you follow me on x, you saw that I built Lovable in two prompts with Fable five, and I also open sourced it.
00:18And normally, for these weekly updates, I like to talk about the most important news and updates from all of the Frontier agent companies, but this week, Claude simply dominate, and there's very little updates from anyone else.
00:31So on this week's episode of agent native, we're gonna be talking about the Claude Fable release. I'm also gonna show you what people are building with Claude Fable five, and then I'm gonna talk about what this means and how we're entering a new era of the building block economy for agents like Mythos and how to prepare for this future.
00:49Let's not waste any more time. Let's dive in. Alright.
00:52So about three days ago, on 06/09/2026, Claude or Anthropic released a new model and it's called Fable five and it is a Mythos class model. So it is in the Mythos class which means it's above Opus and it is the best model in the world.
01:08I'm telling you for 99.9% of tasks, it is the best model in the world. It is a slightly modified version of Mythos that has some safeguards in place and they're saying it's to be made safe for general use.
01:23Many people are upset with how they made it safe for general use but we'll get to that a little bit later. And as I said earlier, this is a whole new class of model. Since mid twenty twenty four, we've been getting three types of models, either Opus, Haiku, and Sonnet.
01:37The most powerful, the middle one is Sonnet, and the least powerful is Haiku. And we've been getting these Dedi updates, and I think the biggest jump in improvement was actually to Claude 3.5 Sonnet, which gave birth to vibe coding in the middle of twenty twenty four.
01:52And then eventually, I think Opus 4.5 at the end of last year was huge jump. Now this is a whole new ballgame. It is a whole new type of model.
02:01Claude Mythos is bigger, it is more expensive, and it just kind of feels like a whole different beast. Based on my own experience of using this for the last three days straight, Claude Mythos five is incredible at using tools and it uses tools for a lot longer.
02:18It's also better at spatial reasoning and it's just more visually oriented. If you can give it a full Mckinsey deck, I did this and it went super viral on x.
02:29I gave it a full Mckinsey deck and it recreated it perfectly.
02:33It made a research report on a different topic nearly perfectly and the deck almost looked identical. If you give Claude Cowork really good examples, it can recreate them and it's super fun to use.
02:46And so with Mythos, I was able to recreate the lovable mobile app. Check out this clip.
02:52Guys, Claude, Fable or Mythos is absolutely insane.
02:58On the left here, we have the actual level Lovable mobile app. And here, we have a Lovable mobile app that I built in two prompts with Claude Fable five. This is the chat thread.
03:10Here's prompt one and then I just screenshotted all the screens on Lovable. I just need you to redesign it to look exactly like this and then that's it. We're gonna use these two at the same time.
03:21So here's Lovable. So on Lovable, I'm just gonna say make a Notion like notes app dark mode. Okay.
03:29So now I'm going to make a Notion like notes app and make it dark mode. Okay. So this is my version of Lovable that I've I've coded and here's actual Lovable.
03:38I'm gonna enter these prompts at the same time. Look at this. Mine's faster.
03:42This is actually running in a sandbox. This is literally lovable.
03:47It's literally lovable. That is good. It's lovable.
03:50It's two prompts. It's lovable. Okay.
03:52Starting the web server. So this one was about seven seconds before mine, but mine's probably using a better model low key. Alright.
04:00So let's open it on the actual Lovable. And there you go. We have this little notes app.
04:04Now I'm gonna open it up in my version of Lovable. Bang. So create a new page.
04:11Look at this. Yo. They don't have any of the features.
04:14Look at this. Hello. I can't change any of the headers or anything.
04:18Look. Look on this. I can literally hello.
04:22There. This was made on Mythos.
04:27Header one. Look. I can literally change all the text.
04:30Couldn't do that on the other lovable. But on both of these, we can make an edit to the app. So I can go back.
04:35Allow me to create add let me add tables in line in my doc. Please allow me to add tables. My team will look it turns out to my docs.
04:42And now they're both editing. Okay. Look.
04:45They're both done. Let's go ahead and open up the lovable. Let's see if I can actually add a table.
04:50I don't know how I would even add the table. Here, let's open up our version. Look at this.
04:55My version is better. Look at this. We have a table.
04:59What on earth is going on? Dude, where's the where's the table option?
05:05Where's the table option? Add? Bro, this is what are we doing?
05:10What are we doing here? This ad is better than Lovable. It's actually better than Lovable.
05:15So yeah, I created that in about two prompts. Here is my app eight prompts later and I've actually open sourced this and I'll put the link in the description. You can actually download this code and I've changed the name of the app to Rylable.
05:27So I will probably get a seasoned assist from Lovable. That's alright. Whatever.
05:32I can create web apps. I can also create mobile apps and I can even use Fable five within my version of Lovable. And this is my whole chat thread.
05:40I will even post the entire chat thread down below. I literally created Lovable. And I'm just gonna use 4.8 because I wanna conserve my Fable tokens, but I can build whatever app that I want.
05:51I say, build a minecraft game. Make it simple.
05:56I even want to make the character fun. I just wanted to show you that I added this voice mode and it works.
06:03Right? I added this voice and I can ship it off. This uses Daytona for sandboxes.
06:09So every time you make an app on one of these five coding platforms, it spins up a sandbox. And you can see all of every single app here, all of these that say live, they are running in a virtual computer in the cloud. And all I had to do was give it access.
06:23I just had to tell Claude to do or I just had to tell Fable five. I said use Daytona for sandboxes and use Convex for database. No need for off.
06:32And because I have a Daytona skill that I've already set up and it knows my account and has the key, it was able to add that seamlessly. And so it's already done. Not only took like forty five seconds to generate.
06:42It is a simple two d Minecraft style mining and building game because I did specify to make it simple. But here we can actually preview. This is a real app.
06:51See I can move the character around. It looks like I can change the character, and this is an actual app, and I can even open it in an external browser.
07:01You can see here it opened up this link just like Lovable. It has a little d pad down here. I don't know what these do but yeah we have this little Minecraft game.
07:09And the next big test that I did with Claude Fable five that also went viral on x and almost got a million views is I uploaded this entire McKinsey document that has a very specific formatting and very specific chart and visual styles.
07:26And so then I just uploaded it and gave it to Claude with that's running Fable five and I said please create a presentation like this except on AI for the second half of twenty twenty six. Make high quality research predictions and create graphics, make them in the same style, and this is what it created.
07:43So here's the report that it made, and it even came up with this new name called Meridian and Partners. And look at this. It organized it to look exactly the same.
07:55It was really well organized, and it even had scoring for each part. And check out these graphics here.
08:03These charts look exactly like McKinsey. And by the way, McKinsey charges hundreds of thousands, sometimes millions of dollars for these research reports. And, yeah, I think this is really good.
08:13It talked to me about data center electricity demand more than doubles by 2030 even as near term US power gap emerges. Talked about open weights, almost near frontier models.
08:24And I mean, come on. These graphs are actually incredible. They just one shotted this entire research paper and I actually just like read the whole thing.
08:33It was useful for me to understand the AI space. And so that is a really, really good tip. And so one useful tip, if you wanna create really high quality like research reports or something like that, you can always simply just go to Claude.
08:46Please give me a list of 20 high quality examples of McKinsey like reports. 10 of them should be McKinsey reports, 10 of them should be other companies that charge similar amounts for their reports.
09:00They should be high quality. You should give me links to down, uh, to download. I shouldn't have to sign into anything.
09:05I should be able to click on it and then click download. Please search for a while, find the best highest quality research reports and give them to me. I can click on the link and download them.
09:15And this is just a good way to have Claude go off, scour the Internet, find high quality examples, and it will give you links that you can download. And there you go. McKinsey Global Institute 2025 in charts.
09:27Oh, I like this one a little bit more. This should take me directly to oh, this is nice.
09:33I'm actually gonna have, watch this, I can just download this right here and now we can go back to Claude. You can simply drag it in, make it look like this.
09:43So way more charts. And what I can do is I can use this new, I'm gonna change the model to fable five, and I want this to go ham.
09:53So what I'm gonna do is I'm gonna change the effort to max, and I'm gonna say, please make it look like this. So way more charts, really internalize the style of these charts and you're using the same context as our report earlier, but you may wanna do some research before and I want you to really, really go hard here and make the report that on the previous topic that we were discussing in this style.
10:19Make the charts perfect and amazing and formatted in a very similar fashion as they are in this report. And then I want you to turn this into a skill once you're done and call this chart report.
10:30So then I can use the chart report again later. Okay. I'm not gonna lie.
10:34This is taking a while. I did not need to use max effort. I just wanted to show you that you can use max effort.
10:40This is just not necessary. It's going overkill. It's smoke testing the skill.
10:46This is insane. 23 pages rendered. I'm excited to see what this looks like.
10:50Okay. Finally, it is done.
10:53So let's take a look and see what this looks like. Oh, let's go.
10:58Let's see what this looks like. Let's check out these charts. Oh, let's go.
11:02Damn. This is fire. Hold on.
11:06Oh, what? Look at how solid this is. This is crazy.
11:11Okay. Here we have the revenue between OpenAI and Anthropic. Look at these charts.
11:16Oh my god. This is a really this it just generated this in one shot with incredible charts. It did all of the research and then generated this perfectly.
11:27And all I needed was an example and it keeps using this Meridian Global Institute. I'm not sure why it's doing that, but this is the report that it created and this is awesome.
11:37I've also been texting Fable five for the past two days straight. But here, I can actually type slash model. And so right now, I'm messaging an agent like Claude Code, except it's running on a computer.
11:48And you can set this up in, one minute on chorus.com. This is our project. We're trying to make agents as accessible as Poke but as as powerful as Claude Code or Hermes.
11:58But I just did slash model and I can actually choose Fable five, please. And it will switch to Fable five and I can actually build whatever I want.
12:07And I had Fable send me this link earlier. So this is an iOS app that I can build. I can just ask my agent to build a native Swift app and it will download it directly to my home screen.
12:19It also manages my emails and is connected to all of my integrations just like Claude code. But here's this here's what Fable built. I can jump around.
12:27It's like a Minecraft game. Switch to a bow and, like, look at how solid this game is. So no.
12:33Please don't die. No. No.
12:35Don't fall off the map. Shoot him. Yes.
12:37Get him. So anyway, yeah, Claude Fable just built me this, and so I've just been texting it all day, and it's incredible at games. I highly recommend that you try and use this for games.
12:47Speaking of games, this guy Pat Simmons used Claude Babel five to create this game. This is probably the most insane game I've ever seen get created. Check this out.
12:56Alright. Map moves. There's our civilization.
12:58This is insane. This is I mean, this is exactly what the, like, graphics were. Like, this is just as good as Age of Empires.
13:04This is wild. I mean, look at the freaking difference. Like, this is So this is not a single This is Opus, uh, 4.8.
13:11So this video, I highly recommend watching this video. He builds out Age of Empires in 04/2008 and still looks really good, Opus 4.8. A lot of upgrade just in terms of graphics.
13:21Look at the difference here. Nothing else. I don't know what the actual gameplay is, and I'm probably not gonna get too much of the gameplay because I never know how to play these actual games.
13:27But, like, look at this freaking thing. This is wild. So okay.
13:31Alright. Let's just try to click some buttons here for idle. Okay.
13:34House. Oh, nice.
13:36Wow. That's cool. Look at this.
13:38Oh my god. Alright. Farm.
13:40Left click to place. Right. That's and then it looks like we have, I don't know, some things over here.
13:45And so you could basically just do anything. You can make any sort of game and you can even get super weird. So this is my friend, Belawful Sudhu.
13:53He said, just use Claude Fable aka Mythos to create this city block simulator complete with multi agent traffic, live detection boxes, tracks, etcetera, etcetera. Look at this.
14:07Like, I don't even know what's going on here, but it's awesome.
14:14It plots all of the coordinates. It's switching from nighttime to daytime, and this is just something you can just create now with Claude Fable five.
14:25Okay. But there are negatives to this model at least in terms of what people are complaining on Twitter and other platform. This person tweeted when Fable five is used for Frontier LLM development, it does not notify the user and instead limits the model's capabilities through methods such as prompt modification, steering vectors, and PEFT.
14:45If you try to do very in-depth LLM research, Frontier LLM research, I'm not smart enough to do this, it'll auto switch your model from Claude Fable five to Claude Opus 4.8.
15:00So topics like building pre training pipelines, distributed training infrastructure, or machine learning accelerator design.
15:07And so this is the difference between Fable and Mythos. Internally at Anthropic, they have no restrictions. They can use Claude to do Frontier LLM research to make their AI models better.
15:18Unfortunately, us, we as as the common folk, we do not have access to that model and a lot of people are definitely sad about this. Okay.
15:27And so if you avoid LLM research topics, securities and vulnerabilities slash hacking, you'll be fine. You'll be able to get a ton out of this model fable five.
15:37So what I wanna talk about to close out this video is kind of this new era of AI agents where AI agents are getting so good at using these reusable building blocks to create anything. Right?
15:48So many of those games that were created use three JS which is a technology that is a building block that AI agents are getting really good at. When I created Lovable, it used it used a certain building block.
16:00It used Convex which is a database. It used Google sign in which is another building block. It used a tool called Daytona for sandboxes.
16:08I used AI gateway by Vercel to be able to create apps with any AI model and then I also used Vercel for hosting.
16:16And so these are all of these different building blocks that can be used by AI agents. And so I wanna play a clip real quick and then I'll close out this video. Did you guys see the there was an article on x by Mitchell Hashimoto called the block economy or the building block economy, something like that.
16:32Like, his argument is that the most useful thing for agents to have now is really powerful reusable building blocks. Because to Max's example, you wouldn't expect your clanker to reinvent a queue infrastructure system every time he needs to send an email.
16:47It needs to bring in the right building block that's right sized for the task that you're asking for, and then say, well, okay. For this one, it's BOLMQ. I challenge the notion that I would want the agent to reinvent the entire universe and first principles in a way that's incompatible with the rest of society and civilization.
17:04Like, it's almost like reinventing highways, laws, policies, etcetera, just for you. What he's saying is, why would you expect an AI agent like Fable five to recreate its own database
17:15when there's tools like Convex, Supabase, Neon that are making these databases accessible by AI agents.
17:23This is a new economy and this is kind of the central purpose of if you go to Claude right now or if you go to Codex and you go to customize and you browse plugins, these are all building blocks for your AI agents.
17:37And so that's why I've been so obsessed with these plugins and why if I was an entrepreneur, what I would be doing is I would be trying to make software as building blocks because so many people are gonna be using AI agents directly through Quad and Codex and these AI agents are gonna get better. And AI agents are just gonna know which building blocks they should use.
17:57You almost need to . Right? If you ask AI to vibe code an app with a database, it's likely just gonna start using Supabase, which is a tool that AI agents almost just default to Supabase, and now they're a $10,000,000,000 company.
18:13And so these are just building blocks that other companies create, but part of running a company now is actually building the building blocks for your company. These are things like skills, like plug ins, any sort of SOP or standard operating procedure you should have written down and you should give access to AI agents.
18:32And you can create these bundled building blocks that your agents can use over and over again. And if you write really good instructions for your AI agents, they'll be able to use them as effectively as any sort of employee. And this is the most important thing right now and what I'll be talking about in my content going forward.
18:51It's like how do you take a thing that you do every day, maybe something you don't like to do and how do you turn it into a building block for AI agents to use. And in the case of cloning lovable, I actually had to create an account on Convex.
19:05So Convex was the database provider and I had to give it an API key. Right? I have a skill that uses Convex.
19:12I actually think it's a plug in and it's connected to my account so I had to manually do that. To set up sandboxes on Daytona, I had to manually set this up and I had to give them a credit card in order to use it.
19:23For Vercel, same thing. I had to manually set it up and get an API key. However, what we're entering is a whole new world once we move to agentic payments.
19:34Very soon, very soon, all of these building blocks will be things that AI agents can just sign up for. And what I think there's gonna be, I think there's gonna be something like a budget command where basically you could say, I want budget.
19:49So like you could just say budget budget 100 clone lovable.
19:54And so basically, the AI agent, because it can basically do anything a human could do on a computer, it could surf the internet and decide that it needs to use these building blocks. It may not even need me to specify which building blocks to use.
20:08It'll just know that it has a budget of a $100 and it will manual it will just automatically sign up for these services and then it will build my app perfectly. And so this is a really exciting time because right now so many of the workflows in my company even when I use Fable five end up hitting a wall because I need to manually sign up or manually pay for something.
20:28As soon as the AI agents can pay for things including maybe going to Fiverr and hiring a human, that's gonna be a huge industry is literally AI agents hiring humans when they need some help. That is going to unlock so many different workflow. And so this is just something I want you to be aware of.
20:46AI agents are so good, basically better than humans at using all of these building blocks, but right now it still kind of requires a human to set it up, but that time is coming to a close very soon because I've talked to many people at these companies doing agentic payments and it's going to unlock so many fun workflows.
21:03And I'm just really excited to keep talking about these building blocks that make using AI agents fun. And, uh, yeah, I'll be covering that in future videos. This was the fourth episode of agent native.
21:15This is the first one where I just covered one kind of big announcement that nothing really happened this week other than Claude Mythos. It dominated the narrative and it is worth trying. It is expensive.
21:26You may run out of tokens pretty quickly, but it is the cheapest it will ever be. I want to point this out really quick. If you go to your chat here and you choose Fable, it is only included in your plans until June 22.
21:41Then after that, it'll switch to API usage. This is as twice as expensive as Opus. To put it in perspective, that lovable clone that I created earlier was $200.
21:51That would have costed $200 in API credits. So it was like eight prompts, $200.
21:57I will say it did go for a very, very long time. But just to put it in perspective, this model is very expensive and now is the best chance to use it with your max plan or your pro plan even. So I highly recommend trying it out.
22:10Be as ambitious as you can. I'm telling you, you can build way more than you think with this model. Have fun and I'll see you here for the next video.
22:18Much love. Hopefully for the next video, I'll be in my studio. I'm really excited for that.
22:23Peace.
The Hook
The bait, then the rug-pull.
Riley Brown opens with a bold claim — biggest week in AI agent history — then delivers: Anthropic shipped Claude Fable 5, a Mythos-class model that he says beats every other model for 99.9% of tasks. He built a Lovable clone in two prompts to prove it.
Named ideas worth stealing.
15:37concept
The Building Block Economy
Mitchell Hashimoto's thesis: AI agents should call existing well-documented building blocks rather than reinventing infrastructure. The agent picks the right block for the task; the value is in the block, not the agent.
Steal forpositioning any software product for an AI-agent-first go-to-market strategy
19:30concept
Agentic Payments Paradigm
- /budget 100 clone lovable
A proposed command paradigm where you give an agent a budget and it autonomously signs up for, pays for, and orchestrates whatever building blocks the task requires.
Steal forthinking about what human-in-the-loop steps will disappear in the next 12 months
08:38model
High-Quality Example Method
- Ask Claude to find 20 high-quality examples
- Get direct download links (no sign-in required)
- Drag examples in as style reference
- Prompt: make it look exactly like this
A repeatable workflow for getting Fable 5 to produce McKinsey-quality outputs: first have Claude curate the best examples from the internet, then use those as direct style references.
Steal forany design-heavy or format-heavy output task: decks, reports, landing pages, email templates
How they asked for the click.
VERBAL ASK
21:30product
“Try Fable 5 now while it is still included in your Max or Pro plan. Be as ambitious as you can.”
Soft urgency close — gives specific deadline (June 22) and cost anchor (200 dollars for 8 prompts) to motivate action without being pushy.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
