21:14Jack Roberts · Tutorial
Every Level of Claude Fable 5 Websites Explained
Seven escalating techniques that take a Claude Code website from a 3-out-of-100 chatbot draft to a one-shot design worth stealing.
July 5thThe argument in one line.
Wiring a frontier model into a personal 'agentic operating system' across five stacked layers, routing, shared memory, multi-model debate, autonomous overnight work, and a voice interface, turns one subscription into a system that plans, argues with itself, and keeps working unattended.
Read if. Skip if.
READ IF YOU ARE…
- You already run Claude Code or a multi-model agent harness like Hermes and want your tools to share memory instead of starting from zero in every chat.
- You're comfortable wiring together APIs, dashboards, and prompts yourself rather than waiting for a packaged consumer app.
- You want a concrete architecture for mixture-of-experts orchestration, not just a headline about a new model release.
SKIP IF…
- You're looking for a plain explanation of what Fable 5 is — this assumes you already know the model and jumps straight into building on top of it.
- You don't want to run local dashboards, OpenRouter API keys, or a custom memory system — this is DIY infrastructure, not a hosted product.
The full version, fast.
Jack Roberts argues the highest-leverage AI project this year isn't picking a model, it's building a personal 'agentic operating system' around one. He demonstrates five stacked levels on top of Claude Fable 5: giving it routing control over a Hermes Agent harness, linking memory across every tool so nothing is repeated, using Fable 5 as an orchestrator that debates DeepSeek, GLM and GPT models and synthesizes their answers, letting it review a full day's activity overnight and propose improvements unattended, and finally wrapping the whole dashboard in a voice interface that can answer questions and edit the UI live. Every layer is built the same way: describe the change in plain language to a coding agent rather than writing it by hand.
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 – 00:36
01 · The Agentic OS Explained
Cold open; Jack frames the video's promise across five levels of an agentic OS built on Fable 5.
00:36 – 01:11
02 · Five Levels With Fable 5
Introduces Fable 5 as the new top model, shown on a FrontierCode accuracy-vs-cost chart, and previews the five-level structure.
01:11 – 02:28
03 · Why You Need Unified Memory
Argues for one shared memory layer across every AI surface, phone, Hermes, Claude, so nothing has to be repeated, and teases dreaming and cross-tool orchestration.
02:28 – 03:00
04 · Level One: Giving It The Wheel
Level 1 defined: hand routing and scheduling decisions to Fable 5 so it can surface unused skills automatically.
03:00 – 05:22
05 · Adding Fable 5 In Hermes
Live walkthrough adding Fable 5 as a selectable model inside the Hermes Agent harness via a Claude Code prompt, then sets its effort level and confirms it responds.
05:22 – 07:32
06 · Finding Skills You're Missing
Prompts the OS to scan every Claude and Hermes chat log and recommend the two highest-ROI unused skills, and shows the onboarding wizard's hourly-rate-based savings calculator.
07:32 – 09:00
07 · Level Two: One Brain
Defines Level 2: one memory core linking an Obsidian vault, Claude Code, and Hermes so every surface draws from the same knowledge.
09:00 – 11:20
08 · Improving The Memory System
Prompts Fable 5 to rebuild the memory UI into a clickable, clustered knowledge graph, topics like YouTube retention and intros, with a live search bar.
11:20 – 12:26
09 · Level Three: Ministry Of Agents
Defines Level 3: an orchestrator model that fans a question out to several sub-models, compares answers, and returns one debated response.
12:26 – 14:48
10 · Fable 5 As Orchestrator
Swaps Opus 4.8 out and Fable 5 in as the orchestrator over DeepSeek V4 Pro, GLM 5.2 and GPT 5.5, using OpenRouter and prompt caching to keep the multi-model debate cheap; demos it settling a debate question.
14:48 – 16:22
11 · Level Four: Dreaming Overnight
Defines Level 4: Fable 5 works unattended overnight across all logged activity and proposes improvements the next morning, demoed as '4 improvements found overnight'.
16:22 – 17:22
12 · Level Five: A Voice You Can Talk To
Defines Level 5: a voice interface that can be talked to like a person and follows the user around the OS.
17:22 – 17:56
13 · Building The Voice Line
Prompts Claude Code to build a voice-driven 'Oracle' widget on top of the dashboard that can answer questions about the OS and edit it live.
17:56 – 18:59
14 · Live Status Report Demo
Live voice conversation demo asking the Oracle widget for a status report; closes by teasing a five-things-to-do follow-up video.
Lines worth screenshotting.
- A single frontier-model subscription can be wired into five layers of a personal AI system: routing, memory, multi-model debate, overnight work, and a voice interface.
- Giving a model 'the wheel' means letting it scan every chat log across your tools to recommend the highest-ROI skills you aren't using yet.
- A shared memory core lets an AI on a phone call, in a coding agent, and in a chat app all draw from the same knowledge instead of being re-briefed every time.
- Mixture-of-experts orchestration works by having one model send the same question to several others, then reading all the answers before responding.
- Prompt caching is what makes it affordable to ask three separate models the same question for an adversarial review instead of trusting one model's first answer.
- An 'overnight dreaming' feature has an AI review a full day of chat logs, code activity, and integrations unattended, then present concrete improvements the next morning.
- Voice interfaces built on top of an agent dashboard can be asked for a live status report and can edit the dashboard's own UI on request.
- Onboarding a personal AI dashboard with your hourly rate lets it calculate, in dollars, how much time each connected skill is saving you per day.
- A notes vault can be reorganized by an AI into clickable, clustered topics with its own search bar instead of staying a flat pile of files.
- Swapping which model sits at the top of a mixture-of-experts setup, e.g. replacing an older orchestrator with a newer one, is a drag-and-drop configuration change, not a rebuild.
- Asking an AI to improve its own memory or dreaming system, rather than specifying the exact redesign yourself, can produce a working new UI from a single prompt.
- The same build pattern, describe what you want changed in plain language to a coding agent, is used across all five levels instead of writing code directly.
One model, five layers, a full operating system.
THE FIVE-LEVEL STACK
The real lesson isn't Fable 5 itself, it's a repeatable pattern for layering routing, shared memory, multi-model debate, autonomous work, and voice on top of any capable model.
02Five Levels With Fable 5
- Model comparisons are increasingly framed as accuracy-per-dollar, not just raw benchmark score, so it's worth checking a model's cost curve before switching your default.
- A five-level mental model, routing, memory, debate, autonomy, voice, gives a way to plan an AI-tooling build incrementally instead of all at once.
03Why You Need Unified Memory
- If you use more than one AI tool, e.g. a coding agent and a chat app, look for a way to link their memory so you stop re-explaining context in each one.
- The value of connected memory compounds: the more tools draw from the same knowledge base, the fewer gaps show up in any single conversation.
04Level One: Giving It The Wheel
- Before adding new tools, ask an AI with access to your chat history which existing skills or integrations you already have but aren't using.
- Framing a request as 'find the two highest-ROI things I'm missing' gets a more useful answer than a generic 'what should I try'.
05Adding Fable 5 In Hermes
- Adding a new model to a multi-model harness can be done by describing the change in plain language to a coding agent rather than editing config by hand.
- Raising the reasoning-effort setting for a build task, versus a quick question, is worth doing deliberately rather than leaving on a default.
06Finding Skills You're Missing
- An onboarding step that asks for your hourly rate can turn abstract 'time saved' into a concrete dollar figure per skill, which makes ROI easier to judge.
- Asking an AI to scan 'everything', all chats, all tools, is explicitly token-expensive; scope the request when you don't need the full sweep.
07Level Two: One Brain
- The pitch for unified memory is 'never repeat yourself': a system with full context on your goals and past conversations should stop needing to be re-briefed.
- Linking a personal knowledge base, like an Obsidian vault, to more than one AI tool means both tools can answer from the same source of truth.
08Improving The Memory System
- A flat notes vault can be turned into a clustered, clickable topic map by asking an AI to categorize and cross-link it instead of organizing it by hand.
- Giving an AI 'free rein' to redesign a UI, rather than specifying every detail, can surface ideas you wouldn't have thought to ask for.
09Level Three: Ministry Of Agents
- Mixture-of-experts orchestration, asking several models the same question and synthesizing their answers, produces a more reliable response than trusting a single model's first answer.
- The technique works for everyday decisions, not just technical tasks; the demo used it to settle a completely non-technical debate question.
10Fable 5 As Orchestrator
- Swapping which model sits at the top of an orchestration setup is a configuration change, not a rebuild, worth revisiting as new models ship.
- Prompt caching is what makes running the same question through three separate models affordable; without it, adversarial multi-model review would multiply token cost.
- Paying for the orchestrator via a flat subscription while paying per-token for the sub-models through a router service blends a fixed cost with variable costs.
11Level Four: Dreaming Overnight
- An 'overnight review' feature works by pointing a model at a full day's logs, code activity, and integrations and asking it to propose improvements unattended.
- Reviewing autonomous suggestions in the morning, rather than approving them blind, keeps a human in the loop on unattended agent work.
12Level Five: A Voice You Can Talk To
- A voice interface layered on top of an existing dashboard turns a click-driven tool into something you can talk to for a quick status check.
- Framing the ask as 'build a voice system that can answer anything about the platform and click buttons' is specific enough for a coding agent to act on directly.
13Building The Voice Line / Live Status Report Demo
- A working voice assistant doesn't require a separate app; it can be built as a widget on an existing dashboard and wired to the same data.
Terms worth knowing.
- Agentic operating system
- A personal dashboard that treats every AI tool a person uses, coding agents, chat apps, memory stores, as one connected system instead of separate, isolated accounts.
- Ministry of Agents (mixture of experts)
- An orchestration pattern where one 'orchestrator' model sends the same question to several other models, compares their answers, and returns a single synthesized response.
- Orchestrator model
- The model sitting on top of a mixture-of-experts setup that reads every sub-model's answer and decides what to hand back to the user.
- Hermes Agent
- A multi-model chat client that lets a user switch between different AI models, GPT, Claude, DeepSeek, GLM, from inside one interface.
- OpenRouter
- A service that provides one API to call many different AI models from different providers, used here to run the sub-models behind a mixture-of-experts setup.
- Prompt caching
- Reusing a previously processed prompt so a system doesn't pay full token cost again each time multiple models are asked the same question.
Things they pointed at.
03:00toolHermes Agent
Lines you could clip.
00:10
“The most important thing that you're gonna build this year is an agentic operating system.”
11:27
“I called it ministry of agents. Some say, Jack, it's mixture of agents. I'm saying, hey, Ministry is way cooler.”
12:25
“Opus four point eight is no longer the king.”
15:21
“You go to sleep with your sleep cap on... and Fable five works overnight.”
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
00:00Fable five is the world's most powerful AI model, and the most important thing that you're gonna build this year is an agentic operating system. And in this video, I'm gonna show you exactly the new capabilities that Fable five unlocks across five levels that is gonna help you get more done, build faster than ever before, and even save money so you don't fall behind.
00:22And if you're new, I'm Jack Roberts. I built and sold my last tech startup with, like, a gazillion customers. Now I'm building my own AI startups, and I share here the stuff that actually works.
00:31So if you haven't already, grab that beautiful coffee, and let's dive straight in. So this is the AgenTik operating system, and in this video, I'm gonna show you the Fable five Agencik operating system across these five levels.
00:43But first of all, of course, when we're talking about Fable five, we're talking about a new class of model. It is the best, most powerful model we have available. There's graph, loads of lines moving in certain directions, and lines going up is a good thing as we all know.
00:57People are saying, Jack, I just say the word Fable five, and my credits disappear. I'm gonna help you with that in this video as well. Really important.
01:03But the bottom line is that Fable five is insanely powerful, and we're gonna use that to massively upgrade our Claude Code operating system. Now there's loads of reasons why you need a Claude Code agentic operating system. Three of the big ones is unified memory.
01:19And once I show you these, you're gonna understand it so we can build these incredible five levels together. Having a memory that connects to everything. So whether you are doing things on a phone call somewhere, whether you're using Hermes Agent or OpenClaw, it enables you to universalize your entire world in one location, which means that Hermes, it knows what's happening with Claude and Claude knows what's happening with Hermes.
01:42If you're on a different laptop, wherever you are, whatever the hell is going on, you have everything connected into one continuous system. That's why it's so so so important.
01:50The other thing is the fact that we can now have, and I'm gonna show you in this video as well, Claude Fable five dreaming for us. You can have the world's most powerful model thinking about how it can make your life better based on everything that you've spoken to about it.
02:03So it can do this for you autonomously. And the reason why it's better than just having that in Claude is that Claude is great, but it's one part of your life. You're using other softwares, other AIs, maybe using ChatGPT, maybe you're using your Hermes.
02:15You should be using Hermes to some extent. Everyone should be. The idea here is that no matter what's happening, we can get Claude to look over everything and create suggestions.
02:23And then we have one location to conduct our entire AI world. Now, this takes us very nicely onto what we'd call level one, which effectively is . And by giving it the wheel, I just mean adding Fable five to our actual operating system.
02:38And what we're gonna have it do first of all is to look at our entire world and find for skills that we should be using that we're not. So what I'm gonna do is I'm gonna shoot over to the Callcode operating system. Now, are many ways that we can access Fable five.
02:50At the moment, at the day at which you're watching this video, you may have one day left of using it inside the Callcode app. After that, it's all API access. Now, the harness that I'm gonna be using to access this is Hermes agent.
03:03You can just yap to it if you want to. Hermes though gives us an unbelievable additional amount of benefits that we get just by using it. So you can see down here, I've got this model selected here.
03:12Right? GPT 5.5, and this will change it.
03:15I've got ministry. I'm gonna show what that means later. It is so freaking powerful.
03:18GLM 5.2, Opus 4.8. I need to add in Claude Fable five. So how we're gonna do that?
03:23I'm gonna shoot over to Claude. Beautiful. Now I'm inside the Claude desktop app, which you need your $20 a month subscription to get access to.
03:30So I'm gonna come down here for example and I'm gonna be selecting Fable five including till July 7. Guys, they are tantalizingly teasing us here.
03:38So I don't need max. I'm just gonna need low for this one. And effectively, what I'm gonna do is to follow.
03:42Hey there. I want you to look at my ClaudeCode AgenTic operating system and in the Hermes section where I can select the models, I would like you to add for me Fable five and I would like you to add for me the effort dial and anything else that you think might be helpful or relevant in doing that.
04:00Go ahead and build that for me. Okay. Gonna send that one off.
04:02Now, actually, on second thoughts, what I'm probably gonna do here guys is come down and change this effort to probably a high if it's a build. It's gonna come down here, send that one off, and then we're gonna be able to rock and roll. And just like that, it's actually done its first version.
04:15So let's come down and take a look at this. You can see now we've got Claude Fable five that has appeared, and we can even select the strength level from low, medium, high, extra high, which is fantastic, and then we've got commands. Beautiful.
04:25So now I've loaded up, let's put it to the test. I'm gonna scroll down into my Hermes agent, click on GBT 5.5. By the way, you don't have to do this in Hermes.
04:32If Hermes is new to you, I'm gonna put a link down below. You can go ahead and grab the Hermes masterclass that I'm just releasing. It goes through foundations, setting up the Hermes, powering Hermes, and also the cool code master class that will take you from a complete beginner to do everything like building websites, power features, memory system, Hermes agent apps, you name it.
04:50I'll put it down below so you can become a pro in that. So we're here. I'm gonna go ahead and select Claude Faber five frontier model.
04:56And we can do extra high, high, medium. Let's go for something like medium. And I might say something like, hey there, dude.
05:03Could you tell me the last thing that I saved on my desktop? Alright. Also, which model is this?
05:08Okay. So we can double check that. I'm using Glider, do speech to text, so we can kick some butt, and that Hermes is having a good sync, and we can get some answers.
05:15And then immediately, we've got this last thing saved on your desktop with the five zero five Agentsia OS, and you're doing Core Framework five. Cool. So the first thing that I wanted to do now essentially is help us understand what skills could we be using that we're not currently using.
05:28So I'm gonna come down and drop the actual strength level down to low, which is gonna be fastest and light reasoning, and let's give it a command. Hey there, dude. I would like to look at all the data that you have available in this agentic operating system, all my call chats, my hermes chats, everything.
05:42And I want you to come back and suggest two skills that I could be using that I'm not currently using that would give me the greatest ROI based on all the information that you have for me. And if the skills to external sources, let me know what they are.
05:55I'm gonna send that one off and let this do his work. And what's really cool about this prompt guys is it does have access to everything. It has access to all the conversations I've had in Hermes, all the conversations I have in Claude.
06:06Because what you may not realize, but every chat you have in Claude, in anti gravity, in codex is all stored locally. So we can use all of this this kind of beautiful lake of data to help refine and understand what skills we have.
06:19Also, and this is really important to bear in mind here, that in the Agentic operating system, you're gonna have a list of all of your skills here. We even have ROI on the skills based on how much for example, in our onboarding system, which I have in the setup wizard here, you can see it. Because when you download this, it'll go through basically showing you exactly how to set up your OS, your name, your photo, etcetera, etcetera.
06:38And it actually identifies all the systems you have on your computer. And one of the cool things that it does, and this is really relevant for your skull section, like, even identifies memories and stuff like that.
06:48If I can come down here, you set what your hourly rate is. And then based on that, we can actually calculate how much your skills are saving you in dollars every single day. It's so cool.
06:58This onboarding itself took me like three or four hours to build. It's so freaking set. And now we come back, Hermes is gonna calculate, look at all the data, and give us some suggestions.
07:06And just like that, Fable five has come back with some pretty incredible suggestions. I blurred them out just for privacy sake, but this is nuts. One of the things I will say though is if you are getting Fable five to look at everything that can be token hungry, so just bear that in mind when you're asking questions and what you're asking it to do.
07:24But this is freaking amazing. It's gone through my skills and it's found ways of how I can improve things based on the models that I'm using and all the different systems. Now, this does take us very nicely onto the second level of this, which essentially is this idea of one brain.
07:38And let me show you what I mean. Now, is the most powerful feature of having an agentic operating system. And let me explain exactly what I mean by that.
07:46So this is the brain, and this is the memory core. This represents the entire memory. Now the idea of the memory system here is that we want ultimate context because if AI had infinite context about you, what your business goals were, what your financial numbers were, what you spoke to about yesterday and a year ago and the meeting you had with Billy Bob and Susie Magoo two days ago, they had all that information.
08:08You would never need to repeat yourself and you would always get the best insights from it. So one of the things that we do in the Jentic operating system is you want to link it with your Obsidian bot. You wanna link it with every one of your ClawCode systems and you also wanna link it with everything that's happening with your AI personal assistants like a Hermes agent.
08:26So you come down, you can see, you can connect this all together. Now one of the things that I want to do here is I actually want to go to essentially Fable five and say, hey, is my memory system. What I want you to do is think of way how we could actually improve this and make it more interactive and maybe even easier to use so I can elevate the entire system.
08:45Because one of the great things about the Obsidian bot I'll put a link on screen if you're not familiar with what that is or how that works so you can get those Kapathi principles. But the idea is that you build a local LLM Wiki that you can access with any information that you want to.
09:00And the key thing here is that both Claude and Hermes can access the same database, and whenever you ask questions to it, it can draw from that Wiki article. So for example, I can come to my agentic OS and say, hey, shoot over to my Obsidian Wiki and just remind me of three best principles when making YouTube videos in terms of intros, how I make great intros.
09:19Just an example, in reality, you could have anything. And then just like that, we can see we've got this back here. Start with payoff, not the setup, open loop in the first ten seconds, cut everything but delays trust.
09:28This is very cool. It's got this all from my Obsidian Wiki. And the cool thing I've got here as well is an overview of my token spend, so I know exactly where I'm at in this conversation.
09:36So we're gonna give a prompt to Fable to actually improve our memory system. This is the prompt that we're gonna give him. Hey there.
09:41I would like to go ahead and improve the memory system within my Called Code Agencik operating system. It is connected to my Obsidian or my Hermes chats or my Called Code chats. I want you to find a way to make it interactive so that I can click through various different aspects, bring that knowledge out.
09:57I want you to be creative and add a section that will help me discover new facets about my knowledge and maybe we could categorize it and cluster it based on what the topics are. I give you free rein. I want you to build something interesting to elevate the memory system inside Magenic OS.
10:13And then we'll just send this off. And so now Fable's actually completed. Let's go now and see what it's done for us here.
10:18So this is really interesting. So let me just pull this over so you can see a little bit better. Cool.
10:22So what I've got down here then, I've got all of my different wikis, my Obsidian wiki, my YouTube vault, and what basically Fable has done is organized it based on topics. So we have Callaway, Ed Lawrence, cool guy that lives out in Dubai, talks a lot about YouTube strategy. Patty Galloway hooks YouTube.
10:37So it's these are the thing that the threads that come up more often. I can expand and see 15 more topics. Really freaking cool.
10:43So if I come down and I say, I wanna learn more about retention. So I click on retention, and I have these different concepts and frameworks. Very cool.
10:50I can click into that further and get more information. That links to intros, and now I'm back at intros, and I've got business YouTube.
10:57I can click on this, and I can keep clicking around. By the way, I just gave you the prompt, and it went ahead and did all of this. I've even got a search bar.
11:04I can search for things if I go for I wanna learn about hogs. And look at this dynamically, guys.
11:09This is ridiculous. This is insane. This is a good retention, and now it's like my own personal kind of database where I can look through all my memories.
11:16You can see how Fable five is taking the kind of Agenetic memory system to a completely new level. But this takes us on to an even more powerful level with the called Fable Agenetic OS, and that is the ministry. I called it ministry of agents.
11:29Some say, Jack, it's mixture of agents. I'm saying, hey, Ministry is way cooler. What do I mean by that?
11:33Well, let me show you exactly what it's about because this is really, really important to understand. So if I come over to Hermes Agent on the left hand side. Now, this is a feature that Hermes released, which is cool and it's in a few different models.
11:44You can do it in your Agenic OS. If you come over to this Ministry of Experts model here, the idea of this is that you have an orchestrated agent. Right?
11:52You have a model that is looking after everything. And what it does when you ask it any kind of question, building code, it could be about your life, like what coffee shop should I go to? Because we know, guys.
12:02That's the most important question these days. What it will do is actually ask three separate models. So I've got here, for example, DeepSig v four Pro, GLM 5.2, and ChatActivity 5.5, which is extremely token efficient by the way.
12:18And this whole system saves you loads of tokens. And it will look at all the answers, find out what the best ones are, and then it will orchestrate and give you a way better response. So So what we're gonna do now is actually put in Fable five in the orchestrated role because Opus four point eight is no longer the king.
12:33So we're simply gonna remove this guy right here, and then we're gonna find Fable five. And since we can't see Fable five on the left hand side, we're gonna go ahead and add it in by shooting over to Claude. So then come back over and explain to Claude that you wanna add it in, if you haven't done that yet, and literally, you should add it into the Ministry of Experts.
12:47Of course, I've got it in mind. Beautiful. So we come back over to the operating system, come over to Ministry Agents, and what we're gonna do now is remove 4.8 and then we should be able to add in beautiful Fable five.
12:58And here it is, gonna grab him and bring him over. So he's now gonna be the orchestrator, which is fantastic. And now we're just gonna save this to the computer.
13:05If you haven't actually added this before, all you do is let you come down, copy this, and let's drop it in the chat window, which I will do up here just so you get a bit of visibility of how it works. And one of cool reasons that we use Hermes for this chat, by the way, is because it actually caches the prompt for all of them.
13:20So ordinarily, if you were to do what we call an adversarial review, in other words, you're gonna have Deepsea gives an answer, GLM gives an answer, GPT 5.5 gives an answer. They would all we'd have to pay for tokens on all of them. But we actually use a cash, which is gonna save us so much in tokens.
13:36And obviously, we're using OpenRooter for DeepSeek and GLM. And ChatGPT 5.5, we're using our OAuth, which is freaking amazing, which means you actually pay pennies.
13:46It's so cheap. I'll put a link to the full system if you've never used this Ministry of Experts, but using Fable five in it is a crazy game changer. And as you can see, it's already complete.
13:54It already actually had the correct information, which is excellent. And then what I'm gonna do is just come down to the selector here, and let's go ahead and actually select ministry, which is just a quick way to do the three experts. And then I'm gonna say, hey there.
14:05Cool. I would like you to go ahead and tell me which is the best country on the planet to look for if you're in your early thirties with a beautiful spring and spiral and you love coffee. Alright?
14:15And then basically, this is gonna go ahead and debate with all of the Ministry of Experts. And then essentially, it's come back. It's done the big debate.
14:21We've had a cafe at Alchemist, and the verdict is Netherlands, specifically ETricht, which is fascinating. So if you're from The Netherlands, let me know down below in the comments. Beautiful.
14:30Just like that, we get our prompt back from Fable five that's had all three models that have debated it, which is fantastic. GBT 5.5, GLN 5.2, and DeepSeek v four Pro. So essentially, Fable five is orchestrating, and then we've got these sub agents underneath it that it's calling, and we're getting the best out of all of those individual agents.
14:47And this takes us very nicely onto level four for our AgenTic operating system with Fable five. So this is essentially the dream. Right?
14:54The dream is that you , don't know who's doing that right now, and Fable five works overnight. And the idea is it works on your problems. It thinks about how can we improve things, it spins up sub agents, it does wonderful things.
15:07Now, we reflect this in our operating system. And this is one of the things that I highly, highly recommend that you get. And it's this idea that effectively based on all this information, right, code activity, anti gravity, memory files, automations, integration skills, chat logs, everything on Hermes comes back and dreams for you.
15:24So what we're gonna do, effectively, is give it a prompt like this, which is I'm gonna shoot over to Claude and I'm gonna say something like, hey there, I'd like to go ahead and look at the dreaming functionality inside the operating system and I'd like you to go ahead and improve that in some way. Find ways how we can make that dreaming functionality better.
15:39How can you look at all the existing data that we have and take it to a completely new level so I get better insights leveraging these models. And it's absolutely fascinating the levels of insights. So Fabrify did a full reroute of the entire dreaming system, how it works, where it gets the data from, and the suggestions and how it comes back.
15:56And what's also really fascinating is the model architecture behind it. Which models are doing the dreaming? What are the data sources?
16:03How does it come together? And just gives you this beautiful overview, this beautiful UI you can look at. You can come down.
16:08You can copy this right now if you just wanna get this done immediately. You can take it off as you go. It is such a powerful feature.
16:14Having a super intelligent AI to review all of your stuff while you're sleeping, working for you, I think it is an incredible benefit. And then this takes us very nicely onto level five, which effectively is taking the brain and giving it the ability to follow us around in the Agenci OS. So this is really cool.
16:31But wouldn't it be really cool if we could actually talk to the AgenTic operating system or have our own some people call it Java, some people call it Maverick, whatever your cool name is for it. But I think it would be incredible to have an actual AI system that could follow us around and basically we could ask questions to.
16:48So looking at it a prompt like this. Now because we already have in the Hermes agent here, for example, if I come down to voice, we have the system. Let me go ahead and set this one up.
16:55So now I have the full voice intelligence. Right? But I come over down here, I go over to something like the classic and I bring up sonar plasma.
17:01I can chat with this, which is really cool. So I've come down here and begin. Hey, Hermes.
17:05What day is it today? Hello there. Let me check that for you.
17:09Just a moment. Okay. So this is a voice interface and it's connected to Hermes.
17:13But what I wanna do is I kinda wanna get a design that looks a little bit like maybe a bit like this that can actually chat to it. So what we're gonna do is shoot straight over to Cloud Code.
17:22And so then we get that prompt like this. Hey there. I want you to build for me an interactive voice operating system on top of the Cloud Code OS.
17:29It should be able to answer any questions about the platform. Just make it interactive and find as many ways as possible to make it as useful as it can be. It should know loads of data.
17:38It should be able to click buttons, move around, do various different things. I want it to be as beneficial to me in my operating system as possible. And then literally, we're gonna go ahead and send that one off to Fable.
17:48Beautiful. So now Fable five has completed it, and it looks like we activated it by selecting this thing here at the top. That will turn it off and on.
17:55And then if I open this up, we have voice lines. So let's click on this. I'm pretty impressed at the actual design it's got.
18:00To have this kind of thing at the bottom, right, is really cool. Click on voice line and we can start a conversation. Hey there.
18:06How's it going? Hey there. I'm doing really well.
18:09Thanks. How about you? It's good.
18:11It's good. Um, give me a status report, please. Sure thing.
18:15Here's your status.
18:16You've sent over 22,000 user messages so far and I've replied about 39.
18:22Beautiful. So you can get the idea now. We can ask you questions, and I deal with this is that we can pretty much build out the functionality now.
18:27So from this, I can ask questions about, hey. Tell me, you know, as I say, for example, I'm looking at a skill. I might say, tell me more about that.
18:34Can you help me improve this skill? And the really cool thing is it can actually improve the dashboard itself and code on your desktop because, hey, change it from blue to green and you're doing it live in the actual dashboard. Now using this incredible operating system is one thing.
18:49But there are actually five things that we need to do with Claude Fable five. They're gonna take your entire business and personal life to a new level, and we can learn that by watching this video right here.
The Hook
The bait, then the rug-pull.
Jack Roberts opens with a blunt claim: whatever else you build with AI this year, the highest-leverage project is a personal agentic operating system. He spends the next 19 minutes proving it, wiring the newly-released Claude Fable 5 into five stacked capabilities, routing, memory, multi-model debate, overnight autonomy, and a talking voice interface, using nothing but plain-language prompts to a coding agent.
Named ideas worth stealing.
00:36list
The Five Levels of an Agentic OS
- Level 1: It Takes The Wheel (routing/scheduling)
- Level 2: One Brain (unified memory)
- Level 3: The Ministry (mixture of experts)
- Level 4: It Dreams (overnight autonomous work)
- Level 5: It Speaks (voice interface)
A five-stage mental model for turning a single frontier model subscription into a full personal operating system, each level adding a new capability on top of the last.
Steal forstructuring any AI-tooling build video or internal roadmap into levels of increasing capability
How they asked for the click.
VERBAL ASK
18:42next-video
“there are actually five things that we need to do with Claude Fable five... we can learn that by watching this video right here”
soft pointer to a follow-up video over the final talking-head close, no explicit subscribe ask
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.
