Big Idea

The argument in one line.

Fable 5.2 is most valuable as a planning and task-writing model, because well-specified tasks let a cheaper model like Opus do the building at lower cost and with less mental effort from you.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use Claude Code daily and want a concrete rule for when to spend on the top model and when to switch to a cheaper one.
You start projects from a vague idea and want to see how a planning session turns it into a tracked task board.
You are watching token usage and want a workflow that gets frontier-level planning without frontier-level bills.

SKIP IF…

You want benchmarks, side-by-side evals or 3D demos; the video deliberately skips them.
You need a finished product; the agent harness is only planned and started, not completed, in this video.

TL;DR

The full version, fast.

The creator calls Fable 5.2 the best AI model yet because it is smarter, more concise, faster, cheaper and more token efficient, which lowers the cognitive load of working with it. It asks multiple-choice questions, challenges weak ideas and reads the intent behind a prompt instead of the literal words. His workflow has three stages: brainstorm and plan with Fable, have it write detailed tasks into a Notion board, then switch to Opus at medium effort to execute them. Good planning is what lets cheaper models build well.

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 →

Chapters

Where the time goes.

00:00 – 00:58

01 · Cold open and promise

Claims Fable 5.2 is the greatest model and promises a master workflow that also saves tokens.

00:58 – 02:43

02 · Why it is better

More intelligent, far more concise, as fast as Opus, cheaper and more token efficient.

02:43 – 04:03

03 · Lighter cognitive load

Faster, concise and interactive answers mean less brain power spent for the same or better work.

04:03 – 05:57

04 · Setting up the project

Desktop app, high effort, and a loose idea for a personal project-manager agent harness.

05:57 – 08:40

05 · Intent reading and multiple choice

The model reads the goal behind the prompt and guides choices with multiple-choice questions.

08:40 – 10:28

06 · Challenge and plan

It pushes back on the idea, asks scoping questions, then produces a phased plan.

10:29 – 12:20

07 · Tasks into Notion

Fable builds a Kanban board of detailed tasks with milestones, acceptance criteria and sizes.

12:20 – 13:10

08 · Hand off to Opus

Switch to Opus at medium effort and start building from the finished task board.

13:10 – 13:24

09 · Sign-off

Next build video, subscribe request and community plug.

Atomic Insights

Lines worth screenshotting.

A frontier model that answers as concisely as a mid-tier one is a different product: the intelligence goes into the answer, not the length.
Lower cognitive load is the real upgrade: same or better work with less of your own brain power spent steering the model.
Multiple-choice questions beat open-ended ones during planning because you pick a path instead of inventing the next step.
A good collaborator model reads the intent behind your prompt, pushes back, and recommends a better route than the one you described.
Spend the expensive model on planning and task-writing, then let a cheaper model execute the well-specified tasks.
A task board with milestones, acceptance criteria, size and done-when fields turns a planning session into something any model can execute.
You do not need a precise prompt to start: a general idea is enough if the model interviews you well.
Notion's free tier works for a task board; Linear is stronger but limits the number of boards on its $20 plan.
Running execution at medium effort instead of high is a simple lever to save usage once the plan is solid.

Takeaway

Spend the smart model on planning

WHAT TO LEARN

The expensive model earns its price in the planning phase; once tasks are specific, a cheaper model can execute them just as well.

Start with a loose idea and let the model interview you, since good questions beat a perfectly written prompt.
Prefer models and prompts that ask multiple-choice questions, because choosing a path costs far less mental effort than inventing one.
Invite pushback on your idea early; a collaborator that challenges the premise saves you from building the wrong thing.
Turn the plan into a board with milestones, acceptance criteria and sizes so the work is explicit before any code is written.
Switch to a cheaper model and a lower effort setting for execution once the tasks are well specified.
Judge a new model by how much effort it saves you, not only by benchmarks or flashy demos.

Glossary

Terms worth knowing.

Agent harness: The software layer that runs AI agents, gives them tools and lets them spawn workers, so you manage one interface instead of many separate sessions.
Effort setting: A Claude Code control for how much reasoning the model spends per turn. Higher effort costs more usage; medium is the recommended default.
Headless run: Running a coding CLI without its interactive interface so another program can launch it, send a task and collect the result.
Acceptance criteria: A plain statement of what must be true for a task to count as finished, which lets a cheaper model work without guessing.

Resources

Things they pointed at.

04:10toolClaude Code desktop app

04:15toolHerder

10:40toolNotion

10:50toolLinear

13:00productVibe Coding Academy

Quotables

Lines you could clip.

01:02

“They say intelligence for human beings is your ability to explain something very complex, but in a very simple way.”

Reframes model intelligence as concision with a human analogy.→ newsletter pull-quote↗ Tweet quote

06:59

“Other models are kind of dumb workers that just do whatever you say. This model is the intelligent collaborator.”

Clean contrast, no setup needed.→ TikTok hook↗ Tweet quote

11:05

“If you have really good intelligence plan out the app and build out the tasks, then you don't need the best intelligence in the world to actually build it.”

The core cost-saving insight in one sentence.→ IG reel cold open↗ Tweet quote

The Script

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.

Let's just get it out of the way. Claude Fable 5 .2 is the greatest AI model of all time and is not particularly close. I have been testing it in early access for a little bit now and my mind is completely blown.

What I'm going to go over in this video is not only what makes it so incredible and what makes it so great, I'm going to show you a master workflow of how to get the absolute most out of this model and also save tons of tokens and usage. My life has been Fable 5. 5 .2 for a little bit now, and I cannot wait to share all the secrets, tips, and tricks I have for you in just a second.

But let's start with what makes this model so unbelievably great. This first part shouldn't surprise you at all. It's more intelligent.

It's more intelligent than any other model out there. Opus, Fable, Astral, all of them, more intelligent. That's what you expect.

That's kind of table stakes, right? You release a new model, it's got to be more intelligent. But here's where things kind of surprised me.

Number one is it is way more concise. second, I'm going to go into cloud code. I'm going to show you how to use it.

I'm going to show you how to get the most out of it, show you a master workflow. But what really surprised me is that it is just way more concise. A lot of the times when you talk to these frontier level models, whether it's GPT -6 Astra on the pro setting, or it's like Fable 5, like the original one, the responses you get from these models are significantly longer, more detailed, more complex than the responses you'd get from a Sonnet.

an opus, a soul. That's not the case with Fable 5 .2 anymore though. You get pretty much the same conciseness you get from like a sonnet level model, but you just get so much more intelligence in that concise response.

They say intelligence for human beings is your ability to explain something very complex, but in a very simple way. This is kind of the first AI model that reflects that human intelligence. Explain really complex things, but in really concise, simple ways.

Before, I think intelligence for these models were explaining things really complex, but in really complex ways. That's out the door. It says really intelligent things, but in really concise manners.

I hated sitting there for like Fable 5 and reading these long, detailed responses. As smart as they were, they were just too long and hard to read. Way more concise.

It's significantly faster. It's just as fast as Opus. It really is unbelievable.

to be some sort of breakthrough at Anthropic over the last few months where like their speed and intelligence both have exploded it is significantly faster just as good as Opus it's cheaper they brought the price down it's way more token efficient it's the trifecta of amazing models Faster, cheaper, smarter, it's all that incredible.

But here's the most important part and here's what I'm going to demo to you in just a second when we get into Claude Code. This is the most important part of them all. All of these improvements plus some more surprise ones I'm going to show you has resulted in a lightening of my cognitive load.

What does that mean? It lightens your cognitive load. When you use this model, Fable 5 .2, what you're going to experience is you get the same amount of...

work, if not better work done, take significantly less brain power. That's because of the speed. That's because of the conciseness, but it's also the way it interacts with you.

It asks you a lot of multiple choice questions. I'm going to show this to you in a second, but what you're going to experience, what I think makes this model so unbelievably good is. Using it and getting great results doesn't take as much brain power So you can use that brain power for other things you can use that brain power for things you love to do I'm gonna show you all that in a second, but that is why this model is incredible.

I could show you benchmarks I can show you cool 3d demos, but I mean, that's what every other channel I'm playing earth's doing So I want to cover the things that truly matter. Alright, so let me show you the best way to use fable 5 .2 for this We're gonna build a real project in real time. I'm gonna show you how you use it to build, show you all the tips and tricks, show you how I got the most out of using it recently in early access.

You'll walk away from this a Fable 5 .2 master and see some of my workflow here. So I'm in Cloud Code. I'm in the desktop app.

I've been using the CLI more recently because I use Herder, which allows me to put all my agents in one place. But for this, we're going to use desktop app. I think it's a little bit more visual.

I'm going to go effort high, even though mediums are recommended. I like to use a little bit more juice than is necessary. Also important thing to know, they added a bunch of resets.

So if you've been using Claude code for more than a year, you get resets. So I'm leaning a little bit more toward using higher intelligence here. Let's get into the first step of using Claude Fable 5 .2 really, really well and effectively is to use it big time in the planning phase, right?

The way I look at it, you got kind of got the brainstorming phase, the planning phase and the execution phase. I think Fable 5 .2 is incredible at the brainstorming and planning side of the face. So let's do this.

Let me show you how this works. First, I'm going to go in with a general idea for a project I want to build. I've always wanted to build my own agent harness.

So let's build our own agent harness here. So you come in, you open Fable 5 .2 and you put in a general idea. It doesn't need to be precise.

It doesn't need to be super detailed. Just put in a general idea. All right.

So here's the idea I put in. If you want to build this alongside of me, I will put the prompt down below. Copy and paste it and work alongside me here.

I want to build my own agent harness in the style of Hermes Grokbot. For this agent harness, I want to be more of a project manager. As in, I talk to one agent and it acts as a project manager for everything in my life.

I can tell it about any project I'm working on and it spins up sub -agents and other workers to do the work. This makes it so I don't have to manage a bunch of threads. It should be beautiful and allow me to monitor all agents and subtasks.

It should support my Claude and OpenAI subscriptions. right? So key part one of using Fable 5 .2, the first stage of kind of brainstorming and perfecting the idea and the concept.

Fable 5 .2 is so smart, so intelligent. It's going to actually help us. figure out what we should be building, what the perfect idea is here.

Another reason why this is so good, this model, and why I believe it's different than a lot of other models out there, it does an incredible job at getting to the core of what you're trying to do. So a lot of models, and I'll hit enter on this as it begins, a lot of models you talk with, they take your prompt and they take it very literally and they just build it out.

The different... here with ClaudeFable 5 .2 I noticed is it reads your prompt sure but what it does really well is it gets to the core of what you're asking it it gets to really understand the intentions of what you're saying.

And then it responds to that in the best, most intelligent way possible. Right. And that's a nuance.

I think that's very nuanced, but the important thing here is like other models are kind of dumb workers that just do whatever you say. This model is like the intelligent collaborator that understands fundamentally what you're trying to do and then recommend the best possible way to do it. Even if you haven't thought of it yourself, that's why this model is so good.

And that's what you're about. to see right here based on its response to this prompt. Building this as a local app, 1pm agent, with tools to create projects, tasks, and spawn workers, where workers are headless, cloud code, and codec CLI runs, which is how it rides on your subscription set of API keys, checking your tooling first.

You'll see a lot of this when talking to Fable 5 .2, which is checklists of everything it's doing, the background tasks it's running, and what you'll see in just a second, this is what I'm guessing, and I'm probably going to be right here, is it's going to ask me a few multiple choice questions. I really like this. about this model is that a lot of its responses are multiple choice questions.

No matter what phase you're in, planning, building, whatever. It tries to ask you multiple choice questions as much as it can. And this is one of the key ways it lightens your cognitive load.

Instead of constantly leaving you with open choice questions, open choice answers, we have to sit there and truly think for a while. It tries to do its best in guiding you down a path. And you don't have to go down that path.

You don't have to do what it recommends, but because it can understand what you're trying to do so well, it is able to kind of guide you down a path, give you multiple choice questions and make building and researching and planning so much easier. This is what, again, what I talk about, lightening the cognitive load.

If everything's a multiple choice question, then you don't have to think quite as much, which means you can use that mental energy for building other things and doing other things you enjoy. All right, so here we go. This is where I love.

this model is it's coming back to me where I challenge it. So it's challenging the idea. The bottleneck is your attention, not agent count.

10 workers produce 10 things review. I don't want to manage threads and let me monitor all agents pull in opposite directions. The hero feature should be a short queue of decisions.

Only you can make the full agent views for debugging and trust. And I scroll down a bit and we get these multiple choice questions. Which wedge should chief nail first work projects, code research and writing life admin, both in the start.

Let's start with work. projects. I'll do next there.

Where do you want to talk to chief web app and phones? I've got permission control telegram. I message.

Yep. We'll do web app and phone. Where should projects and tasks live?

Let's start with chief's own database. Who's in this again. This is what's great.

A lot of models. When you talk to it, it kind of gives you some questions, some choices. But it's very open -ended where you have to come up with what's next.

Fable 5 .2, again, feels like it's really aiming to lighten your cognitive load a ton. So it gives you this kind of handhold path you can go down. If I want to, I can say other and put other things in.

But it just makes it so nice when it's like these really well laid out multiple choice questions. Who is chief for? Just me.

Let's start with just me. Maybe I'll open, hey, do you want me to open sources? If we build something cooler, let me know down below.

Let's do just me. Okay. So now it's going to do this.

Now it's going to start building this plan for us. All right. And now it is giving us a plan.

So phase one plan where we are, lock the brief, write the spec, create the task. Then there's the building phase, improve it. Here's how you use Fable 5 .2 really well.

We're going to do one last thing with this model. Then actually we're going to switch off of it afterwards. So here's the first we're going to do.

Please. Build out all the detailed tasks in a new Notion. board that we can track.

And I'm going to hit enter on that. And now I have, cause I have a notion connector in here. It's going to build a new board.

And this is what I recommend for you, whether you're using linear or notion, I'm doing this in notion because notion is free, but linear spectacular as well. It just has like a limited, like four boards you can do. If you're on the $20 plan, have it now Fable 5 .2 go and actually build out the detailed tasks.

So it's going to make like a Kanban board of all the detailed tasks, because here's the thing. If you have really good intelligence, plan out the app, build out the tasks, then you don't need the best intelligence in the world to actually build the app out. You can have cheaper intelligence like Opus or even Sonnet do the building, but we're going to have Fable 5 .2, which is the smartest model ever.

actually do the planning out and build us out all the tasks, then we'll hand it to a cheaper model to do the work and we'll get just as good, if not better results. It's asking me some follow -up questions. It asked me how I should reach me on my phone, telegram, where it should run on my Mac.

So it's asking me more follow -up questions because everything it tries to make is easy for you to answer as humanly possible, which is so nice. And if we take a look here at Notion, boom, look at all of these tasks it put in and you can see it has. milestones, status, done when, area, acceptance criteria, size, all of the hundreds of tasks here, all built by Fable.

And now here's the key. Now we can move over to cheaper model because these tasks are so well built out. Going back into Cloud Code, I'm going to switch to Opus 5 .5.

We're going to switch the model. In fact, I'm not even going to go high. I'm going to go medium.

And now I can say, you can begin working on this. and I'm going to hit enter and it's good to go. Now Opus is going to take all those tasks that Fable built out.

This is the thing with AI. As long as you have really good intelligence doing the planning, building the tasks, everything like that, you can use the cheaper intelligence to actually do the execution. That's the way to get the most out of Fable 5 .2.

Use Fable for the planning. for the tasks, store those tasks somewhere like Notion and Linear, then use cheaper models like Opus. And you can even go Sonnet for this and build the app out.

That's how you get the most out of this incredible model. In the next build video we do, we're gonna continue building this. So make sure to subscribe, turn on notifications.

If people like this and they enjoy it enough, we'll open source this and get this out into the world. Let me know. Also make sure you join the Vibe Coding Academy.

Link down below for that. We have Claude Codemaster classes. Chad GPT masterclasses, Hermes masterclasses, Grokbot masterclasses.

Hit the link down below for this. You'll learn so much more. I hang out there every week so you can ask me questions one -on -one.

I hope this was helpful. I'll see you in the next video.

The Hook