Transcription
Claude Fable 5 is officially back today, and I'm so excited because it literally felt like magic using this model for the few days that we had access to it. I felt like I could do what used to take me weeks in just a few days. So, I'm so excited to reintegrate this back into my business. So, to make sure you're ready, I'm going to be showing you the four things that you need to do the moment you get access to the model. These are the things that you must do if you want to extract the maximum value out of this model, because Claude Fable 5 is very different from Opus 4.8. How to get the most out of Fable is very different from Opus. So, if you do these things, you're going to absolutely crush it.
But before we get into that, there are a couple of caveats. Firstly, they're going to limit the extent you can code on the model. So, they are going to automatically route some coding back to Opus 4.8. This is likely due to the US government's inquisition. And secondly, you'll only be able to use Claude Fable up to 50% of your Anthropic plan. So, you'll have even less usage limits than last time. And last time, people were running into lots of issues. So, that is definitely a constraint, but I do feel like the strategies I'll be teaching you in today's video are going to actually help you use up less of your usage limit. And then I'll also be dropping a video in the future, which is a follow-up to my last video on how to actually save tokens using Claude, because that is something you're going to have to be cognizant of when you're using this model.
But, let's get straight into things here. The real beauty of Fable and how you got to think about Fable is it's so good with autonomous work. Claude Fable is the smartest Claude model yet when it comes to thorough proactive testing of its own work, when it comes to automatic agentic workflows, when it comes to looping any autonomous agentic flow is so much better on Fable than it was on 4.8. And that's why it feels like magic using Claude Fable because when you're using Fable, things are just happening autonomously if you set it up right, which I'll teach you how to do today. So with Opus 4.8, it still feels like there's lots of manual intervention. It still doesn't feel like it's great at checking its own work. When I was using Fable, it really felt like you could hand over the goal, hand over the system, and it would be able to execute and delegate an outcome with reoccurring loops, with skills, with research overnight. And I'm going to walk you through how to set all of this up in today's video. A lot of you are probably going to be watching this by the time Fable's out, and that's fine. This video is probably going to be relevant in weeks or months because these are the principles that Anthropic is literally telling you in the Fable documents that you need to be using if you want to extract the maximum value out of Fable.
So, the first thing that you are going to need to master with Fable is loops. These were popularized when Faber launched because people realized that it was just much better at iterating on loops than the previous models. Well, what are loops? Well, loops are basically the way that you can get Claude to do a task autonomously and repetitively from the very beginning stage of finding the work to handing it to an agent to checking the result of that agent. So, the final product to recognizing and recording what happened to reviewing that work and then deciding the next move. So loops enable you to start an entire process from start to finish and then recursively complete that process again and again because Claude is much better at checking its own work now. This is now super effective and it's probably the best and most efficient way to use Claude Fable 5, and which I'll show you today. If you configure your context and your skills in the right way, the agent will remember and self-improve over time based on the exact feedback that you give it.
So let me just give you some examples. You could essentially run a developer loop which develops an application, checks the work of the agents and automatically delegates more work once it's finished with one particular goal. So you have the goal prompt which I'll speak about today, which essentially is the way to get Claude to do something until it's done. So you know, research until you can answer these five questions or build this application. You're telling it to do a goal, it'll do it, and then it will stop. Loops. So, if you use `/loop`, it can complete a goal, but then it can loop back around and it can keep going on an interval basis. For example, if you're researching markets every morning at 6:00 a.m., it can send you a market brief and it can research throughout the day. Instead of hitting a wall like it would on goal and just, you know, stopping. The loop keeps it going. So, it can keep iterating, improving, reviewing its own work, and sending it to you every day. Or for example, you could create a loop which scans your email every day and every 30 minutes it could flag an email that actually needs your attention. So `/goal` is still powerful because it tells Claude what you are trying to achieve. But `/loop` is even more powerful because it automatically recursively completes the task, which is so important for autonomous work, especially if you're going to start to set up agents to run your personal life and also help in a business context.
And this is personally really important for me, someone that's deep in the media weeds and runs a marketing agency. We do a lot of creative work, and Fable is fantastic at creative work. So running loops enables me to constantly generate content ideas, constantly generate ad creatives, and constantly iterate on what's working based on real data that we plug in from YouTube, that we plug in from Instagram. So it's self-improving and autonomous. So you know, if I step out for the day or if I go to sleep, it's still learning. It's still improving. And you, of course, I can check in and give it feedback, but it's not just reliant on like me prompting and then receiving an answer to a certain response. It basically works by itself, which is really what you want out of an AI model. And for those that are interested, you can pause the video right now and see the full looping process, how it actually works, and how you can actually run a loop. But I'm actually going to show you a live demonstration right now so you really understand. And I'll probably do a dedicated video on this later 'cause it's a rabbit hole, but hopefully you understand after seeing this.
So I'm going to set it a goal. So firstly, I want to tell it what it's aiming towards of researching the latest developments in AI. So this is an example of a prompt that would help me research developments for X ideas for content ideas. And so I'm in the loop. You know, I'm launching product soon which requires me to be really in the loop and understand what's happening. So this is going to build an intelligence brief and under the file, it's going to give me eight solid distinct entries. Then that's the goal. I'm going to use `/loop`. So every 10 seconds. So on an iterative basis up to eight times, it finds one genuinely new development. So it's tracking the news 24/7. It appends a timestamped entry to the brief. It skips anything in the file and it puts the new entry after each pass. So if I enter this, you'll see the loop in action. This is basically how to construct a loop. You have a goal and then you tell it on what basis you want it to, you know, do that work reoccurringly. So this is a small example of, you know, a market intelligence loop which you can keep running every day, right? If you want to build like a market scanner, if you know you're researching stocks, if you want to build a scanner for, you know, content.
But some more ideas that you could do with this, for example, if if you were vibe coding an application, which Fable is also very good at, you can have the loop check its own work and automatically like while you're sleeping create new features overnight, which of course you can step in at any point and approve, but it will automatically be able to review and approve its own work and work out where to go next. So you kind of create this automatic development path. Obviously you can step in and course correct so you know that it's like developing 24/7. That's obviously going to burn a lot of tokens, right? Doing an app versus scanning the market is very different. And of course you can stack, you know, multi-modular uh systems. So for example, I could scan on Sonnet. By the way, the new Sonnet came out, it's almost as good as Opus 4.8, but it's much cheaper. You could do like market scanning there and then you could have Fable reviewing the work using the 108010 approach that I've spoke about on the channel. So 10% the first 10% you want your smartest model executing. So Fable, 80% the grunt work you can use, you know, a model like Sonnet or something cheaper, even Opus. And then the last 10% you want to go back to Fable to review the work, and then that will kick off the next loop. So that's a bit more advanced. I'll be talking about that in a future video, but I think it's uh good for you guys to know.
So you can see here it's working in real time. It's done pass one, pass two, pass three, and it will automatically keep following the loop. You don't need to actually use `/loop`. But if you do use the command, then it's obviously going to use the bundled information under that command. So it's helpful. Okay, so that's Claude loops. But loops are just a command for it to do something over and over again. It doesn't have taste. It's not informed. It doesn't have data. So how do you give it the taste, the data, and the feedback loop so it can execute better? Well, that is where Claude's skills come in. Probably the most important thing that you need to master alongside loops to get the most out of Fable 5, because they literally tell you in the documents that Fable 5 is a thorough proactive model which tests its own work. So, it's actually the best placed model to iterate upon skills and build up really strong reusable skills.
So, the way to think about skills is it's a recipe that you teach once and it reuses forever. For example, I can create and I have created a video planning skill which helps me plan and research for videos like this. And every single time I film a video, I take the engagement data and I give it my own thoughts on how it's performed. I give it that feedback. It updates the video production skill. So, every single time I workshop a video with Claude, the video production skill gets updated and the next time I go to create a video, it's a little bit better. And then the next time I go to create a video, it's a little bit better. Think about what you could do for writing. If you were starting an X account for example, you could train it on a bunch of data. You could take, you know, thousands of creators. You could run a scraper. You could give all of these tweets to a Claude skill. You could be like, "Hey, mimic this tone of voice and mimic these thought patterns and ideas." And every single time it writes a tweet for you, you can give it feedback on that tweet, what it did well, what it did badly, what's doing well right now based on the current meta on X, what is doing badly, and you can even hook it up to an API to give it live data or an MCP, which I have to talk about in a future video 'cause it's freaking cool and we use it in our content systems. But that will essentially create a a self-evolving system, which Fable, as we know, is very good at. So your quality of work just gets better and better and better. So loops are a way to have reoccurring tasks, but skills are the way to train Claude on how to do those tasks well.
So if you want to build a skill, and once again, I'm going to do a full tutorial on this 'cause it's quite a rabbit hole, but I'll try and summarize it here. There are essentially three ways. Firstly, you can just take a past chat. So if you've done any sort of work in a chat, financial analysis, you know, video scripting, writing, whatever it is, you can actually ask it using the skill creator `/skill creator` to analyze that entire chat, look at your preferences and create a skill to more effectively next time you use a new chat, for example, execute. And the best part about skills is you can open any chat, you can open, you know, any new session. And if you have your skills and you own, which I'm going to talk about later, how to actually set up the context files for this. If you own these skills, then you know you can use them on any chat, which is amazing. You can even put them into other LLMs. Like GPT, for example, has custom GPTs. This way, you own your own IP. It's not a black box. You actually have access to the to the training that you've actually given Claude over time, which is so important 'cause so many people just use Claude or GPT and just rely on it to keep your memory, which I think is nonsensical because you want to own your memory because you want to be transportable and plug it into different models.
So, you can use a past chat. The other thing you can do is you can actually start a skill creator from scratch. So what you can do is go into Claude, select skill creator, click enter, and it will literally just guide you through the process of creating a skill. So what you can actually do is you could even create like a Google sheet and you could just have like a list of everything you do in your daily life or everything that you'd like to do. And then you can just create skills for them and just train them over time and then you can create a loop so Claude can iterate on that skill to get better and better and better. So the skill creator will walk you through that process. It it's such a powerful way to use Claude, especially Fable, which is going to be even smarter at uh recognizing and and course correcting itself.
So now you know about loops. Now you know about skills. The next thing you need to know is what Fable is actually good at. Why would you use apart from that stuff Fable over another model? And one area where Fable really shines and something that I'm going to do a lot of is the image detection and visual detection and visual capabilities in general of the model. It's pretty insane. Let me let me show you what I mean here. For example, you've got Pokémon Fire Red, which was one of the demo videos that they posted, where it actually went through the entire game and completed it. Now, how is it doing this? It was actually reading the visuals. So, it was reading the game in real time. So, Claude is the best model ever. If you drop it an image, if you drop it a video, if you what if you drop it like real imagery, it's the best model ever at actually looking at that imagery and detecting that imagery. So if you even if you send it a screenshot of a spreadsheet or you're you know, you're working, you know, you're a videographer and you're trying to get feedback on framing, it's it's the best model ever at doing that stuff. And you can see it in real time here actually completing a video game using that primary quality.
And think about it in terms of design. Like let's say you were designing an application or you were a developer or I want feedback on a product that I'm building like the user interface. It will be able to look at that user interface, judge it, and then iterate and create using Claude design, which is what someone did here with an application that is like a calorie tracker application that they built. And this is just so much better with Claude Fable than other models. So of course other models like Opus 4.8, it can do design. Claude Fable is just better. The graphics are better when you actually render something and just the overall intelligence of the model is just better. So obviously you're going to get better results. So the summary from that is that Fable is the best model in terms of vision. It has state-of-the-art vision. You can easily create and review charts, tables, messy PDFs, screenshots of UI. It's like really a partner architecture even if you're into that, you know, interior design. Like if you send a photo of your room, it's just got way better image detection. And that's one of the things that I think you should be using to get the most out of Fable.
Now, the final thing I want to talk about today, because all of this sounds great, right? But the final thing I want to talk about is how to actually set up the right context system. Now, this is universal across whatever model you're going to use. This is not just for Fable. This is for Opus. This is for GPT. This is for future models that come out. I just want to reiterate the importance of this. It's actually giving it your context, your world, and your instructions. So it knows how to give you better outputs. So skills are one thing that you can store within a local folder for example. But the other thing that it needs is context and memory. This is how it's going to make every single response tailored to you or your business. And Fable's a really smart model. It's a great strategic partner and strategic thinker. But, you know, if I'm going to brainstorm a funnel strategy for a new product launch that I'm doing, I need to make sure that the model has the entire map of my business, how it's going, the data, the people in the business, who they are. So, when I mention things, it instantly knows and it can put different connections together to create an outcome that is it's only going to get if if it's nuanced. Like, think about how your brain actually works. Why can you be a good strategic thinker? It's 'cause you know everything about your own business. Even though maybe you couldn't list it all off the top of your head, you're subconsciously remembering all these minor details. Times you've got burnt in business, time things um times things have gone well, times things have gone badly, and you're remembering all this stuff, and you're able to come to a strategic decision. That's likely going to be different if you just give the question in isolation to what Claude will be able to come up with because it doesn't have the context.
Now obviously you can't import your entire business context, but what you should be trying to do is import the major context, the stuff that actually moves the needle, the stuff that's actually going to enable Claude to be a good thinking partner and, you know, a good executor for everything we've spoke about, loops, design, skills, etc. So, what you want to do, and this is where I recommend people start, and then you can go more advanced later with Claude memory systems. What I recommend people do is they you guys just set up a local folder. You just voice prompt. You use Whisper Flow, use the voice transcription feature, whatever you want. You voice prompt everything about your situation, everything about your business, everything that it needs to know. You put that into a documents folder. So, you can see mine has my company map, it has my production SOPs, it has, you know, one pages for important meetings. And then you create a memory file, which is `claude_memory.md`, which basically stores your memory over time. So anytime you say something to anytime you use this folder connected to Claude or Claude code, it's going to reference and update this document. And then you want an instructions file where you basically set up the instructions where you tell Claude how to behave in terms of storing its memory. So you'd have a line in there, "Every time I give you major context, make sure to update my memory folder." So then it becomes a self-updating system. And then you anytime there's a material change in the business, you can just voice prompt Claude and it will update that system. And every time you come up with a new strategic document, which I do all the time in terms of new systems for clipping, new systems for content distribution, new systems for product launches, it will just generate a document and then you create a log here that it's always building upon and always learning from.
So this is for business, but for, you know, your personal life, you can do the same thing. I mean, you know those crazy Obsidian second brains that I built and other people have been building? Well, they are just basically this, just with a visual representation. Uh, I could connect this to an Obsidian brain and it would look really cool because I've got all the MD files and documentation. Obsidian is just a front end to be able to like see everything and connect things in in like a neural network essentially. But the underlying documents are all that matters 'cause the underlying documents are what's actually transportable into other models. If you want Claude to be effective, you need to give it your world. And that is the approach that I would take. And then obviously in the future, like what I did for my personal habit tracker and uh personal assistant, I created a Claude memory system. And Claude memory systems are better for like some use cases 'cause you have more portability and they're also better for businesses because you can't run, you know, business files off a local computer. Multiple people might need access to it. If your computer goes down, you basically don't own the IP to your own business. It's kind of crazy. So start with a local folder while you're learning, and then you can move to like Postgress or uh Super Memory or one of these cloud systems, which I'll speak about more in a future video, but I recommend people just start with local folders. It's like the best way to start and you can get more advanced over time.
So that is how I'd be using Claude Fable as soon as it's live. I'm very, very excited to implement some of this stuff. I'm personally going to be experimenting with a lot of stuff. In fact, I'll show you the exact things. I'm going to work on my new AI clipping system. Once again, it's great at visual stuff. I'm going to work on the loops which I showed you today. But personally for me, I'm going to be using it for business development and product development. That's really high leverage for me. So business development in terms of investment opportunities, sponsorship opportunities, anything that essentially makes me money. I want to run loops for being able to identify uh potential targets to be able to retarget, to be able to enrich data, to get information um about potential clients. So for me, that is really high leverage. If you're running any sort of sales business where you're trying to get new leads, that's really great.
Another thing I'm going to be doing 'cause I love investing, I love trading as well, I'm going to be running some trading bot tests. I only got 2 days into my last test with Fable. And then what I'm also going to do is start generating some ad creatives. Also using Higsfield, using uh Claude Fable as basically the brain behind it to create a system for AB monitoring creatives. We're doing a lot more stuff on like Facebook meta ads uh as well as some YouTube and um X stuff as well. So, I think that's going to be really helpful to be able to like tweak ad spend, compare creatives, AB test, give feedback on creatives. Of course, you know, you still want designers and you still want some sort of manual stuff and you still want to be a strategist here, but having AI be able to help you synthesize data quicker is massive, and that's something that I really want to test out once I have access again, which should be very, very soon, if not by the time this video is out.
So, this is what I'm going to be doing. I'm going to be updating you with videos on everything. I've got some bangers planned soon. So, very, very excited. Thank you guys for watching. Let's uh let's smash this new uh relaunch cycle together. And uh yeah, subscribe to the channel if you haven't already, of course. See you in the next video. Peace.