Transcription
If you're still just prompting AIs like Claude and ChachiBT, you're doing it wrong because the way to use AI now is completely different to what it was just a few months ago. We've moved away from a prompt engineering environment into a loop engineering environment. And in today's video, I'm going to give you the masterass on how to properly loop engineer. This is exactly how you need to be using AI if you want the best results either in your personal life for your tasks or in your business.
Instead of prompting AI, waiting for a response, giving feedback, waiting again, giving feedback, waiting again. Looping means that you can run tasks autonomously. And it also means your outputs are much better because the AI can self-correct. But to fully understand how loop engineering works, I need to explain how an LLM, a large language model like Claude or Chachibbt actually works and the very specific change that's happened now that we have the new class of models, Claude Fable 5 and Chachib 5.6. six soul.
So AI is a genius, especially these new models. They're so smart. They're brilliant at every single task. But the issue is if you're just prompting these models, they essentially have amnesia between every session. They simply forget what you've told it. The context window on chats, whether it's in clawed code or whether it's just in a normal chat tends to fill up and that's what leads to error messages and lowquality responses.
Now, in the past, the only way to use AI was really prompting. Even just a few months ago, you'd have to manually prompt. Even if you had a memory system plugged in, you'd need to keep creating new chats, and you'd need to rely on yourself to keep giving feedback on the AI so it could better complete a task and take feedback into account. So basically, you would prompt, it answers, you check its prompt, you would then reprompt, and then you'd go around again and again and again. Essentially, you were the loop to get the outcome that you wanted. you had to control that entire process. And the reason prompt engineering was the skill is because the better your instruction, the better the way you worded the context to the AI, the better its response. And that's why prompt engineering was a thing. If you consumed lots of AI content throughout 2024 and 2025 and even the start of this year, it was all about prompt engineering. How could you prompt an AI to get the response that you wanted?
But what's happened in the last few months is that the models have genuinely gotten really smart. the new breed of models, the mythosclass model in Fable 5, the open AI frontier model in 5.6 Soul and all future models, they excel in a different area. And if you've been actually listening like I do to the actual experts, the builders that are actually building anthropic and have been giving tips on X and giving tips in master classes, the number one theme you will notice is that they're constantly talking about the loops as the best way to use these models because these models are now actually smart enough to run unsupervised and they have much better strategic thinking.
So you will actually notice if you've used Fable 5 or if you've used 5.6 six. When you actually prompt these models, you don't even need to give that much context anymore. If your memory system is set up correctly and if you use loop and goal, like I'll show you today, you don't really need to have a perfect prompt. In fact, often I'll give it really messy prompts, but I know I'm engineering it in a way that I'm still going to get the output that I want. It's cuz these models are so smart. They're smart enough to figure out how to deliver what you want, reading between the lines.
So the real reason we're having this conversation today is really because the tech has changed and how to get the most value out of the tech has shifted and thus I do believe prompting as an art form is mostly dead and it's now moved towards a new model which is looping. So before it was you in the middle you had this genius which you know kept forgetting things on the left and then you had the work that you needed to complete on the right and you were essentially just the middleman between these two things. You had to apply what AI told you to the work and then you had to relay that and give it feedback. And honestly, this is probably how you're still using AI. Statistically speaking, most people are still doing this. They're still in the middle asking AI for things, then they're applying it to their work, and then they're going back to AI, giving it feedback, and they are basically the loop.
Whereas now you can have a system with these new models with loop engineering where the AI essentially tries something, it completes the work, it self-corrects, so it's smart enough to know how to critically evaluate the output and it fixes errors and then tries again. So it actually runs internal loops to complete work. Once again, this is only possible now because the new models are so smart. So before you were in the loop, now you're outside of the loop. You're able to give feedback. Of course, you're still able to course correct, and I still actually encourage that. I often do that. But work is actually now happening autonomously.
So now I'm going to show you how to loop engineer. Instead of micromanaging the laborer, you can basically just hire a contractor in these new models to do the work for you instead. So essentially what happens is you have the contractor, let's say it's a model like Fable 5. It uses a different model as an inspector. So you could call the smarter model the orchestrator model. It will for cheaper tasks call upon different models. It will review the work. So it's basically the site manager in this case reviewing the work of other laborers. And once it's approved the work, it will mark as done if it hits the goal that you set out for the AI to do. So you can essentially design these loops and maybe they take a bit of iterating if you're doing a reoccurring task in your business or in your life. But once you design these loops, then you can have confidence in the AI to be able to fulfill them autonomously. This is true hands-off work and it's essentially comprised of four pieces. Firstly, you need the spec or the goal of what you want built. Secondly, you need the checklist. You need to define what done actually means. Then you have the inspector which the model will automatically orchestrate and you can tell it also how you would like it to inspect work. It will actually go off and grade the work on its own. And then you have the budget. So, so the loop doesn't like infinitely run, you can tell it after an amount of tries or an amount of token expenditure where you'd like it to stop. This is how you can stop. because I've seen people fall into this trap of running these like infinite loops where you just absolutely burn through your subscriptions or your API tokens.
Now, I'm going to give you a demonstration live of how I actually run loops inside Claude code. But firstly, I want to distinguish if you're confused between the two types of loops because I'm using the word loop overarchingly, but really it's two things. Firstly, you have the inner loop. This is, for example, if you give the AI a mission like you know review my website, let me know how I can improve. This could technically even though it's one prompt constitute as a loop if you use forward/loop because the AI will orchestrate a set of actions looping internally in order to achieve the goal of reviewing your website. So it will try something. So you know it might scan the website. It'll check if the scam was correct. If it failed it would fix it and then it will try again and it will go through this process. Then it will give feedback. It'll check if its own work actually was good and if the feedback was correct and then it will try it and then it will fix it again. So this is only once again possible because the models are now smart enough to recognize whether their own work is good or not. And you've probably experienced this even just interacting with the new models already like if you ask it for feedback on its own work it's actually quite self-aware much better than the old models um which frankly were just dumb compared to the new ones. And then once it's confirmed that it's reached the goal then it marks it as done. So this is the inner loop. So even single tasks can be defined under the looping model if you use for/ loop which I'll show you.
But then you also have the outer loop. This is essentially a job which happens indefinitely and you can also set it on a schedule which I'll show you. So it'll run the job at certain time intervals and it will just keep running and it will just keep self-improving. These are purely autonomous things. For example, if you are on YouTube, you could use a loop on Claude Code to scan your competitor channels and make a log of their videos and what's outperforming and the statistics after certain time frames and it can just keep updating a document around the clock. Or if you are researching for market alpha, let's say, you know, you're in the stock market or the crypto market, you can use an X scanner every day and on Claude code, you can set a loop to run and check for the best viral posts in your niche or or the most, you know, impactful post that might impact your portfolio, which you can feed to Claude, and then it will run the job on an automated schedule. It'll check its own work, fix it, and try again. So you can have an outer loop which is a reoccurring scheduled task operating with multiple inner loops. And you don't need to get too confused here by the way. I'll show you the exact way to do it. Claude actually does this on your behalf. It's not like you need to tell I'll do an inner loop within an outer loop. No, it automatically does this. When you use for/ loop, it can automatically orchestrate. So I'm just trying to explain, you know, how it actually works so you have the the understanding before I show you a real prompt in action.
And one last thing before I show you a prompt in action, I need to define what goal is, forward/goal, and I need to and I need to define what forward slashloop is because we've discussed both those terms in today's video. And I want to be very clear. When you use forward/goal, you're essentially telling Claude in this case, but you can do this on chatbt. It works the exact same. They both have this functionality what you want to be achieved. So build this thing and this is when you're done. You give it the exact conditions for when it needs to complete its work. Technically, you know, you can just prompt an LLM and tell it, oh, you know, I want you to do this. And then technically that's setting a goal, but by using forward/goal, you set very defined parameters as to what you want completed exactly. And this is very important because it's the only way, as I'll show you in a second, that a loop actually knows when to finish. If the goal is not clearly defined, then you know, how does the loop know when to stop? So forward/goal tells the LLM what you are trying to achieve. Forward sloop shows you the interval in which you are trying to achieve that goal. So run this loop every 1 minute, 1 hour or 1 day. And this addition that the newer models now have because how much smarter they are allow you to achieve goals by running automated tasks around the clock. Or you know you could run a loop just for one specific task and it will simply run an inner loop. So to summarize, forward/goal is the mission. For/loop is the the interval, the routine. And this has been uh an amazing addition because it's made work so much more autonomous. And you've probably heard the term AI agents. This is essentially an agentic workflow. If you do what I'm about to show you, you are running AI agents. You don't need to over complicate it. Claude will just spin up agents on your behalf.
So, by understanding looping and scheduling, I'm also going to give you a bunch of loop ideas as well to actually show you some examples of what you can do in your, you know, life or your business. Um, you're actually using AI agents already. You don't need to build an agent. They're built into Claude. All right. So, I just used for/goal. I wanted to scan my website. I have a website called marilesd.com.au. It's mainly just for our newsletter. It's not, you know, a fully fleshed out personal brand website, but I wanted to audit it. You know, maybe I could convert better. So, I asked it to audit it like a conversion consultant. I'm paying $5,000. Write the findings to roast.mmd when it lists at least 10 specific issues. So, this is defining the uh specificity for the goal here. Every issue has a severity, the exact location on the page and a fix I could ship today. And the top three are ranked by revenue impact. Every item must reference something actually on the page. No generic advice or stop after 12 turns. If you don't know how to write a goal from scratch, you can just use AI to write a goal. You can basically say, "Look, I want you to review my website. What would be the best prompt for forward/goal to have a defined set of deliverables? This isn't as important if you're just asking for a single task like this, but if you're actually running a loop, it's very important to have a defined goal. By the way, this is what my website actually looks like. It's for the Miles High Club newsletter. Basically, it's a very uh basic website. I'm sure it'll have some feedback. What it's actually doing is it's reading the HTML. So, it can actually look into the site, not just the visuals, but also like the back end. So, it can also help you with optimization here. So because this is forward/goal, this is essentially just operating like a normal prompt but with more specificity. I'm just telling it exactly what to do.
Basically, while that's running, let's set up a loop and then we'll check back in on it. So a loop starts with a goal. So I'm asking it to fetch the latest uploads from my YouTube channel. So for example, if you were a YouTube content creator and you wanted to look at your data from your competitors, so you wanted to see what topics overlap with your niche, their angles, their titles, look for patterns with their thumbnails, you know, to try and spot outliers. You can essentially give it the channel and you can run a loop. So you can tell it what intervals to look at the channel. So you might do this daily, but for the example of this video, I'm going to show you on the 1 minute time frame what this looks like on my channel, which is obviously AI Edge. So you can see that it's setting up a one minute loop and running the first iteration now. So I've given it the loop. So I've given it the loop which is the interval and the goal what I want it to do. Now just imagine the power of this. For example, here are some loop ideas. And if you do want access to these ideas as well as the entire guide from today's video, all of the assets from this video so you have a full understanding of looping. You can use the link in the description down below. Become a part of the AI Edge newsletter. We post a lot of really cool stuff there. We work really hard on our newsletter and you'll also get access to everything from today's video for free. But let's talk through some ideas while these are running.
If you're a content creator, you could run a loop for looking at data and then coming up with ideation for new content every day. Not just if you're a content creator, if you have any product, you need to create content. So I think distribution is the key in the modern era. So this pretty much overlaps with everything. All of these goals can be run on intervals, remember? So if you have a goal of writing YouTube title options, you can also run it um with a loop. So the AI reviews its own work and then you can run an outer loop basically scheduling an interval for a task to be completed. I'll show you in a second what to do when your computer's off because obviously you know if your computer's on looping will work but if your computer's off you'll need to run schedule. So you can still run loops on a schedule which I'll show you in a second. But if you're marketing you can actually run scraping on people's ads or you know content that's working well. This is really cool. If you're running a business and you have the data for example let's say Slack you can put an agent into your Slack. By the way, subscribe because we're doing a video on exactly how to do that. And then you can run a loop which reads all of the updates from your team channels and automatically compiles a a list of blockers in the business and a summary of the business. So that's how you can use claude code to actually audit your business. It can run a loop to audit it which is I think like a really really amazing use case of a loop.
Now you might think but I'm not running a business. You know you can just run a personal business. You know you are a business. any side hustle you have, whether it is, you know, content creation, whether it's digital products, whatever, whether it's ecom, you you're running your own business. So, why not set up an environment? It could be Telegram, Discord, or even Slack, which I'll do a video on, to run your agents like real employees. You can even give your employees names and run it like a real company. I I feel like that formality actually helps people get better results out of their agents because it changes their thinking around agents. Cuz most people have this word like, you know, they see the word agents, they're like, "Oh, that's really, you know, ambiguous." But in reality, an agent's just like an employee. Just like a real employee, it needs to be properly onboarded. It needs to be given and empowered with information so it can act. If you have an employee and you onboard them into your business and you don't give them any information about the business, no context about the business, how do you expect it to perform? If you don't define what you want it to do, in other words, forward/goal, how do you expect an agent to perform? Just like you wouldn't expect an employee to perform. If you don't tell an employee how often you want something done, how do you expect them to complete that task on that interval? So, I I feel like people have like this idea of agents and AI as if it's so different from the real world, but it's really not. They're really just digital employees. And if you treat them more like real digital employees, then you'll just get better outcomes across your agentic work in general.
All right, now let's flick back to our chord code session. All right, so it's done. It audited my website. It basically summarized it. So firstly, there's no identity or value prop. I agree. Um two, there's zero social proof. I agree. We're actually working on a new website to add that. Three, we have live template placeholders. So it's actually, you know, seen that if you click through, they don't even actually work, which is true cuz it can scan the HTML. Um so this is all really good feedback. And then it created an MD, so a markdown file of all the things we can actually ship today. And the best thing about claude code is now I could just ask it to ship these things. Of course then I need to connect you know Versal or um the domain my website is being hosted on the provider um but it can actually do things on your behalf. So when you start think about AI like this you know defining it um giving it a goal and then empowering it. So giving it the ability to change things it can like just overhaul your entire business overnight. And we're constantly running loops in our business. I'll actually show you my uh Slack environment in a future video. It's very very cool to just achieve stuff essentially while I'm sleeping.
So that was the first goal we set. For the second demo that we're looking at, basically where it's scanning my YouTube channel to see results, it's now running a one minute looping process. Some other really cool use cases would be X. Let's say you're training a writing skill because you want better copywriting for whatever you're doing. You could actually look at an account, let's say an account like mine, and you could basically scrape the account to try and replicate the writing style. And then you can use that to create a skill. loops or any prompt include are much better with not only context or a memory system but also with with skills. So you can have specific skills for specific things. For example, I have a video production skill. I have a writing skill that knows how to write like me. So you could use loop to um automatically scan different profiles and then you could get it as part of the goal. Also in for/goal you'd say every single time you run a loop update the skill based on your findings and then you're actually automatically creating a skill which can empower work in other areas. If you have employees you can then give that skill to your employees to work on your behalf. You see how how powerful this can be as a business owner.
And by the way, it's not just for business owners. It's also for your personal life. Let's say you have bank statements. You can run a loop to look at your bank statements. Of course you know it's still manual with the bank portal. So you might need to download it and get a CSV file. But there are all sorts of things you can do in your own life. You could run a loop for Claude every hour, let's say, or every day, whatever you want to look through your downloads folder or your desktop and just clean it up in the categorization that you tell it with forward/ goal. So, there are probably all sorts of things in your personal life that you could clean up and audit that uh that that you can also run on Claude alongside your business stuff. And I'm using it in both areas of my life. You guys know I built a personal OS. I have my habit tracker. I have all of these like gamification things that I've done um you know bank statements all that stuff with claude code but then I also have of course the business applications which are you know more content and product focused cuz that's the areas those are the areas I'm focused on alongside my uh my agency.
So you can see the loops completed. It's just meandering while it's waiting for the loop to fire again in 1 minute. And you can see the loop counter is out of eight cuz I've defined how many loops I wanted to complete. And you can see if you click on my niche MD, it's created a template which at the end of the process it will update where it you know scan the videos on the feed, the audience, the topics etc. If you want to stop the loop, you can just say stop the loop like this and it will stop it. What's very cool is scheduling. So if you want to schedule at a particular time every day, you can also use forward/chedu every day at 6:00 a.m. because the goal covers the mission, the loop covers the interval, but the schedule covers the exact times. If you want to do things on the exact times, if you want to cancel a loop, by the way, you can always go like for/loop cancel as well.
What I actually recommend that you do because I'll give you the generic list because I'll give you the list that I created using the link in the description below. But if you really want to get the most out of loops in your business or your personal life, just have a chat with Claude. Just get a Claude Fable 5 chat or even GPT if you're on GPT and just explain your lives, your goals, where you'd like to use AI, where you're not sure that you can use AI. Just do a 1-hour brainstorming session. Do it this weekend. Do a brainstorming session and then ask it, hey, out of all of the tasks that I do, out of everything we discussed, what are the highest leverage things that I could automate? And then it can start telling you. And then you can just go into Claude and apply the exact same steps that we've talked about in today's video. And that's how you go from like, you know, a turnbased system where you're just always prompting and you're always in the middle to becoming uh essentially automated with AI and having proactive AI that's able to self- audit and complete work on your behalf.
I think the natural follow-up to this video is going to be the agents video where I show you how to use it inside a real business environment because you can actually, you know, people think of Slack as some business tool or Discord as some like chat or gaming tool, but these are simply applications that you can use to plug in agents to run your personal life. So you could actually have a Slack, right, with a channel for personal finances, a channel for your habits, a channel for whatever you want. Or you could vibe code a dashboard for it, but you get the concept. And then you could have agent employees doing jobs under each thing. So you might have a let's say like a health channel in Slack. You then you might have an agent under the health channel, which is your personal trainer, your nutritionist, and then you might have, you know, a finances CFO, uh an adviser, and you can run loops within that. So, you start to see the power of this if you actually treat your life like a company. There's so many cool applications of AI, man. It's super super exciting. What a world we live in. But that's it for today. Now you understand looping. Make sure to subscribe to the channel for more. Link in the description below if you want to get access to the full guide from today's video uh with all of the graphics that I showed and um the examples that I also show showed on the spreadsheet. I think there are some really cool ones in there. And yeah, I'll see you in the next video. Have a lovely rest of your day. Peace out.