Speaker 1 (00:00):
So over the last six months or so, I've been testing quite a lot of different ways as to how to build out specific agents on manners, on open claw equally, building out websites and platforms on lovable. So I'm going to take you through a bit of a process as to how to build out specific agents, specific prompts to make sure that you get the best from what you are trying to ship essentially. Hey, it's Oliver Bruce, and welcome to the Unlock. Previously known as Success is in the Mind. I'm a UK entrepreneur, angel investor, and neurodivergent founder, and I recently exited my first business, which I scaled from my university halls into a multimillion dollar agency with no backing, no funding, just grit, mistakes and determination. I want to pass on some of the lessons that I've learned, the barriers that I had to overcome and the challenges that I'm still coming up against today.

(00:46):
This podcast doesn't grow by itself, it grows with you. If you could possibly share this with friends, family, colleagues, anybody you're in business with, somebody that you think might find this useful, I would be greatly appreciative. Anyway, let's get into it. So if we're starting off with building agents, then we need to start off with what I like to call the three Ps. So the three Ps simply put, you need to start off with purpose. So you need to understand exactly what the purpose of your agent or your platform that you're trying to ship is. Without understanding that purpose, there'll be hallucinations, there'll be bits within the code that maybe you don't want to have. So actually the agent won't specifically do what you need it to do. So start off with the purpose, add that into product.

(01:37):
So the second P is product. Now what is that? That's specifically what you are wanting your agent to actually do. So once you've understood the purpose of where the agent and what the agent should be doing, the product is how you package that up, how would you explain that to an individual, to a client, to a colleague with regards to what the agent is actually going to be doing? And that then equals the third P, which is prompt, arguably the most important. So without the prompt, you're not going to be able to brief in the specific platform, lovable, Manus, open floor, whatever it might be that you're wanting to use. So understanding the purpose of your agent. So let's just say it's going to be a diary management agent. So that agent is specifically going to look through your diary every single morning, send you a brief, send you an update, send you an agenda of what the day ahead looks like, almost like a pa.

(02:27):
The product is exactly that. It is a PA type agent, and then the prompt should be five or six pages of very detailed actions that you will then put into whatever platform you're building this agent out on. Now before we go into the types of platforms that we want to build the agent out onto, I'm going to talk to you about how we can actually build out the prompt. So once you've got your prompt idea in your mind, you then need to build out a project on Claude. Specifically, I find Claude a lot better than other channels. So assuming we're going to use Claude, you would log in as you normally would, and you need to build out a specific project. So once you're into Claude, you build out a project.

(03:16):
So the project piece is critical because that's where you can train the memory of Claude essentially to learn and understand exactly how you're wanting to essentially bash this agent, to build this agent out. That can be very, very simply built by uploading knowledge documents, URLs, images, things that you are wanting it to learn and to understand context of naturally the prompt that you have built is crucial to put into that project folder because that's going to be the kind of heartbeat of what you are trying to essentially build. So what you do is you build out the project, you will then upload your prompt, and that is the heartbeat of the project. Once you've built out that prompt, you will then ask your project folder to remember. So it must be prompted to remember that that prompt is the heartbeat, is the mind, is the brain of whatever agent you are trying to build out. Once all of that is done, you are largely ready to start building out your agent. So assuming for the purposes of this demonstration that we will be building every prompt out and every type of code out through Claude or Claude code, we need to maintain consistency in that project.

(04:43):
So Claude project equals agent structure. Okay, so we'll start off first with building out an agent in open Claw. Open Claw is going to be the platform of choice here. So open claw, there is a barrier to entry with open claw, so you do need to buy a Mac mini, you can buy those for a thousand dollars. You'll then need to spend some time on actually installing Open Clause specifically on the Mac Mini, and it does come with some security risks. So it's worth just understanding exactly what exposure you are comfortable with before going down this process. But assuming you are happy with the setup of Open Claw, you have the terminal installed on your Mac, you are now essentially ready to build, what you will do is you will take that prompt that we had in the project here and you will feed that into open claw.

(05:43):
Now, the best way to do it, and most people will build out specifically an open clause, sort of main portal, main terminal if you will, but to get the best results, I find it's better to build it out in the actual Mac terminal specifically. Now, that is more about the code than it is about the prompting, but the actual prompts that you put in initially in the open claw portal is kind of the ramifications, the ring fence, the kind of walled garden if you will, as to what you're trying to specifically build out that specific agent, right? So once you put the prompt into open, you're ready to start building. So you'll need to upload, sorry, no, you'll need to open your terminal.

(06:23):
So you'll need to open your terminal on your Mac to be able to essentially start the build properly ensuring that everything that you're building out within that terminal is saved as you go along to make sure that actually open claw is as robust as possible. Now, it's really important to understand with Open Claw that you must always have that Mac on. So if you switch that Mac off, for instance, an open claw or your agent isn't necessarily going to live, isn't going to breathe, isn't going to work. So that's the importance of having a Mac mini because Mac Mini's typically have a far longer lifespan, are able to stay on for far longer than any other Mac, even if you were to turn off that sleep mode. So open, sorry, Mac Minis for Open Claw are the primary sort of objectives here. Once you've got that terminal built out, you will then need to go into your claws code project.

(07:14):
So this is where the fund begins clawed code. Now, this is within your project's folder. It's within the knowledge of your clawed infrastructure, if you will, and this is where the fundamental backbone of everything that you're wanting to build out on Open Claw will live, and it is super simple. What you will do with clawed code is you will start to bash it and you'll start to prompt it and you'll start to talk it through your objectives. Hey guys, sorry for interrupting the podcast, but I thought you might find this useful. If like me, you found that one of the biggest headaches when running a business is managing money across different tools, currencies, and expenses. Then I think I've got the solution for you. Incar is built to fix this headache. Incar is a new financial platform designed specifically for high growth modern businesses. It gives you multicurrency accounts, connected banking and smart spend management all in one place.

(08:14):
The really cool part is the rewards with In card, you earn up to 2% cash back on things that you are probably already spending money on, like advertising, SaaS, tools and travel. Every time you spend, you generate points and you redeem them instantly for cash directly in the platform. So if you are building or scaling an online business and you want a smarter way to manage your finances, check out in card using the link in the description. Thanks so much for taking the time to listen to this. Let's get back to the episode. So you will basically treat Claude Code as a consultant, as a colleague, as a friend. You will ask it to do things for you in a way that you would consult, say an advisor. You will say, this is the objective, having uploaded the prompt, this is what I want to do.

(08:59):
Now, the most important thing when you're uploading the code from Claude Code into the terminal is to ensure that you have a two-way conversation with both. So you need to take the code from Claude Code, you need to put it into the terminal. Once it's finished, populating, screenshot or copy and paste what it has populated and put it back into CLA code. Claude Code will then understand if there's any issues, if there's any glitches, if there's any hallucination before then telling you what you need to do next. Bearing in mind that this process, the projects, the Claude Code piece will understand what your end goal is because we've built out the purpose, we built out the product and we built out the prompt. So ensuring that you keep dialling back into Claude Code will keep the agent build straight and narrow. Essentially, that process will then go on until you get to a point that you are happy, smiley face for that.

(09:52):
So once you are happy with your build, so to speak, you are then able to start to test stress test and ship the product. Now, it's really important when building out anything through Claw code or anything frankly on Open Claw that you start really, really small. There's no point in trying to solve World Peace straightaway because that's just not going to happen. So something very, very simple like diary management, like email management, like scheduling content onto social is the best place to start. As you become more FA with how Claw Code works and how Open Claw works, you will start to become a little braver. You will start to plug in different platforms, if you will, APIs, et cetera, to be able to then build a flow that's fully autonomous. What's really important when building out on OpenCL is that you put the gate parameters in place.

(10:42):
So you tell it to not essentially think for itself to not be overly autonomous because that's when it will start to hallucinate. That's when it'll start to do things that maybe you don't want it to do. IE it might schedule something incorrectly, it might reply to an email incorrectly, it might read something you don't want it to specifically read. So being super, super focused when building out that purpose product and prompt is key to make sure that both clawed code doesn't give you the wrong code, but also Open Claw doesn't hallucinate and start to do things that you don't actually want it to do. Now, I won't go into too much more detail around how to specifically build out on Open Claw because that'll be a whole nother video in terms of building something and taking it, taking it to ship. So what I will do however, is build out a flow in terms of Manus and lovable because they're two very different things.

(11:31):
What's important to understand with Open Claw before I get into that is that Open Claw has its own sort of terminal slash chat function, and you can communicate to those chat functions through WhatsApp and other associated third party communication channels. So could be Discord, could be Slack, could frankly even be email if you wanted it to be. So OpenCL gives you sort of more opportunity to talk through different channels to patch in different people, to patch in different agents. You can build multiple different flows through Open Claw specifically. Whereas when we get into Manus, it's a little bit more refined, it's a little bit more one size fits all. So there's less kind of freedom, if you will to build out what you want. So going into Manus, Manus is a different beast. Manus was bought in late 25 by Meta, so it's very, very meta focused.

(12:36):
It's very good at certain things that you would think meta are wanting you to do, such as content, such as scheduling and such as copywriting. It's less good at other things such as building out multiple different agents. So in terms of Manus, you don't need to worry specifically about building out code. So when you go into Claude code to build out an agent with an open claw, you can somewhat disregard that process for Manus. However, what you do need to do is you do still need to have your Claws projects open because this is setting the parameter for what Manus can and can't do essentially. So always dial it back to Claude code, always dial it back to Claude Projects specifically. That is the kind of hub of whatever you want to build out. Treat that as your consultant, as I said. So once you've downloaded Manus and you have to download it to your desktop to get the best output, you can simply use it online, but I find it better to have as a desktop app, you will then want to put that prompt directly.

(13:37):
So if we write the prompt, you'll then want to put that prompt directly into Manus under the agent's tab. So you'll see within the Manus dashboard that there is a button that says Agents. Now that is essentially the area that you'll start to train, talk, communicate, and work with Manus on. So whereas in Open Claw you can build out different agents, as I've already said, and they can work autonomously in Open Claw. In Manus are very much working in one siloed way, and if you haven't because it's in a sandbox, uploaded the information into that specific agent section, then it won't have the level of knowledge that you want. So it's really important to make sure that in that agent section within Manus, you are training it and giving it the information that it needs. Now with the prompt in Manus, it will then start to build out essentially a flow.

(14:30):
So once you've got that Manus prompt uploaded and you've clicked go, it will start to build out what it thinks is the right way to deploy whatever the objective is that you've tried to train it on. Now interestingly with Manus versus Open Claw is that you can plug directly into platforms like Instagram, Gmail, et cetera. But the one thing that I would say with Manus is that it's a lot more expensive than Open Claw. Maybe not initially to download because it's free to download and you do get free credits, but in time when you're asking it to do deep tasks, even reply to emails or even analyse for instance, your calendar, it can cost a lot in credits. So just be mindful that actually for me, when I'm using Manus, I'm using it specifically for creative tasks. I'm using Open Clause specifically for deep thinking tasks.

(15:19):
So there's two very different objectives there. Anyway, once Manus has built out that flow, you can start to work with it. It really is as simple as that. With Manus, once the flow is built out and you've plugged in the specific areas that you're wanting to work with it on, it will then start to think. It'll then start to do. Now with all ai, all AI agents and all kind of prompting, you will have to keep bashing it. It won't be perfect straight away. So it's really important when you are building creative out on Manus to make sure that you are uploading either the perfect file, exactly what you're wanting it to look like, and then saying this is what it needs to look like, don't deviate from, or you are working with it in some way to ideate and to create content that won't be perfect straight away, but you are happy to invest the time in getting it to that point.

(16:03):
Now, Manus plugs directly into Nano Banana. Nano Banana is one of the best kind of gen AI platforms from a creative perspective. So when building out creative in Manus, typically it will pull from a nano banana. Yes, it will cost a bot credits, but the quality will be far better than if you're building that out specifically in open Claw, in my opinion. So that's what I would do in terms of Manus, that's how I would work with Manus. Now the difference again with Manus is that you will have an app on your phone. You'll have an app on your desktop, and you will literally consult it as you would do chat, GPT or Claude for instance. Whereas with Open Claw, you'll be able to communicate directly with it through WhatsApp if you wanted to, but you wouldn't have a specific app on your desktop or a specific app on your mobile.

(16:47):
So it's a very different workflow, but the best thing to do is to get open Claw to talk to Manus for the creative and then Open Claw to run independently for any administrative tasks I mentioned earlier. But I do think something that you guys might find super useful if you are running a business and managing multiple transactions across multiple platforms is in card. It's a new financial platform for modern online businesses giving you guys multicurrency accounts, connected banking and smart spend managements all in one place. You can open up an account in minutes, attach cards for expenses and earn up to 2% cash back on everyday spend like ads, SaaS and travel. Check out Incar using the link in the description. Thanks for taking the time to listen to this. Let's get back to the episode. So what we've now ticked off is we've gone open claw for admin slash automation.

(17:42):
We've then got Manus equals creative equals creative. Now, as I said to you, open Claw into Manus works really well, Manus into Open Claw, not so well. So it's really worth bearing that in mind. You can run them independently if you wanted to. You can just have open claw running your agents and doing what you need it to do. That is absolutely fine. It just takes a little bit more building or you can have Manus again running independently doing what you need it to do, but it's more costly. So there's that kind of, is it worth spending the money on Manus because it's kind of already built or is it worth building out so that it actually works specifically for you as an independent sort of individual business? In which case, maybe using open floor is the way to go. There is no right or wrong way of doing this.

(18:27):
This is just what I've experienced over the last six months of trying to build out and automate workflows. These are the agent side of things. When it gets to actually deploying and shipping platforms or websites, then lovable is the place to do that. Now, lovable is very different to Manus because it's not specifically an agent. Again, it's very different to OpenCL because it's not specifically an agent. What lovable actually is is a platform where you can put no code in or you can put code in if you want to be able to build out bespoke platforms, workflows, apps, whatever it might be, it's super powerful. I'm sure you guys know about it. So how do you build out lovable? Well, again, it doesn't take a genius to work out that you need to go to your clawed projects. Again, you need to go back to your knowledge, you go back to the brain, the heartbeat if you will, and you get a prompt.

(19:20):
After you've put in the purpose the products, you'll get that prompt. That prompt is what you're then going to put into directly lovable. So lovable again will have an interface within it. You can access it online. You don't have to download an app, you don't have to download anything on your desktop. You can access it simply by typing in lovable online, and you'll be able to upload that prompt directly into the chat function, at which point lovable will then pretty quickly spin up a wire frame of what you need your site or your platform or your app to look like. However, what you need to do is to get the best from lovable is you need to keep uploading ideally code because you get a far better build. But again, prompts are actually fine. So what is the difference between prompt and code? Well, prompts are kind of just words, if you will.

(20:02):
So you'll give it a paragraph or a couple of pages of what you're wanting it to actually achieve, whereas code obviously is what it says, so it's going to be more intricate, it's going to be more detailed. It's going to have different types of symbols in there to be able to actually build out specifically a key objective. Typically, when you're building out on Open Claw, you'd be using Python. If you're building out on Manus, you probably wouldn't be using Python, you'd use a bit of Claude Code. It wouldn't necessarily be as robust. And then again, when you're building out on Lovable, typically that's going to be just Claude Code as well. It's not going to necessarily be Python, but again, you can decide what that needs to look like. So once you've built that prompt out on Claude, you will then put that into Lovable, and then it will start to render out essentially a dashboards or a website or whatever your key objective is here, and you will simply go through the motions of taking the response from Lovable and you will put that response back into your Clawed project, at which point it will kick out another prompt and you guessed it, you put that prompt back into Lovable and then it will carry on building out your website, your platform, whatever it might be.

(21:10):
Now, it saves online so you don't have to kick save every single time. It's not going to disappear if your computer crashes, for instance. But what it will do is it will allow you to see before it goes live exactly what you're trying to build out before you ship it. And again, like with all of these agent builds or autonomous builds, you need to make sure that you start really small and really intricately to make sure that actually what you're shipping and what you're delivering is going to work because you're able to actually critique it, you're able to analyse it and look at it rather than just trying to build everything out when you're not actually quite sure what to kind of critically analyse at that point because there's just so much going on. So once you've done that, you're then able to ship your lovable website, if you will.

(21:44):
Now, I think it's really important to understand that there's different levels of building out prompts on Claude, depending on the level of detail and intelligence and knowledge you're wanting Claude to pull on. So for instance, Opus is one of the Claude operating systems that you'd probably want to build out on because it's slightly more intricate, slightly more detailed, thinks a little bit longer, and goes a little bit deeper. Equally, if you didn't want to burn as many credits, you could build it out on sonet. Now that's a slightly quicker, slightly like to touch way of doing it, but again, it depends on your budgets, it depends on what you're actually wanting to achieve, and it depends on how deep you're wanting your code to actually be and how much you want to get to think. So it is worth understanding and doing a bit of research onto the different types.

(22:24):
They're being released every week it seems at the moment. So one week Opus is the best one. The next week it might be something slightly different, but it's worth making sure that everything that you're doing, be it on Manus Open Claw or lovable, is being built out into a clawed project. You're using Claw code when you can to be able to put that into whatever the workflow is, and then you're shipping it very, very slowly and very methodically. Now, if we just break down some numbers, so barrier to entry for Open Claw would be roughly a thousand pounds for the Mac Mini.

(23:00):
I'm going to disregard the second screen. You don't need to necessarily have that. It would probably be on average 500 pound for credits or tokens per calendar month. That typically would be enough to get you started and to get you rocking and rolling with probably anything that you're wanting to achieve right now. Naturally the credits will increase as you're starting to burn more knowledge or more tokens through Claude. And again, that plug straight into philanthropic, and you can pull Andro tokens from that. That's the best way to work it. But again, for 1500 quid, give or take, you can build out open Claw on your map mini and you can start to deploy and you can start to ship some product. Now, if we were to look at Manus, for instance, Manus is a bit more variable in terms of how it's priced. Manus at the moment.

(23:50):
Burns credits ridiculously, you get 300 credits to start. So if you were to get your 300 credits, typically you'd be able to build something very light, touch out straight away. It wouldn't necessarily be that detailed or really that autonomous, but it's a good starting point. Thereafter, meta have decided to bill in different cadences. So you can go all the way up to, I think it's 8 million tokens, and again, that costs thousands of pounds a month. Equally, you can be spending as I do $400 a month, which is about 110,000 tokens, which for me does enough of the heavy lifting. But again, Manus is a little bit different. There is no upfront cost for that. You can download it totally free. You just then need to be able to actually afford to pay for credits. If you're wanting it to do more 300 credits, you'll burn through very, very quickly, but it does refresh if you're on the lower tier quite regularly. So really good from a sort of low barrier to entry perspective. Lovable, however, is if you can read that lovable big heart, there we go.

(24:52):
Lovable is one of the most, the easiest access, sorry, the easiest platform to access. Again, you just go online, it won't cost you very much at all to subscribe to, and you don't burn through tokens that quickly. Again, for a couple of hundred dollars a month, you can build a pretty robust platform, certainly a lot cheaper than getting a developer to build out a platform for you guys, which would probably cost you tens if not hundreds of thousands of pounds. So lovable, again, has different pricing mechanism. Barrier to entry is very, very low, but it's not an agent. It's a platform build. So when you're looking at the different types here from coat, sorry, from open Claw through to Manus, through to lovable, open Claw is the most expensive, but I think it's the most robust and you can do the most with it. Manus is the best from a low barrier to entry perspective, has a level of automation, but you will very quickly burn through credits and probably end up spending the same as you would on OpenCL if you just fronted the cost, and again, lovable because the purpose is very different.

(25:46):
The price point is also very, very different. So worth bearing in mind exactly what your objective is or indeed your purpose, your product. And then the prompt. Again, you can ask Claude specifically what the best platform to build out your specific agent is. So if you are building out that purpose product and prompt, it may well tell you that actually building it out on Manus is a total waste of time, and you should just double click into Open Clause. So again, treating Claude as that kind of consultant is super, super important to make sure that you are not burning credits, you're not burning money, and you're not burning time, right? So I guess that's a really light touch way of understanding how to build out a really simple agent. So in summary, you've got open Claw, which equals automation slash scale. You've then got Manus, which equals creative slash ease.

(26:48):
And then what you've got down here is you've got lovable, which equals platform slash product. If you were looking at how you could monetize these as a business, you'd be able to monetize, for instance, lovable far, far quicker, because when you ship that, you can charge a SaaS fee, people can sign up to whatever your platform is and it can do stuff for them that is useful. So lovable will be one that you could invest time in and probably very quickly make a return. Open Claw is very much operational, so it's being able to actually do essentially projects or processes within an organisation or just to make your life a little bit easier. And similarly, Manus kind of falls into the same bracket there. In terms of operational slash creative, I prefer Manus from a creative perspective. So yeah, over the last six months I've been testing, this is kind of my rough workflow.

(27:38):
It's working quite well for me. Hopefully for you guys, if you haven't yet dabbled in the world of agents or autonomy, you hopefully learn a little bit from this video. But yeah, go and build. It's good fun. Appreciate you listening to the podcast. Hopefully you found it useful. For those that want to read up or learn more, head over to my LinkedIn page, Oliver Bruce online where you'll find a weekly newsletter called the Brucey Bonus where we double click into more detail and give you more tips and tricks around how to scale your business. If you want to share this with friends, family, colleagues, business owners, people that are in your circle then might find it useful. I would be super appreciative if I said at the beginning of the podcast, this does not grow on its own. This grows with you and we do it for you.

(28:20):
So thank you so much for listening and catch you next time. I mentioned earlier, but I do think something that you guys might find super useful if you're running a business and managing multiple transactions across multiple platforms is in card. It's a new financial platform for modern online businesses, giving you guys multicurrency accounts, connected banking and smart spend management all in one place. You can open up an account in minutes, attach cards for expenses, and earn up to 2% cash back on everyday spend like ads, SaaS, and travel. Check out in cards using the link in the description.