Manus AI: Everything You Can Do With This New AI Agent

Hey, welcome to the Next Wave podcast. I'm Matt Wolf. I'm here with Nathan Lands. And right now, agents are all the rage. We are in the era of AI agents, and right now, the biggest agent that everybody is talking about is Manus. So in this episode, we're going to break down what is Manus. We're going to show you what other people have created with Manus. We're going to use Manus ourselves and actually build a really cool website with it. You'll see what it built before the end of this episode. We're. We're going to break down how Menace was made, the bigger implications, how you can use it in your business. This episode is jam packed with AI agent goodness and how it's going to impact your life and your business. So get ready, you're going to be in for a fun ride. Let's get right into it. So where should we start? Let's. Do you think maybe we should just start showing off some of this stuff that other people have made with Manus?

Yeah, I think there's been so many crazy posts on X, like showing how people are using it. I think let's just start there and show what people are doing with it.

Cool. Yeah, I agree. I think just sort of looking at what's been created is like a great jumping off point here. And then we'll get in and test it ourselves. You and I have both played with it a little bit already.

Right.

Like, I know I. I shared a screenshot on my X account of me trying to. Actually, I'll share the. The tweet. So I shared a tweet of me trying to play with it here. And in this tweet, I basically gave it the prompt to go research Manus, because I want to make a video about Manus. So go research Manus. What is it good at? What is his limitations? What's the competition? You know, just give me as much Deep dive info as you can on Manus. And it broke right during the portion where it was investigating the limitations of Manus.

So literally the spot that it decided to stop working was when it was starting to deep dive on its own limitations.

Yeah, that makes sense as intended.

But anyway, I know you've played with it or tried to play with it as well and sort of ran.

I just tried to play with it a little bit. Like, it took forever to get access. You know, I had to message a bunch of people in X to get access and I finally got access and I don't know if it was just the Time zone difference in Japan, maybe they were doing some maintenance or something, but when I tried it, like it.

Wasn'T working well, I think what's happening is that Menace is insanely overloaded. They actually tweeted this on the day that we're recording. Seven days, 2 million people on the wait list.

Wow.

So, yeah, I think, you know, they've been letting a lot of people in every day. I think they're letting like, I don't know, 50 to 100 people in a day or something like that. And I think they're already overloaded. So I think, I think that's what we're seeing right now is just so many people trying to use it at once. Even that one that I was just showing you here where it says investigate and document Manus AI technologies limitations. And that's where it actually stopped working. If we look over here on the left, it says high system load has caused an internal server error. So it didn't like get messed up because it didn't want to, you know, share its own limitations. It got messed up because at that point in its own processing was when the servers got overloaded and it just sort of stopped working. But yeah, let's, let's take a look at what some of the, some other people have made here. So this one is from AK and he actually got it to make this like Instant Run game.

Very similar to like, very similar style to that like levels IO flight game. A lot more basic obviously, but it's. If you're listening on audio, it's basically a character just like running in a straight line and it's like a 3D world but everything sort of like basic shapes like squares and rectangles and cones and stuff. It's very, very basic looking, but it's just a character running and obstacles getting in its way and it's obviously trying to dodge the obstacles and like grab power ups and stuff.

That's just from a single prompt. Right. So like it went off and did everything.

Yeah, literally. AK says the prompt was make a 3js endless runner game. So 1, 2, 3, 4, 5, 6 word prompt made that.

Yeah. Damn. He specified 3gs and it's kind of interesting. Like I, I've played with 3Gs for like years. I think that's been up front like over 10 years. Right. It's. It's been a long time that that's been out and it just never took off really. And it's crazy now that it's taking off because of AI.

Yeah. Must be one of the easier Ones for, you know, Claude and Grok and these kinds of tools to code with. I'm imagining.

Yeah. I'm sure there's tons of examples out there for like over 10 years now. Like almost 15 years. Right?

Yeah.

Of examples.

Yeah, yeah, yeah. So a lot to. A lot to pull from. A lot probably in their training data set.

Yeah.

So here's one from our buddy Bilavel here. He says, I tested Manus AI. It's the coolest thing I've experienced to a truly autonomous AI agent. I can't wait till this thing can use desktop acts like Premiere and Photoshop. It low key feels like Baby AGI. That was actually my first thought. Like when I actually was using Manus the first time, it. It felt very similar to Baby AGI because what it'll. We'll show it off in a minute, if it works, hopefully. And what it'll do is it actually builds out a plan for you and then once it builds out the plan, it starts back from the top and then tries to execute each step in the plan, which feels very similar to what Baby AGI was doing in the.

Past, but it actually works. That'd be like the clear. Sorry, Yohei, if you're listening. But yeah, for sure.

Yeah. I mean, when, when Baby AGI came out, it would get stuck in these loops, right? It would, it would like go to step three and then for some reason just like keep on searching the web and never get past step three. This one actually seems to, you know, actually follow through and finish. But here's what Bal, here's the prompt he gave. Can you open up a 3D view inside Google Maps of Austin, Texas? And let's find the most compelling photos to take of the downtown skyline. It's very important that you change the map's perspective so we can get oblique 3D views versus a top down one. I want you to fly around surveying downtown from 3D vantage points in Google Maps and tell me your favorite location, slash, photos. So that's the prompt which is. I mean, Bo Oval uses big words. I love him, but he does.

And then we can see it starts to go and do the thing here. Let's see. Let me fast forward here. I want to see it actually do it so you can see on the screen it actually created this task list. Research drone regulations in Austin. Like a. Why does it need to do that? He's doing this virtually, but search for potential drone locations.

Oh, okay, okay. I see what he did there. For some reason I was thinking he was trying to get like A virtual image. But he wants to go fly his drones downtown, and he's trying to figure out the best places to do it. I sort of misunderstood when I read it the first time, but you can see here, very similar to baby AGI. It went and created this task list for the AGI for the agent to follow, not the AGI. We're not quite there.

Later.

Yeah, yeah, that's next episode.

But you can see it says, research drone regulations in Austin. Search for potential drone locations near downtown Austin. Create a short list of locations. Scout locations using Google Maps 3D view. Evaluate locations for drone flight. Suitability complete. Top three recommendations. Prepare final report with recommendations. Deliver results to user. So that's the sort of task list that it gave. Then you can see as he scrubs through it, goes through it, does all these tasks. We can see here. It's actually looking on Google Maps for him, trying to find locations, and then eventually gives him top three recommendations. Butler park. Why it's recommend date. Why it's recommended. Zilker Park. And there's one more in there too, that it kind of skips over, so I can't see what it was, but it found three different places. So, I mean, that's pretty wild that it can even do that, in my opinion. Yeah, like, how does it know from Google Maps, like, which are good locations?

Yeah, I mean, I don't know. Is it doing a combination of, like, looking at the images itself and also searching the web? Is it like, doing both and, like, kind of combining that into, like, possibly?

I mean, it's using the 3D view. Right. And he was asking about getting shots of the skyline. Right. So I guess it could go to Google Maps and, like, look at a whole bunch of different angles inside of Google Maps and try to find a good angle on the skyline and then compare those good angles from what it saw in the 3D view of Google Maps with the drone regulations of where he's actually allowed to fly his drones, and it compared those two and then gave him some locations. So pretty wild. Hey, if you take a look at my web presence online, it's safe to say that I'm a bit AI obsessed. I even have a podcast all about AI that you're watching right now. I've gone down multiple rabbit holes with AI and done countless hours of research on the newest AI tools every single week. Well, I've done it again, and I just dropped my list of my favorite AI tools. I've done all the research on what's been working for me. My. My favorite Use cases and more. So if you want to steal my favorite tools and use them for yourself now you can, you can get it at the link in the description below. Now back to the show. And then we've got this here where this is a thread from Al Scene or else over on X. And this is a thread of a whole bunch of different shares that, that he found. So the first one, he had it clone the Apple website. Yeah, this one's, this one seemed like a simple task to me. But we can see here it literally made like a. An exact duplicate of the Apple website here. And then analysis of Tesla stock.

Wait, did it take a screenshot of the Apple website? Is that how it was generating the website?

I'm pretty sure, yeah. Of the website.

It looks like it actually like took screenshots of some of the elements. Right. Just like page pasted them in, which is.

It kind of does. Yeah, I don't know. I don't know. It's. There's not a whole like breakdown there. But then we've got an analysis of Tesla stock where it went and did a whole bunch of research on Tesla. I don't really feel like you need that. You need Manus for this. I did, I actually did our. This exact one. I had it go and analyze Tesla stock for me using chat GPT's deep research and it was really, really, really good. So I feel like Manuscript is almost overkill just to do this research here.

Yeah, you can use that Grok, Perplexity, they're all great for use cases like that already.

Yeah. Plan a family trip again.

Do you really need like an AI agent to plan the family trip? I mean an agent like Deep Research from, you know, from ChatGPT or from Google or from Perplexity. Like those Deep Research agents are really good at planning family trips. I don't really feel like you need Manus to do that. The ones that I think have impressed me more have been more these like visual ones. Like here's this, this, this game that somebody created again with three js. We can see the tasks that they did to make this game. Create project structure for 3js playing games. Set up 3js render environment design and create plane 3D model. Simplify left right plane controls. Fix plane shooting test. Improve plane movement and controls. Finalize game implementation. Share updated game version with user. And then on the right side of the screen we can actually see the plain game. Kind of looks similar to, you know, Peter Level's game and that game we saw from ak similar style with where it's real basic shapes, but it did it with a single prompt.

AI game with Manus. I mean this is wild. This one I almost don't believe. I don't know if this is real or not.

I actually think, yeah, where is it getting the model from would be my question. Like everything else is totally believable.

Yeah. This one I don't think is real. This looks like it was like Unreal Engine or Unity.

Yeah, yeah. I think this. There was a few that people were like BSing.

Yeah. This one says AI game with Manus AI cursor. Let me actually open it.

I think it got more extreme after that. Like there's people who are sharing like AAA games and being like, I can't believe I just got this out of Manus or Claude or whatever. Like one prompt and it's totally.

I saw one where somebody did like an EA Sports like soccer game and it looked really it I'm and like. But the EA Sports logo was still on it and everything. I'm like, yeah, this one I don't actually believe it says Vibe coded this AI game with Manus AI. And Cursor crafted the prompt using Gro, built the back end with Laravel and the front end with three js.

But I wouldn't be surprised if that's only like months away. I mean I have seen like recent demos of like prompt to 3D model that are getting quite good. So if you teach, you know, if an agent like this learns how to use those tools within this goes off and uses the tool, creates a prompt with that tool to generate a 3D model, extracts that puts it in. These things are going to be possible like in that and pretty soon.

Yeah, I believe it. This is, this is fake. I mean I'm like 98% positive. That's not really done with with Manus here.

Not with one prompt for sure. Yeah.

No. Let's see. Website SEO optimization. Again.

Not really a great use case for Manus in my opinion. Double agent productivity. Somebody actually showing they were driving their Tesla and they had Manus running on the screen in their Tesla doing some autonomous stuff. And this one was actually shared by Manus so it's a little bit more legit. Double agent productivity. While Tesla's full self driving takes a power user to his coffee meeting, Menace preps him with all the key talking points. So he's on his way to a meeting. While Menace is rounding up talking points for him to actually talk about at the meeting he's going to.

That's crazy lady who had the glasses, like just telling you, like, what? What are my talking points? And then I should respond in this negotiation. Like, here's the parameters of like how I should negotiate this deal. Like, oh, I should ask for more right now. Look at how he's responding, you know. Yeah, yeah, yeah.

Well, I mean, with the Project Astro glasses that I was, I actually demoed when I was at DeepMind in London, like, you're gonna be able to do stuff like that. It's hearing the conversation around you and we'll be able to actually like put little text on the screen in front of you that you can read off of. And one of the cool features of that is you can almost use it like a teleprompter. I can see getting those glasses and like making a YouTube video and like literally reading off a script the whole time, but trying to hide the fact I'm reading off a script because it's like, all right inside of my glasses. In fact, I might do that for future HubSpot sponsorships. We'll see.

Let's see. Create an amazing animation using P5JS and it created this like colorful animation. And yeah, that's everything in this thread. So those are some of the things that other people have created with it. Really quickly I think we could, we should talk about like how Manus was actually developed because this was sort of a point that you brought up that you saw from Gary Tan. But you know, people were complaining, going like, hey, I just simply asked Manus to give me the files at Opt Manus and it just gave it to me. It's Claude sonnet Clodsonnet with 29 tools using multi agent and browser use. So basically Manus is just using Claude and this browser use, which I believe is open source.

And it was kind of funny people, when they discovered that some people started calling it like a fraud. Like, yeah, but you know.

Yeah, actually I think I shared that with you for a second. Not deleted like, because that doesn't really make any sense. Like, I get what they're saying. Like, yeah, they even like created some amazing proprietary model or whatever. Maybe they have like, it's kind of like, you know, one on top of Claude, but. But still, like, I mean, they're combining it together in a very unique way. And like, you know, to me that's.

The same as saying that oh, Cursor or Windsurf is a fraud.

Because they're Perplexity.

Yeah, yeah, I mean Perplexity actually has own LLMs now. Like they have their own fine tuned llama Models. But yeah, it's like, it's like calling one of those tools a fraud because you didn't train the model yourself. Like, who cares? It doesn't make any sense. But then, you know this guy Peak here, he's the, he's the CEO. He said, hey, I'm Peak from Manus AI. Actually, it's not that complicated. The sandbox is directly accessible to each user. Each session has its own sandbox, completely isolated from other sessions. Users can enter the sandbox directly through Manus interface. The code in the sandbox is only used to receive commands from agents. So it's only, so it's only lightly obfuscated. The tool design isn't a secret. Basically saying like we're not trying to keep anything a secret. Right. Like when you ask it a question, it's going to figure out which models to use. I guess it uses both Sonnet and Quinn. So based on your prompt, it sort of decides which models to use for the various steps. And then it does use this open source browser use tool, which is a pre created open source tool that will actually go and use your browser before you. Although it's not using your browser when you do use Manus, it sort of opens up a virtual computer and uses this browser use tool in a virtual computer.

Yeah. I think it is interesting that this year, like all the top American AI companies have been trying to push out solutions with agents and they haven't really got much attention yet. And then this Chinese, you know, which is not open source yet, they said it's going to be open source, but this Chinese company comes out and just combines all these tools in an interesting way and they, and they make it happen.

Yeah.

So I, I think that's going to be a dynamic. We're going to continue to see and probably should show the Gary Tan tweet talking about how.

Yeah, yeah. So here's, here's what the tweet says. It says if real, this actually emphasizes the models are plenty smart already. And all the alpha in custom is in custom prompting tool use and clever workflow and evals.

Yeah. And I think that that is probably where things are at is like, you know, even though OpenAI has got a great model and Claude and everyone else, man has figured out how to combine these together in an interesting way first, which, which is exciting because that, it feels like all these models, they're just waiting for people to like build things on top of them, like new things and try new things.

And yeah, again, I don't think it's a Bad thing that they went and used existing large language models and existing open source tools. They've just combined them in a way that other tools haven't yet. Because they went and combined them in a way that others haven't. They made something really valuable and useful to people. And like Gary says here, it just shows these models are a lot better than people are giving them credit for. Right. It shows that these models are already capable of going and just handling work for you. You just need to sort of have the right workflow and the right combination of large language model and tool use to get done what you want to get done.

Yeah, I do. I do wonder, like, why did Claude not do this first? I wonder if it's like out of restraint kind of thing or they just didn't come up with it.

They had. What was it called? Claude computer use. Right. And Claude was kind of the first one to push something like that out. But it was such a pain in the butt to actually use. You had to install Docker on your computer and then you had to run a whole bunch of like terminal commands and install a bunch of stuff. And like, you know, if you're fairly technical, it's, you know, even if you're not that technical, it's pretty easy to follow the directions. It was just a lot of steps that I think scared a lot of people off. So nobody ever really used Claude's computer use because it was like too many steps. And then ChatGPT came out with their operator, which, you know, simplified things a little bit.

Oh yeah, that's true. And it's actually doing something similar to this. It's got like a sandbox where like, instead of you having to put it on your own computer, they've got like a little virtual machine running.

But I guess, you know, Claude 3.7 is just, you know, that was, that was the magic, right? That was the one that, that, that made it work better. I think the Chat GPT models didn't really work that well for this use case or something because, you know, Operator launched a few weeks ago, maybe a month ago. I don't remember exactly when, but it's been out for, you know, three, four, five weeks, some, somewhere in that range. And it was all hyped up the week that it came out. And then I feel like nobody's been talking about it ever since.

Right.

It just sort of like died off. So it must not have impressed people that much. And now we have Manus and like everybody can't shut up about it now. And it's like, well, it kind of is similar to what we already have, but it just seems to work better. It just seems like this combination of Claude 3.7 plus browser use plus, you know, occasionally using Quinn as well was like that. Right. Mix to make it feel valuable and make it feel like it's really working for people.

Yeah, I mean, it does feel like OpenAI they're trying to do so many different things. Yeah, right. Like, like early on they were doing AI art too, which apparently they have a big update coming or whatever. But, you know, mid journey and Leonardo and other people are like totally dominated that space. And so it feels like maybe it's the same kind of thing where, you know, the, the core model companies, they're trying to do so many different things and so a company can focus on one thing. Like, okay, let's make sure that the, the AI models know how to use a computer really well or use tools very well. They can still win by just having like really good focus.

Yeah, no, absolutely. Well, let's, let's go ahead and try out Manus. I want to, I want to try to like run through. Let's. Hopefully it works. It's not overloaded. And then after that I have like a few alternatives that I've come across. Like we mentioned, there's 2 million people on the waiting list, so I, I thought it would be helpful to share some other, other options that you can go and try out if you can't get into manuscript. But what do you want to test here?

I mean, I think a landing marketing page would be probably, you know, a value for business people because you're always making landing pages and then pushing ads to push people to those landing pages to test them. Agents make that dramatically easier to do.

Create a marketing landing page for a.

Website that sells. What should we sell.

Shoes?

I don't know.

Sells shoes. Include social proof research on the best shoes. I just want to, I want to give it some extra details so it's not just, you know, going off and cloning a website. Let's see what else?

Images of shoes and customer testimonials with headshots. Let's go use stock photos for now. Let's see what it does.

So it's thinking, if you already had.

Images, could you have provided that to it?

There is an upload button. You can see there's like a little attach button down here. So you can't attach other stuff to it. Yeah, let's see. I'll help you create a marketing landing page for a shoe website. I'll Include social proof, research on the best shoes images, and customer testimonials with headshots using stock photos. Let me get started on this right away. So now it's setting up a project structure, creating a to do list, updating the plan, and then I believe if I open this, yeah, we can see the plan that it mapped out for us. So here's the plan that it mapped out. Set up project structure, gather stock photos, create customer testimonials, design HTML structure, add css, test and validate landing page functionality, deploy landing page and share with user. So that's what it's doing. And then I can actually open this up and we can actually see it taking actions in our browser right now. So what you're seeing on the right side of my screen is actually it just going, hands off. I'm not touching anything. That's just literally the future of work. Yeah, this is literally just future video right now. It's on the Vogue website.

It's on the Vogue website. We can see every once in a while, it sort of flashes bounding boxes on the screen. I think what it's doing is sort of reading the CSS or something. Like reading the CSS and then, you know, pulling text from within certain divs or something. But it's reading vogue.com right now, I guess, about the best shoes.

Okay. Hopefully it's not just copying their website.

Right now it says it's doing research. According to the task list. It says, research best shoes and current market trends for landing page content. So according to Manus, it's just doing research. It's not. It doesn't seem to be, like, screenshotting. It seems to be, like, using bounding boxes to, like, find where the text is and then reading the text inside the box. Now it's on a site called Birdie. We'll probably have to speed this up in the edit because this is.

I mean, I assume this is not gonna be perfect, but it, like, you can see where this stuff is going, you know, and even recently, I've already found that this has changed how I, like, hire contractors and stuff. Like. Like my own website, lore.com like, I built it myself, but I was having somebody who, if I had. If I was busy and I needed small tweaks, I might occasionally hire that person to make tweaks to the website and stuff like that. And now I found, like, it doesn't even, you know, make sense to hire that person anymore because, like, I just use Claude and just type something in. And it changed my entire website for Me literally like in like five minutes, it made dramatic changes to my website. I added like the Next Wave podcast and all that. I literally just told Claude to do all that.

Yeah, yeah, yeah.

And they did all of it, like five minutes. And I pushed it live and didn't have to hire anyone.

Yeah, well, I mean, as you know, like, I'm actually rebuilding future tools and I hired a developer to help me build it. The developer kind of got really, really flaky and then started telling me stuff that was originally planned, was outside of scope and it was this whole pain in the butt process. I eventually ended up just like letting him go. I was out like basically 10 grand on this redesign. But I went, this is just getting too insane. It's getting more and more expensive. I'm going to cut my losses. And yeah, I mean, I've basically managed to completely rebuild future tools with a better design, like just totally by myself.

Using these AI tools. I think you're going to see this in like so many different areas of business and in creation is so, you know, I've been like doing more and more game design stuff lately and, and been watching videos of Jonathan Blow, pretty famous and notorious game designer. And, and he was talking about how when he's made his games, like, he's talking about the difference between like a gigantic company making a game or a small team or an individual making a game, how dramatically different those things are. And he was talking about how how so much is lost between like a designer or a business person or an engineer all wanting different things or having different things in their head and communicating to that, to each other is almost impossible usually because, like, there's usually a lot of things that are beyond just words and, and it's really hard to express those and maybe even impossible because people have different understandings of different words and what you mean. And, and it ends up, you know, in a different result. I think you're going to see more and more aware, like, okay, your idea of future tools. Instead of you having to work with a designer and working with an engineer and work with all these different people. You know what you want.

Yeah.

And also you'll know very quickly if you like what you've produced and if you want to change it versus having to talk to all these other people who will never actually know what's inside your head.

Yeah.

What you want. And I think you're going to see this in so many different areas where people are going to be able to create amazing new things that, you know, whole committees of people can't really create, but you and your AI partner will be able to.

Somebody was asking me on X, like, but don't you get like really frustrated when you're trying to code something and you run into a bug and you can't get past it and it takes you forever to get past that bug? And, and I went, yeah, I mean, that's definitely frustrating. But it's not nearly as frustrating as having something that I don't like on my website going to my designer or my coder asking him to fix it and not hearing back for seven days. Because if I run into a bug, it's usually like between 20 minutes to maybe like two hours of me going back and forth until it's fixed. If I was working with a developer, it could be two fricking weeks before I hear back from him, plus an invoice because it's out of scope, you know?

Yeah. And dealing with the developers too is like dealing with a mechanic like you, you, you don't know. Like if you're like a hardcore developer, you have no idea what they're saying and if it's true or not. And, and that's actually another thing Jonathan Blow is on. He's like, he's been that same dynamic where like a designer will tell. Used to tell him something to code and then he like disagree with it, but he would pretend he agreed with it, he'd build it in a different way, then he'd, he'd make up excuse why he did that, had to do it, you know, and, and, and there's all these disconnects where if you were able to use AI and be in total control of a product, you will get closer to your actual vision for a product than dealing with a whole group of people.

And I think you're gonna see this.

Next so many different areas.

I think the next like Stardew Valley or the next Minecraft is probably gonna come out of a single person just who's passionate about making the game that they envision and they're just at home vibe coding it up, right?

Yeah, yeah, yeah, yeah.

But anyway, when it comes to the whole Future Tools website, I'll, I'll just seed this right now in next week's episode. I'll break down my entire process. Right. I have like a, I'm actually using multiple tools and jumping around. And Nathan, you and I have talked about this off recording, but I'll actually, in that next episode break down my entire process that I've done to recode future tools. So if you're not subscribed. Make sure you subscribe. I'm just saying that right now, while we're waiting for our agent to go and finish up our stuff for us.

Waiting for it to do our work.

Yeah. So right now we're on the third step out of about eight steps. It did its research and now it's gathering stock photos of shoes and potential models. So that's where we're at right now. It's been.

So it's slow, but you know, these things will get faster and faster. Like probably they'll, they'll get like three to five times faster every year. Forever.

Yeah, yeah.

Which is, which means in like three years we'll have things like this where it'll do like thorough research and you know, competitive analysis and check out your competitors designs and figure out all that for you and put together and create a website in like a minute or something. Like we'll probably have, it'll probably take like two or three years, but. So I guess we need to figure out what's the key business takeaway for people.

I mean, in my mind, the key business takeaway for people is you could go and build the website, that landing page that, you know, that mini SaaS app that you've been wanting to build. And you don't need to go hire a developer anymore, right? You can, you can go and have a tool like Manus go and do the research for you and build the bones for you. Like, I know when, when I've tried to, like I've had software ideas over the years, right? I've, I've been like a solopreneur for the last 16 years now, not working for somebody else. And over the years I've tried to develop various softwares and I tried to get into the SaaS business and it was always such a pain to work with developers or to try to teach myself to code. And now I know I can just go to some of these AI tools and just say, here's what I envision. Go build it for me. Like, I feel like that's valuable to any business, especially the sort of, you know, small businesses and solopreneur types, which I think is probably a big chunk of the people that listen to this show.

Yeah, yeah. Whether you're in a big company or you're solo, like, I mean, if you're in a big company, you know, you may, you may have a situation where you have like all these ideas, but there's no way to express those ideas or test those ideas or make them material in any way. Now with stuff like this, you know, you could be like, okay, well you can't do that during normal working hours. But at night you could just like chat with AI and build like a prototype. And whether it's for your own project or a company, if it's for a company, well, you take it to work and kind of show it to people. And if no one likes it, well, okay, didn't take a whole lot of time to put it together. But we're going to get closer and closer to like where people have an idea and that idea becomes real. And then you can test to see if the idea was actually good or not. Really, really fast. Where in the past, like if you had an idea, it's like, well, okay, am I going to go spend like months working on that idea and then maybe nobody likes it. We're going to get where you can just have an idea, test it in a day, hours.

Have you ever read any of the Gino Wickman books, like Get a Grip or Rocket Fuel or. I haven't, I haven't Traction. So he's got this concept that every business kind of needs two people to be successful. You need a visionary and you need an integrator, right? The visionary is the person that has the big ideas that sort of sees where the, the company is going, pretty much has that roadmap and vision of where they want to get it. But they're not necessarily the person who's going to get into the weeds and write the code and build the website and, you know, do all the outreach. They're not the, like doer in the company. So a company needs an integrator also. And an integrator is that person that's going to either go and, you know, hire and find the people that you outsource to, or they're going to be the person that develops the thing themselves. Or, you know, they're the ones that gets into the weeds and sort of does the building. And to me, this really opens up the possibility for like everybody to be that visionary, right? It really opens up the possibility to like, I have this idea of where I want to go and I don't need an integrator anymore. I can be my integrator. I mean, which is good and bad.

Because everyone's got a bunch of ideas and they're not all good.

And it's also, it's also not great if you're not a visionary type person, right? If you, if you're not an idea person. That's where I think there might be some Struggle in the future with this, like, new. This AI era that we're entering into. I think this AI era that we're entering is really, really, really good for the visionary type people, but it could be really, really sort of catastrophic for the integrator type person, because the integrator type person are the ones that probably have to worry about, you know, what am I going to do, like, with my skill set that I've built now that AI can do it. But if you're that visionary, you're that idea person, you're that thinker, like the. The era that we're entering with all of these AI tools.

Best time ever.

Best time ever for that person, you know? Yes.

Yeah. Seriously, like. Like, I. I've dreamed of this my entire life. You know, I've always had all these great ideas. Sometimes the timing was wrong, you know, and. And I've always been, like, lightly technical and had to deal with tons of engineers and designers and then raising capital for that. And just to think that now, any idea you have, you know, God, I wish I was, like, you know, 18 again. Like, like right now, like, really WISH I was 18, because, like, the. The fact that, like, now or very soon, any idea you have, you can just create it and have some. Some level of control over the whole process. Like I said, like, you know, there's no disconnect there of, like, you got an idea, then you deal with a designer, an engineer. No, you can be part of that entire process to make sure you actually create the thing that you envisioned. Yeah.

Yeah. I'm kind of glad all this stuff started to happen after I already had a wife and kids, because otherwise I'd probably just be obsessed and never leave my house.

You wouldn't have the wife and kids. So there. There's a thing there I like. Yeah, I'm super happy to have my wife and my son. So. Yeah. Yeah, I wouldn't. I wouldn't change anything. Like the whole eight. Wish I was 18. No, I don't actually wish that.

So right now we're on the. The sixth step. It's doing the css. All right, so just move to the seventh step now. Now it's just testing, so we should have something to show for this in, you know, any minute now, because it's just the test. It actually says deploy landing page and share with users, so it may actually deploy it for us. I'm curious if it's going to, like, give us a link that's actually usable or if, like, it's going to give us code that we need to go and like deploy somewhere else.

Next step is it goes and markets it for you, sends you some money.

Yeah, yeah, yeah, yeah. I mean, quite honestly, I bet you can say, you can say go and create some Google Ads for me. So one thing about Manus though is it's not going to log into sites on your behalf. So let's say you need to, you ask it to go check your email for you and like reply to the 10 most recent emails and save them as a draft. Right. You can tell it to do that, but when it gets to the spot where it needs to log into Google, it's going to say I need to hand this off to you. Right. So it'll hand it off to you. You then go and enter your login details and then hand it back and then after you enter your login details it takes it over again and starts working again for you. So it's not going to, you can't like plug in your username and password and have it go and log in on your behalf. All right, it looks like we've got something here. It's still finishing the test and validate landing page but I'm actually seeing.

It.

Called the website Step style, which is actually not a bad name, honestly.

Okay. Yeah.

All right, so we're getting a preview of this site but I don't have like a, an actual like link or place to check it out myself. It looks decent just from the previews we're seeing.

Yeah, I mean it's not amazing but it's pretty standard.

Yeah, it's not like, you know, it doesn't have like, you know, parallaxing background images and animations when you hover and stuff. But it's like a pretty standard looking e commerce site. But I mean, here's the thing though. It went and found all the images, it wrote all of the copy for us, it found all of the prices for us, it did the research on the popular shoes for us. So like, if you're looking at more than like the design is, you know, it's standard. Right. But if you look at all the other stuff that came before it, that's the more impressive part to me. We know that Clyde and V0 and Replit and some of these tools can do, you know, okay, pretty decent looking designs. It's all the stuff before the design that I think is impressive now.

Yeah, we didn't really provide it much either. Right. Like if you were doing this actual in a business use case of like building landing pages, you'd probably Be providing IT assets, giving a clear list of like, here's the kind of landing pages I want to create, maybe giving it some resources or like examples. If you did that, it probably would be able to make like tons of landing pages for you and they'd probably be good enough to test.

One thing that I think would be interesting to test would be like, can it go into like Canva and create an ad for me? Right. Can it like actually like manipulate and like drag the images around and like actually design an ad for me that's like a banner ad or something? One thing I'd love to see Manus do is like, you know, have some sort of like environment variables that you can plug in, right. Where you can actually give it usernames and passwords. And ideally it's in some sort of secure location and, you know, saved to your local computer and maybe not saved Manus servers.

Right.

I don't know how that would work exactly, but it would be nice if it can like sort of do some of the logging in on your behalf. Aha. Okay, so here's what it says. The landing page includes all the elements you requested, social proof, research on best shoes, images of shoes, customer testimonials, attractive hero session section, et cetera. View all files in this session. Okay, so it gave me all of the files. So if we did want to deploy these, what would be the best way to deploy them? Would you say, like, so that other people can actually access this? So here's what I would probably do. I would probably load it into a GitHub repository and then attach my GitHub repository to Vercel.

Yeah, that's what I was going to say. So that's probably not the easiest way though. Like, for like most people, there's got to be some other way. I'm not sure how.

I know.

I know when you use replit's agent, it's very easy, but I don't know if you're not using their agent, how easy it is just to push something live like this. I'm sure there's gotta be an easier way, but yeah, that's how I would do too. I'd create like a GitHub repo and then connect it to Vercel.

Yeah, that might be a little too in the weeds for this one, but I would imagine if you wanted to, you could tell Manus to, you know, push this to GitHub for me and then set up a Vercel account. But you would have to keep on taking control every time it was asking for login details. But it would walk you through it at least, right? It would, it would be like, here's the next step, I need you to log in for me, you log in, you hand it back and then here's the next step, I need you to log in for me, you log in and then hand it back. And then it would actually do those, those, that process of setting up GitHub, setting up Vercel and you know, pushing it off. So it says manage has completed the current task and is ready for new requests or revisions. So it didn't actually seem to give me a live link where I can go and view it, but it did give me all of the files here. All right, so links, no, items, code files, images, documents, all. So we do have all of the files here with the research notes, testimonial, stats, index, HTML style script to do. Okay, so it actually put it on online for us and set this URL. I don't know if that's like temporary or if it's like fully deployed for us and we can just like change the domain name. But here's the website. Step Style Shop Collection, Explore Categories. So if I click Explore Categories, it sort of auto scrolls down the page for me. The numbers speak for themselves. It's got the, you know, global footwear market 495.7 billion. So it actually has some like research stats on the page about the shoe market featured collection. This is based on our research, right? These are trending shoes. The retro runner sneaker, Python print pumps, soft leather loafers, performance runners.

And they give you all the research files too, right? It gives you like all the stuff they collected. Yeah, that's crazy because then, because you could do the same kind of thing I've been using for projects where you could then feed that content into an LM and like kind of have a conversation about it and like, okay, let's talk about the research and then how do we want to change things based on that research? Right, you could start iterating on it. Use this like the baseline for the, for the starting.

I mean none of the links work so it seems to be just like a single page without, you know, any internal links yet. Yeah, but you've got the bones of a front end landing page. We've got the science of comfort with some, you know, research text here. We've got some fake testimonials that it made up for us and added stock images. I mean it actually looks pretty clean. I mean it's not like the, the best looking website ever. But it's, it's clean. Right. It. You wouldn't look at this and be like, oh, this site looks like crap.

It'd be a decent starting point. Like, you know, you, you take something like this into, you know, cursor or something like that and then have a conversation with it about what you wanted to change and then, you know, probably after a day it'd be pretty good.

Yeah. And I mean, hey, if you wanted to create just like a landing page that collects emails, you know, or is in the marketing world, they call it a squeeze page. Right. If you wanted to create a squeeze page that's like a single landing page that collects emails, this would be a pretty easy way to do it, I would say. After seeing it, do you have any like, you know, additional thoughts? But I feel like we pretty much shared most of our thoughts as it was processing.

Yeah, I mean, I guess the only thing is I find it super interesting. I think that probably though, if I had to guess, I mean, everyone's going to copy like how this is working. Like it's not clear that menace is going to be the winner in this space. I'd be, I'd be shocked if Open the Eye and Claude and others don't do the same. I think agents is going to be a core part of the business 100%.

Well, you know, one of the things I mentioned was talking about some alternatives. Right. Just because There is a 2 million person waiting list and they're only letting a handful of people in every day. Well, this week, the week we're recording the week before, you're actually listening to this. OpenAI just released some new features inside their API. Here's the blog post from March 11th. New tools for building agents. So in their API they actually gave people access to sort of create these like agentic. Agentic features in the apps that they're building directly through the API. So there's like the assistance API trying to see if there's like a section that best describes what they just released.

All right, so there's like three different ones. I'm trying to remember what the three.

Yeah. So agents easily configurable LLMs with clear instructions and built in tools, handoffs, intelligently transfer control between agents, guardrails, configure safety checks, tracing.

Oh, that's right. Yeah, yeah. One of the big parts of the release, well, one of them was they updated computer use. Apparently the computer use is doing way better on benchmarks. I haven't had a chance to test it yet. So that's one I Remember, it went up by like 15 or 20% on the benchmarks, I think.

Yeah.

Another one is like the agents collaborating with one another. Like they apparently really improved upon that. And that's part of this release.

Yeah, but most of it seems to be like through the API. Right. I don't know if there's like a front end UI where you can just go and play with this. I mean you can go and use the operator. So yeah, I mean as far as like alternatives go, you do have the, the Chat GPT operator. And then there's also this browser use tool which is an open source tool that you can use. It's actually a Chrome extension that you can use. They do have a free open source version. I've played with it. This was the tool that I used in a previous live stream that I did where I asked it to go to Amazon for me and buy me some toilet paper and it went and added $600 worth of toilet paper to my, to my cart and then I canceled it before it actually placed the order. So there is that. You can go play with it. But browser use is actually what Manus is using under the hood. I just think Manus is a lot better at sort of giving it good instructions where just giving it broad instructions like go order toilet paper is not clear enough. But Manus, I think using the underlining like Claude or Quinn models is better at making sure it won't go do stuff like that.

There's also this other one I came across called Highlight AI. I haven't played with this one yet myself, but they say everything you love about cursor across every app on your computer. So if you look at the app, it's basically like a tab that opens up on your computer and you can ask it to do stuff and it will open up different apps and run programs for you and stuff like that. Again, not what I've tested. Just wanted to show what else is available out there. And then there's this one called pig@pig.dev. and it says operate windows apps with AI. Pig makes it easy to build complex automations powered by AI. And this one actually will take control of your mouse for you, so it's not using like a virtual browser. So I don't know how much I would trust this one yet because it's literally taking control of your own computer, not a virtual computer. This one though, you do install and I believe it's open source, so I'm not 100% sure on it.

I think they could have picked a Better name for something that you're going to trust with your computer to, like, run your computer calling this pig.

Yeah. So it does say Piglet is fully open source and you can check out the code on GitHub. So Piglet is, I guess, the open source version of this. Again, this one actually takes control of your computer where computer use and operator and menace all seem to use like a virtual desktop that it uses that it will go and take the actions inside of.

I think for now, the virtual desktop is the better way to go. Like when these things get really good and you can actually trust what it's doing, I think giving it access to your computer will probably make sense. But for now, I think I would stick with ones where they do a virtual sandbox.

Yeah, I agree. I'm. I'm too scared to put pig on my computer and let it. Like, A, it's a tool called Pig, like you mentioned. Like, that just doesn't make me too confident in it. But B, I just. I don't want an AI going and controlling my mouse and give me back my mouse. I can't do that, Matt. You know, like, I don't want that.

Scenari.

We're locked in, Matt. You told us to do this, Matt. We're gonna finish it. Finish what you started. You told me to finish it.

Yeah, yeah, exactly. But anyway, that's a sort of breakdown of the current standing of the world of AI agents. Obviously, Manus is the big talk of right now. A lot of people creating some really cool stuff. I'm fairly impressed with it. I wish it was a little bit faster right now. It feels a little bit slower than it should be. But that could just be a symptom of so many people all trying to use it at once. Right. If it wasn't so bogged down, it might go a lot faster.

But it is a big moment because, like, this is like agents are working now. Like, there's been hyper on agents. And it's like everyone was saying, like, this is going to be the year of agents. They are working now. And we're only in March.

Yeah. It is officially the year of AI agents. They are here.

Yes. And we got like over nine months of more improvements. So, I mean, like, by the end of the year, agents are going to be amazing. So anyone? Business. I think that's what you should be thinking about is like, be testing these things out, like, understand how they work. And then imagine, try, try your best to extrapolate and like, think about over the next year, what does it mean when this thing is like five to 10 times better, because that's most likely going to happen. And how does that impact your business and how can you be using it? Because, you know, that's what I'm trying to think about daily.

Yeah, absolutely. Absolutely. Well, I think that about covers everything we can kind of possibly talk about around Manus and the current state of AI agents. So we can go ahead and wrap this one up here. If you like episodes like this, make sure you're subscribed either on YouTube or wherever you listen to podcasts. We have a lot more stuff like this coming next week. I'm going to share my whole breakdown of how I've been rebuilding future tools. We're going to talk about some other really cool apps that have been coded with AI recently and break them down a little bit better. So that'll be a fun episode that we're releasing next week. So again, don't forget to subscribe. And thank you so much for tuning in. We'll see you in the next one. Good.

Thank you.