From the lessons learned while building (and sunsetting) an AI browser, to how Manus executes complex multi-step tasks in the cloud from just a simple prompt, this talk is a must-watch for anyone interested in agentic AI, no-code productivity tools, or the next evolution of human-AI collaboration.
Transcript
Tao: So my name is Tao, and I'm co-founder of Manus AI. And right now, I'm acting as our chief product officer, and also in charge of this go-to market strategy and the partnership scene. And yeah, that's me. And my nickname is Hi Cloud. Actually, you can find me on any social network with this nickname. Yeah, I've used it for more than 20 years. So with this nickname, actually, you can find many dark histories of me on the internet. So let's get started. So I just pressed button. OK, cool. So the first topic is always about what is Manus? Because later, we're going to show a lot of demo videos to you. So I won't describe the whole product scene. Because you'll see demos, and we will send many free credits tonight. So after this event, you can try yourself. But why I still want to explain what is Manus, because this word, Manus, actually, after I read this Manus, I spent 22 days in the United States. And I found out that even in the United States, not many people know what this word means. So every time I have to explain it, actually, it's a good story. We choose this name, Manus, just from MIT's motto, which is "Mens et Manus," which is an old Latin word. It means man and hand. Manus is the hand in Latin words. Why we choose this name from MIT's motto? Like all three co-founders of this company, we never went to MIT. But we just borrowed this sentence from their motto. Why we want to choose this word is because we think for the whole past two and a half years, all these frontier AI labs, like OpenAI,, Google DeepMind, they are all building the smartest man, which is the brain, for the whole world. We believe that all these frontier models, they're super smart. They're super smart. But the problem is that, as a human, we can't make real impact just with our brain. We still have to use our hands to interact with the physical world and then we can make impact into the real world. So we think right now the problem is that we've already have the smartest brain in the world with like the frontier models from OpenAI, from Anthropic. They are super smart. But the problem is like you just you hire some like PhD level intern in your company but you never give him or her a computer. You just give him a paper and pen and ask them to solve very complex tasks for you. That's exactly what is happening for the past two years. It's like we already have a very smart brain, but we never prepare the proper tools for them. So actually, right now, these smart brains can make real impact into the world. And the Manus is us. We want to build hands for these smarter brains so we are not just another model company. We don't train models at all. Yeah, at the foundational model layer, we are using like Anthropic and the Google Gemini models. We don't train models. But we want to be the hands of this smart brain. Then we can make a real impact into the world. That's why we choose this name. Yeah, I think it's a good story. Yeah. So let's move to the next page. It's about we choose this name, Manus. And why we want to build this thing in the first place. Actually, from last year, I think it's last March, we started an internal project, which is, because this company, Manus' parent company, the name is Butterfly Effect. And Manus is our second product. Before Manus, two years ago, we have our first product called Monica2IM. It's kind of like a Chrome browser extension. just leave inside your browser. Then you don't have to do a lot of back and forth copy pasting before Chat GPT and your working apps. So you know, Monica is just a plugin, a browser plugin. Leave just in your browser. So after two years, actually, Monica is kind of a success. We have like 20 million users, and it's generating 15 million AR for us. So it's kind of, I think it's the best in its category. But the problem is all these three words, a Chrome browser extension, each of them just lowers the ceiling for us. Because not everyone is using Chrome. Not everyone is just using browser for their work. And talking about browser extension, I think for many normal people, they don't know their browser can install an extension. So we want to see what's the next big thing for us. So from last March, we started a new project, which is our AI browser. Right now, there are a lot of companies that are doing AI browser. But we started doing that from last March, which is like 15 months ago. So after seven months of work, which is from last March…wow. So maybe someone can close that door. Yeah, thank you. So from last March to last October, we put seven months of work into the AI browser thing. But we never released that. That's why all of you never heard of that. But why we sunset it just before one week, which is the original release date, is because after seven months of work, we are the creator of the AI browser, and we are using it every day. And we found the whole experience is kind of weird. Because when you are using the AI browser, it's like the AI can control your browser. It may look cool if you are just watching in YouTube, yes, wow, it's cool. AI is controlling my computer. But if you are the one who is sitting behind your computer, in front of computer, you'll find the whole experience is kind of weird. Because once you hit some button, the AI started. And you have to keep your hands off the keyboard. Because any movement to the computer will break the AI process. And in the meantime, you still have to keep staring to the screen. Because you don't know when AI will finish its job. So the whole experience is like, you click some button, and it's like this for minutes. It's totally weird. The AI, he is competing with you to use your computer. It's like you hire someone into your company, but you don't give it a computer. And you two use only one computer. It's totally weird. So we think this is not the experience we want to deliver to our users. So we decided to sunset that thing. But during we're building our AI browser project, which is from March to October, just in the middle of our project, there is a very big thing coming, which is the Cursor. Like how many of you have tried Cursor? Oh, a lot of you. Yeah. They told me today is not about tech industry. But I don't know why most of you have already tried Cursor. But we three co-founders, we are all coders. Even I'm a chief product person. I'm a product person. I started coding when I was nine. So it's 30 years at coding. So we found Cursor is kind of very, very cool product. It's like, help us write some language we never know, you know, things like that. But the most interesting part is not us coders using Cursor. The most interesting part is watching our friends, watching our colleagues who are non-coders. They are starting, like my wife, she's starting to use Cursor to solve her daily tasks. Like, you know, just give it an Excel file and ask Cursor to use some Python library to do data visualization. And me also, yeah, I often just put a lot of video files to Cursor, asking it to convert them into audio files. So you see, in these use cases, these non-coders, they never write one line of code in their whole lifetime. So when they are using Cursor, actually, they don't care about the left side of Cursor at all, because they can't evaluate the code. The only thing they do is keep pressing the Accept, Accept, Accept button, right? I see a lot of people are laughing. I think that's how you use Cursor, right? Keep pressing Accept, Accept, Accept button. So actually, that's the moment we come up with some ideas that maybe we should build the right panel of cursor and the companions are learning from our AI project, which is we shouldn't let AI to use your local computer. We should let the right panel of Cursor, but that is running in the cloud. So it won't compete with your computer, right? It's like you just assign a task to Manus, and the Manus will just finish its all job in the cloud computer. So it's like once you assign your task to Manus, you can just put your phone back to your pocket, maybe like five minutes or 10 minutes later, when the job is done, and we will send you a notification, and you get the result. That's last October. That's the original idea of Manus. And after four months of development, we released this thing in this year's March 5th. That's kind of like how we come up with the idea and why we want to build Manus in the first place. And right now, I think I build this product for one of my personal intentions, which is I'm a very typical adult ADHD-er. For being an adult ADHD, which means you will have a lot of ideas, I have a very long to-do list, which contains all my daily inspirations. Every day, I was like, oh, I want to do this. Oh, I must do that. Oh, this is great, this is awesome, I should do this. But for that very long to do, I never get it started. Because for being ADHD, it's very hard to organize your life and get everything done. But after building Manus, it's like every day when I have some new ideas, I never put it into my to-do list. I always give my ideas to Manus. Manus will get my ideas into some working prototype, given it's a slide, a video, or just a website. But this is a very good starting point to get started, but not just an item on your to-do list. So I think that's the whole reason about why this company wants to build this thing in the first place, and why I want to build this thing in the first place. So that is the story behind Manus. And after we released Manus, we spent a lot of time around the world, like me personally. I spent 22 days after we did Manus in the United States, from west coast to east coast. We hosted many user meetups there. And after that, I traveled to Japan, like Singapore, Saudi Arabia, Dubai, and Europe, a lot of places to meet all these users. Why we want to do all these things is because we think Manus just defined this category, this general agent category. But to be a general agent, which just brings some kind of other programs too, which is we can do anything. But when you say to users that we can do anything, the first program coming from users, OK, what's my first task? Yeah, that's a problem. Why we want to meet users around the world is because we want to hear from you. It's like, OK, what's your problem? Especially for some industries we are not very familiar with, we just want to hear from you. What's your problem? Can Manus solve your problem? We want to work together to say that. Yeah, that's why we host a lot of user meetups around the world, like tonight's event. And the whole thing about Manus, I just saw a lot of people are just raising hand, like you already are Manus users. But I just want to expand a bit more for someone who haven't tried Manus. So Manus is just like another chatbot. Manus is just the LLMs. It's like a large language models, but plus agentic architecture we designed by ourselves. And we some, I don't want to use the modular word. We often use the atomic capabilities. Yeah, so in Manus, it's like we have a lot of modular capabilities. Like we can generate slides. We can generate audio, video, images, blah, blah, blah. We have a lot of things like that. But I think you must see some AI image generation, AI video generation tool on the internet. So what's the difference in Humanus? It's like we are not just a simple tool. We will use our agentic architecture. And with the help with all these frontier LLMs, we can get the task done, agentic by you, which is like you don't have to care about it. You just assign the task to us, and you can leave. And the management will have its own plan and execute the plan step by step by itself. I think maybe it's hard to understand, so I will just start with some very simple example. Like last week, we hosted a hiking event in San Francisco. And when our colleague are hosting this event, they use Manus to prepare the whole event. Like, first, they did a poster to put it on Luma. And there are many AI poster, AI image generation tools out there. But the whole thing needs you to write a very good prompt, right? OK, what elements? What's the style? A lot of things. But when you are using Manus, you can just keep it very simple. This is the original session. Yeah, I can just replay here. Which is like, you just give our logo to Manus and ask it to generate a poster so we can put it on Luma. So Manus will first analyze the style of our logo and decide what styles he should use to generate the image. And then Manus will use our image generation capability to deliver this whole poster to you. So it's like a very simple prompt. It's like, OK, just use my logo as an inference to generate a poster for this event. You don't have to care about how to do prompt engineering for the image generation tool. Manus will solve them for you. And the second example is because we have to plan a trial trip about the path, where is the start, where is the end, how long it's going to be, blah, blah, blah, a lot of things. And you can just use a Manus. You give it a very simple prong. And the Manus will, in these use cases, you will say…oh, sorry. This is the last one. Yeah. It should be this one. Oh, sorry. Isa, do you know how can…oh, yeah, yeah. I can move the mouse here. Yeah, sure. Yeah. Yeah, OK. Yeah… So in this case, you can say, after you ask a master to prepare a trail, master will go out to search all these websites about what is the best hiking trail in San Francisco, and then use all the information he found on the internet and deliver the final plan of the whole hiking trail, and then turn this hiking trail plan into images so we can send to all these attendees. Yeah, just a very simple prompt. So this is kind of the difference between using Managed Days General Agent as a chatbot. It's like we focus on the deliverables. We focus on the results, which is like you assign a task to us, we can guarantee that you will get some result, not the intermediate materials or steps. Yeah, so that's a difference. And also, my people, they are also using Manus. It's like, OK, go to my Luma account and download all the attendees' information. And maybe you can just analyze all of them and give me some insights with the deck. So in this use case, you will see, because we have a fully functional virtual machine running in the cloud, so actually Manus can use its browser to log into your Luma account and find all these attendees' information, download it into his local computer, and use some Python scripts to analyze them, and then convert all these things into a very insightful deck. So my marketing team can have some insights for the whole attendees. And in this session, you say, just use its own browser to browse some information, download it into his local file system, and use the Python script to do some analysis, and then turn all this information and the charts into a very insightful and good layout slides. And all these have been done under one prompt. So it's a totally different game. So yeah, this is for the… I should yeah... So this is another example. Yeah, I think the host just mentioned WinServe, right? Actually, I'm a paid user of WinServe, so that's sad. Yeah. But my colleague said that he just saw that WinServe is now on bound calculation. But we just heard some news last week. This should be acquired by OpenAI, right? And then Google came out. So I wasn't following the news that close. Can you research on the news and tell me what happened? And in use cases, you will say, Manas will go out, in this use case, you will say, Manas will go out, search all these views for you, and blah, blah, blah, blah, say a lot of views, and come back with all these informations. And then Manas just decides to give you a slide, deliver your slides, that with the exact timeline of this whole event. You say, yeah, what are the key players? What's behind the story? What's the timeline? It's very easy to understand something. It's not just another deep research tool, give you a very long article to read. Later we will have.. So it's not like another deep research tool, which gives you a very long article to read, but we can just express all this information in a very good format. So yeah, that's the case. Yeah, I should click Next. Where's my mouse? OK, here. Oh, no. The mouse moves so fast. I can't catch it. Yeah. OK, here. Oh, yeah. OK. Yeah. And in these use cases, it's just another example. Because we have a feature we call the schedule task, which is that this task can run on a daily or weekly basis. So you can just set up a schedule task on a daily basis, like scan some marketing news for blah, blah, blah, blah, for me, and every 7AM or 8PM, and convert that into a podcast. So every morning, you will have a customized, personalized podcast just ready for you every morning. Yeah. So in this use case, you will see it's like, it will just go out to fetch all this information out there. And..oh, sorry... Yeah, it will go out to fetch all this information and build all these views into a podcast script and use our text-to-speech capability to turn all of this information into a podcast like this. But yeah, I don't have time to show the full podcast. But in today's demo, we all have the original session link here. So after this event, we can just share this. You can just, all these link sessions replaced just by yourself. Yeah. Yeah, so the whole thing, I think, how Manus works, is really like to work with a real human, which is like, you can get started with some very simple things. It's like, OK, Manas, can you just generate some slides for me, just generate videos for me? And this is very easy task. And Manas can deliver that. And after that, Manas just did some very simple things. And you will see, behind all these simple tasks, actually we have a lot of tools and the data APIs behind that. Like if you are working in financial, you will see we've already paid for some private databases and paid APIs. So actually, you can access real-time financial data through us. And then how can you trust another person? Maybe you see how he works, and oh, he works really great. So actually, we have full transparency with our sandbox and the virtual browser. When you are using Manus, you can see each step of how this agent works. And also, eventually, we have a knowledge system, which is like after you use Manus more, and Manus will learn your preference. So next time when you ask Manus to do some type of task, Manus will answer the task with your preference. Like for me, I always like reading files in PDF format. So in Manus, you don't have to ask the team to add a new feature, which is like output all this content into PDF format. You can just tell Manus, next time when you deliver some documents to me, I want it in PDF format. And then Manus will suggest a lot to you. After you accept the lot stored in your personal database, next time, oh, not next time. Every time you ask Manus to deliver a document, Manus will deliver it in PDF format. So this hosting just works like that. And I know today not everyone is from tech industry. And actually, we think Manus actually is very great for every industry. And tonight, we want to show some different use cases in different industry. And I will ask our chief of staff, Parker, to introduce all these use cases here. Yeah, so Parker, here's your time. Yeah, so Parker, here's your time.
Parker (24:09:17): Thanks, Tao. Thank you. Hi, everybody. I'm Parker. I'll give a quick intro on myself first. I joined Manus the day after we launched the product. And what convinced me to jump on board was the product. I was a venture investor before when I met the Manus founding team and I got my hands on the product. The first thing I did was just take a project I would have had to do myself previously at work, research a bunch of startups, write an investment memo. I just gave that to Manus. And in minutes, Manus accomplished what would have taken me maybe half a day to do myself. So that kind of blew my mind. And I thought, let's see what else this tool is capable of. Let's try something I could have never done myself. You know, I'm not a software engineer. But I said, hey, Manus, make me a game that I can play. I don't even know what went into that process. But a few minutes later, Manus spits out a playable game in the browser that I share with all my friends. We had a competition, a tournament that day. I bought milk tea for the highest score in my group chat. It was a blast. And I thought, I have to be part of this. This is the greatest product that I've ever seen in my life. And so joining the team have been helping out a lot with growth and operations since then. And tonight, what I'm going to do is show you some use cases across industries. Manus really is for everybody. Manus is for not just people like Tao, who've been coding since they were nine, but also people like me, who want to get stuff done personally or at work and have never written a line of code in our lives. So I'm just noticing that what's on the screen here is not what's on my screen. And I'm going to see what happens if I click Next. OK, perfect. This works. Yeah, we're going to go across some industries. Whatever it is that you do, Manus can help. And let's see how it's working for people. In education, if you're a lecturer, previously, you might have had to. I'm going to make sure our video is playing while I talk. Whoa. There we go. Maybe? OK, cool. You might have had to create lesson plans yourself, done things like that. With the Manus agent, you can see a 16-week syllabus for a freshman-level biology course here. The research happens automatically. The documents are created automatically. It's delivered for you in minutes. We're going to do this quick, because I know people want to go into that head-to-head comparison with the ChatGPT agent later. A personal use case. A lot of us do shopping. In this example, this person was in a bookstore, saw a bunch of different interesting books, took a picture of them, asked Manus, hey, which one meets these criteria for what I might be interested in? And you'll see Manus recognizes the image, goes onto the internet, looks up reviews for each of the books included, goes across a bunch of different data sources, and evaluates, based on what the user asked for, which book might be the most interesting for them to read. Every time I buy something now, I'm asking Manus first to go compare all my options. I tell it exactly what I want, and I get a really high quality suggestion on how to make the best purchase. So super useful thing for Manus to help with. Admissions officers. I know we're in San Francisco, so probably all of you have jobs with Zuckerberg knocking on your door trying to give you a $100 million offer. You're not thinking about grad school, but some people are. And if you're an admissions officer at one of these grad school things, you want to hunt hot talent. So you ask Manus, find me candidates that meet some criteria. And let's see what Manus does. Manus is going to go off. It's going to access both public data, as well as, I think in this case, the person had logged into the Manus Cloud browser. And so Manus can look at their LinkedIn. And Manus can find high potential candidates, create a list. Let's wait till the end of this video and see what, I think, format the results are in. Might have been a website, might have been slides, might have been a document. Manus can do anything like that. You can ask, or you can let Manus surprise you. Either way, it's super valuable. If we get to the end of the task here, yeah, we can see document here with some information on the work that Manus decided to do, and then a list of candidates at the bottom. But not just in education. Realtor. In this example, what we've asked Manus to do is go out and find some luxury listings. Anybody here who's considering that $100 million offer might also need a $15 to $20 million home in San Francisco. And Manus is really great at, here it's on Sotheby's real estate site, evaluating different options. It's going to take all the data that it finds online. It's going to build a website that just makes it super easy to understand the options that it's found for you here. So another use case for real estate agents, I'll say something we personally did. Manus, we're all based in Singapore. I was apartment hunting for our CEO there recently. And when the real estate agent found out that I was looking for an apartment for our founder and CEO, the price of the apartment she was showing me went up. But she also got excited about showing us what she could do with the tools. So she took the information she had about listings she'd shown me that day, she dumped it into Manus, and she said, create a custom slide deck for this client to pitch them on the few locations that I showed them today. And so we got a Manus deck with super detailed information telling us why the neighborhood was so exciting, why the exact places that we'd toured that day were so interesting. You know, Manus went on the internet and found floor maps of every place that we had looked at itself, showed that to us. So real estate, another industry where people are getting a lot of use in the real world today from using Manus. In finance as well, stock pitch, personally and professionals, you know, people want to know what kind of stocks to buy. I'm not going to say you should make your investment decisions based on what Manus tells you. In fact, don't do that. We don't want to be liable. But if you want to organize some information, build some financial models, maybe you don't have that skill set, but you want to see how different assets compare, just ask Manus to do it. A 10-minute stock pitch on Duolingo is a super easy task with Manus. It's going to take a few minutes here. And I really want to get to the end of this so I can show you what the results look like. There we go. This is what I've been trying to do all night. Cool. Here you can see some slides that Manus has built. This interesting thing about Manus slides is not just that it is beautiful. These are pretty good looking slides for taking a few minutes to build them. Manus thinks you should hold the lingo stock. But also the content in all of those slides is real. You know, Manus has access to financial data. Manus has access to the web. So Manus first does the research, then builds the deliverable slides in this case for you. We're almost there, I promise, to the comparison slides. Competitor research. If you're a marketer, let's watch this one. And let's skip right to the end to see what Manus does. Sorry, everybody. I'm really struggling with the mouse. Cool, here we go. Maybe this is a blast from the past. Somebody was trying to compete with Uber on rideshare, and they wanted to analyze the marketing work that they'd done so that they could come up with a response. Another example of Manus slides with really high quality content based on proprietary information about a competitor's marketing campaigns that Manus went out and found itself. So a multi-step process that might have taken a marketer a lot of work themselves, now Manus gets it done just like that. And in the interest of time, I'm going to go all the way here. Cool. Let's watch this. And is this the first one? Yeah, OK, interesting. So we built Manus. We launched it early March. Like I said, I joined the day after that. And it's been a wild ride since then. We were really excited about what we've built. We imagined a lot of other people would build similar things. We also were talking to other agent builders, investors. And the question was always, what happens if OpenAI just builds this tomorrow? So we had to wait four months for it to happen. But finally, just yesterday, we got the response, what happens if OpenAI just builds this? And it's so early for both of these products, for the whole market in general. There's so many people that have never had an agentic experience. And so the story is far from over. But let's just watch side by side on a few tasks and see some of the different results. In this case, the prompt that's been given to the Manus general agent and to the ChatGPT agent is to create a data visualization dashboard web page to showcase Labubu's product performance. For anybody here who is not terminally online or maybe a little older, it's a toy. You can buy it for your kids if you can get one in stock. But anyway, watch Manus here. It's going to show you what it's doing. Same with the ChatGPT agent here. Little bit of a different decisions being made on user experience. But let's skip forward, because some of these tasks take some time. Both of them writing some code here. This one took ChatGPT four minutes. And if you download that, it'll open in your own browser. I tested this on the links myself earlier. And we can get to the results on the last slide. But in this case, Manus took a little longer. I think Manus spends, what is it, seven minutes on this task or something. Let's see the results of each. Never before shown, by the way, our team in Singapore has been up all night running tasks on all of the competitive agents out there and seeing how they compare. But yeah, Manus on the left here. I'm biased, but I like the report a little better. You should go try both of them and see how the quality of the research, the quality of the output, the beautifulness of the presentations, that's a technical term, see how it compares. I'm going to pause. Next slide. There we go. Okay. Same thing here. Another comparison. And it's going to be asking these agents to go out on the Internet, find some publicly available data, analyze it, and then present it to you in visuals. We can watch them work. And in this case, I'm curious how long it took. Manus was done in 10 minutes or so. And the ChatGPT agent took more than 30 minutes on this task. So interesting. But let's compare the results. The request was to build a financial model. You can see several tabs on the work that was done by Manus over here. And then the presentations. The OpenAI version. You know, I'm not even going to tell you my biased take here. But it's interesting to watch them complete similar tasks. You can hear the team talking. Video is filmed overnight from Singapore. I've lost my, I've lost my. Okay, I'm on the left. All right, everybody. We're going to skip right to the results of this one. More slide presentations as a result of some research done in the direct-to-consumer skincare sector. Madison OpenAI, never before seen head-to-head comparison right here. Cool stuff. I'm way over time. We're going to go ahead and finish up. The other fun thing here is that if you all scan this QR code, free credits. Go try the tools. I don't know how many of you are pro-subscribers to the ChatGPT version. That's the only way to try it today. But you can try the Manus agent for free. And if you scan this QR code, you'll get even more free access. But thank you, everybody. Yeah. Give a round of applause for yourselves. Appreciate.
Comments