The AI Daily Brief: Artificial Intelligence News and Analysis - Is the Manus AI Agent Another DeepSeek Moment?

Episode Date: March 11, 2025

Manus, a new AI agent from China, is creating excitement similar to DeepSeek by showcasing advanced autonomous capabilities. It went viral overnight, offering functionalities such as website building, financial analysis, and business assistance. Although it uses existing models like Claude 3.7 Sonnet, Manus impresses users through exceptional user experience and task integration, marking a pivotal shift towards highly effective multipurpose AI agents.Brought to you by:KPMG – Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://kpmg.com/ai⁠ to learn more about how KPMG can help you drive value with our AI solutions.Vanta - Simplify compliance - ⁠⁠⁠⁠⁠⁠⁠https://vanta.com/nlwThe Agent Readiness Audit from Superintelligent - Go to https://besuper.ai/ to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Subscribe to the newsletter: https://aidailybrief.beehiiv.com/Join our Discord: https://bit.ly/aibreakdown

Today on the AI Daily Brief, is this new agent the second deep seek moment? Before that on the headlines, a massive new valuation for Cursor. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Welcome back to the AI Daily Brief Headlines edition, all the daily AI news you need in around five minutes. We kick off today with some fundraising news. Cursor developer, AnySphere, is set to close a new funding round that would value the company at $10 billion. Bloomberg is reporting that Thrive Capital is expected to lead the round,

adding to their position in the company. Now, it's been clear for a while that Cursor appears to be one of the big winners or at least leaders in the AI startup boom, but the numbers really spell out how unprecedented their success is. Assuming this deal closes, Cursor will have achieved a 4x multiple from their Series B, which closed at a $2.5 billion valuation last December. That round was itself a massive 6.5x jump from their Series A, which closed at 400 million in August. And the revenue growth really tells the story. Curser was reportedly at 20 million in annualized revenue in August, quind doubling to 100 million by December, and is now humming along at 150 million in ARR. The multiple is now 66 times revenue, up from 25 times during the

series B. That said, for people who have the appetite, there are just very few opportunities to bind to this type of growth, even if you look historically across the history of venture capital. Any sphere achieved $100 million in ARR are just 18 months from the launch of their flagship product. That puts them in extraordinarily rarefied air. Mostly, cursor lovers are excited about the deal. Linear CEO Kari Siren writes, growth shows the power of integrating well-designed and purposeful AI into existing product paradigms. The AI-native empty page plus chatbox approach can work, but it's often far from the practical

workflows professionals and businesses actually need. point being that as opposed to a lot of the text of code tools, this one is actually purpose-built for developers to integrate AI into their existing processes. At the same time, some people are skeptical of the multiple. Autograph CEO, Hari writes, I know lots of edge teams already switching away. An ARR multiple of 66x would be reasonable for the growth rate, but this isn't ARR. It's more like pilot revenue. I hope the cursor founders' employees and early investors are taking some secondary. Ultimately, the short of it is, cursor is very hot right now, and investors are more concerned with getting in than overpaying. Staying on this white-hot theme of coding generation,

a pair of top researchers from Google DeepMind have unveiled a new startup working on next-generation coding agents. The company, Reflection AI, emerged from stealth at the end of last week to announce $25 million in seed funding and a $105 million Series A. Sequoia Capital and CRV led the seed round, while Lightspeed Ventures in CRV anchored the series A. Reed Hoffman, Scale AI, CEO Alexander Wang, SV Angel and Nvidia all participated, valuing the company at $555 million. The two founders previously did things like leading reinforcement training for Gemini, and also helping create AlphaGo, which was, of course, the paradigm-breaking AI that was the first to beat human experts at the board game Go.

The goal for the company is to create autonomous coding agents, which the founders hope will be a step on the path towards superintelligence. Misha Laskin, the company's CEO, said, this is the problem we've been thinking about for over a decade. our team pioneered reinforcement learning and large language models, and we decided that now is the time to kind of bring both of those advancements together and build out a practical superintelligence that will do work on a computer. Reflection already has paying customers in fields that maintain large coding teams,

such as financial services and the tech sector. At this stage, the product is focused on replacing the most tedious work involved in programming, things like migrating software databases and refactoring code. Lightspeed partner, Reverez Jane, said, I don't think it's about replacing engineers jobs. I think it's more about instead of engineers doing grunt work They'll become like architects who will oversee lots and lots of autonomous agents. If you are interested in that way of thinking,

please go check out the Doctor Strange Theory of Agent Work, which is an episode I put out about a week ago now. Moving over into the world of big tech, Apple has acknowledged that AI Siri isn't coming anytime soon. John Gruber at Daring Fireball Report of the News, posting a statement from a company spokesperson which read, Siri helps our users find what they need and get things done quickly. And in just the past six months,

we've made Siri more conversational, introduce new features like Type to Siri and Product Knowledge, and added in integration with chat GPT. We've also been working on a more personalized Siri, giving it more awareness of your personal context, as well as the ability to take action for you within and across your apps. It's going to take us longer than we thought to deliver on these features,

and we anticipate rolling them out in the coming year. John Gruber, the blog's author, commented that it was a Friday sort of wah-wah sad trombone news drop. He added that, reading between the lines, his sense is that AI Siri is being pushed to next year's iOS 19, rather than coming this year. Bloomberg, with the assistance of insider sources, is reporting the news even more strongly. They claim there are, quote, new heights of turmoil in the

AI division. And honestly, with how staggeringly behind they are, there should be turmoil over there. Bloomberg had previously reported that there was a sprint to squash bugs in hopes of pushing a new version of Siri this year, but they now report that those efforts have been unsuccessful, adding, in the lead-up to the latest delay, software chief, cred, Federigi and other executives voiced strong concerns internally that the features didn't work properly or as advertised in their internal testing. Some within Apple's AI division believe that, that work on the features could be scrapped altogether, and that Apple may have to rebuild the functions from scratch.

The capabilities would then be delayed until a next-generation Siri that Apple hopes to begin rolling out in 2026. Bloomberg also received a leaked memo to AppleCare support staff issued on Friday, which said, if customers ask about the timing of these Siri features, reiterate that we anticipate rolling them out in the coming year. Keep in mind that AI Siri was pretty much the central sales pitch for the latest iPhone, meaning that this is not going to be a particularly satisfying explanation

to a lot of those customers asking questions. In a final admission that things are not going according to plan, Apple has apparently pulled their advertisements for the iPhone 16 from YouTube. The series featured the user asking Siri to name the person they had dinner with a couple of months ago, recalling and summarizing a pitch meeting, and creating generative video memories. Apple Intelligence has exactly zero of these features more than nine months after they were first advertised.

Just brutal, man, I am not rooting against this company. All of the products that I use basically are Apple, but boy, they have got to get it together. Anyways, for now, that is going to do it for today's AI Daily Brief Headlines edition. Next up, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC2 and ISO-2701.

Centralized security workflows, complete questionnaires up to 5X faster, and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back, so you can focus on building your company. Join over 9,000 global companies like Atlassian, Kora, and factory who use Vanta to manage risk and prove security in real time. For a limited time, this audience gets $1,000 off Vanta at Vanta.com slash NLW for $1,000 off. There is a massive shift taking place right now, from using AI to help you do your work, to deploying AI agents to just do your work for you.

Of course, in that shift, there is a ton of complication. First of all, of these seemingly thousands of agents out there, which are actually ready for prime time, which can do what they promise. And beyond even that, which of these agents will actually fit in my workflows? What can integrate with the way that we do business right now? These are the questions at the heart of the super intelligent agent readiness audit. We've built a voice agent that can scale across your entire team, mapping your processes, better understanding your business, figuring out where you are with AI and agents right now, in order to provide recommendations that actually fit you and your company.

Our proprietary agent consulting engine and agent capabilities knowledge base will leave you with action plans, recommendations, and specific follow-ups that will help you make your next steps into the world of a new agentic workforce. To learn more about Super's agent readiness audit, email Agent at B-Super.A.I, or just email me directly, NLW at B-Supor.A.I. And let's get you set up with the most disruptive technology of our lifetimes. Welcome back to the AI Daily Brief. If you want to have a sense of just how impactful Deepseek was on the collective psyche of the AI industry. The conversations that have been happening around Manus over the last couple of days offer some pretty interesting insight.

Deepseek, of course, fundamentally changed how people thought about not only the relationship between Chinese and American AI, but their sense of the speed at which American AI companies were evolving, in other words, thinking it was somehow too slow. And it also really reinforced just how important giving even regular users access to the most advanced reasoning, models was going to be when it came to popular perception. Part of why Deepseek temporarily ran ahead of ChatGBT in the App Store was that they were offering access to an advanced reasoning model at that free level. Now, of course, the other reason that Deepseek was such a moment that extended outside of

the builders in AI was the notion that they had trained this model for just $5.5 million. That news came right at the time that Wall Street was trying to figure out if AI infrastructure is being completely overbuilt with a major correction eventually coming down the pipeline. Obviously, those questions remain unresolved, and in the meantime, people are paying a lot more close attention to what's coming out of China, not just to see how far behind it is, but instead, just on the merits of whether it might have something totally different and more advanced. And that's the context for this moment that happened with Manus. Over the weekend, a new agent called Manus went completely viral.

The Discord for the project swelled to 138,000, and some reported that Invite codes were selling for thousands of dollars on Chinese social media platforms. The demo showed a computer use agent that was capable of things like building a website from scratch, planning a trip, analyzing financial markets and generating reports, designing interactive courses for teachers, as well as comparing insurance policies and assisting with business sourcing. In the viral demo video, founder Yi Chao Peak G said, Manus isn't just another chatbot or workflow. It's a completely autonomous agent that bridges the gap between conception and execution. We see it as the next paradigm of human machine collaboration. The Manus team also claims

top ranking on the Gaia benchmark, beating every rival agent in AI autonomy problem-solving tool usage and web interaction. And once people got their hands on it, the rave reviews started to pour in. Indeed, I think the word breathless would be appropriate for how people initially were talking about this. The rundowns, Rowan Chong writes, I think China's second deep-seek moment is here. This AI agent called Manus is going crazy viral in China right now, probably only a matter of time until it hits the U.S. It's like deep research plus operator plus cloud computer combined, and it's really good. Now, as this will come up in a minute, Rowan also makes clear that this is not any sort of paid endorsement. He continues, we noticed Manis gaining some traction at the rundown and wrote about it

in the newsletter this morning, shortly after publishing one of the co-founders reached out with an invitation code, so I dropped my work for the morning and tested it out. Rowan's test included creating a biography on himself and deploying a website based on that. He said that the info was 100% accurate with info up to date as of today, and then he tried a number of other tests as well. Dean Ball writes, it is wrong to call Manis a deepseek moment. DeepSeek was about replication of capabilities already publicly achieved by American firms. Manus is actually advancing the frontier. The most sophisticated computer using AI now comes from a Chinese startup, full stop. Bill O'AlCitu writes, I tested Manus AI. It's the closest

thing I've experienced to a truly autonomous AI agent. I can't wait till this thing can use desktop apps like Premiere and Photoshop. It low-key feels like baby AGI.I. Menlo Ventures Didi Das Manus, the new AI product that everyone's talking about is worth the hype. This is the AI agent we were promised. Deep research plus operator plus computer use, plus lovable, plus memory. Lovable, by the way, is a text-2 code generator that's become very popular, in case you were wondering. D.D. continues, asked it to do a professional analysis of Tesla stock, and it did around

two weeks of professional level work in around one hour. One of the more interesting tweet threads came from McKay-Rigley, who you can watch get more and more excited throughout the thread. He started, Watch a 14-minute demo of me using Manus for the first time. It's shockingly good. Now imagine this in two to three years when it has over 180 IQ, never stops working, is 10 times faster,

and is run and swarms by the thousands. AGI is coming. Expect rapid progress. He continues, Manus is way better than everything on the market, but it's also not going to automate you away right now. I hope the U.S. labs respond with a great wave of releases. I do really want to emphasize that both the agent under the hood

and the actual UI are both incredibly well done. It's legitimately impressive, and as a reminder, I don't do paid posts. I saw the viral posts and kind of went, yeah, doubt it's that good, and boy, was I wrong. He then continues later. Okay, after further use, I'm doubling down. If OpenAI released an equivalent called Deep Task

and charged $1,000 a month for unlimited usage, I'd pay for it in two seconds, creating an entire research report and spec based on my preferred tech stack from latest versions. WTF. Next, he writes, all right, I'm starting to freak out a little bit. I may have undersold this.

LMAO, it's writing a literal step-by-step guide from up-to-date docs with all the code for every, everything. Finally, he writes, Without exaggeration, I'm genuinely being super earnest about this. I think this experience has shifted my worldview a bit. That was basically 80% of what I imagine experiencing AGI will be like.

Literally thought this was going to be vaporware, and now I'm amidst an existential crisis. Now, McKay also pointed out that under the hood is Claude 3.7 sonnet. In other words, Maneth didn't invent some new model that we didn't have access to yet. That'll become important in a few minutes when we get a little bit deeper into our analysis.

Brian Romley sums up, we just moved from chat AI era to agent AI era. The China Deepseek R1A.I. moments done the world. Now we have the Manus moment. But if this is the Manus moment, what does that actually mean? In my opinion, the Manus moment, to the extent that we're calling at that, is not really about China catching up with the U.S. or China exceeding the U.S. Instead, it's about people seeing some of the first expression of what a real agentic experience is going to be like. Agents are, of course, every other word out of everyone's mouth at this point in AI, but we are still very, very much in the infancy of their capabilities.

The agentic type experiences that we have access to, particularly Open AI's deep research, have definitely started to give some people a sense of just how differentiated they're going to be. And in a lot of ways, I don't think it's inaccurate to view Manus as deep research, but kind of for everything, instead of just for research. And one of the really important things here is that the innovation is not, as I mentioned, about the underlying model. It's about how the pieces have come together. Professor Ethan Malik writes, The current frontier LLMs are very good, and their abilities have not been fully explored

even by the labs making them. Too much waiting for mysterious new LLMs, too little pushing what we have. And I think that's a really accurate description of what's going on with Manus here. Manus has embedded these LLMs in a UI that allows them to really push. both be pushed by their operators and to push themselves to achieve more. Ada McLaughlin writes, If your opinion of Manus changed after discovering it's a newer sonnet wrapper and not some trained-on-potatoes underground Chinese lab leak, you've lost the plot.

I don't care if it's a wrapper, it created value, it deserves my respect. Care about capabilities, not architecture. And I think this is a hugely important point. A huge amount of what we interact with from here on out is going to be an experience or data wrapper that sits on top of an underlying model. A lot of the things that feel most innovative and unlocking are not going to be that because of model performance increases, but because of the specifics of how the user experience is brought together. 0.005 seconds writes, the whole Manus AI thing is showcasing what a lot of people have already

internalized. The models are already AGI grade. And the last steps are how nicely we scaffold perception, context memory, and the for loop. If you were at all serious about building literally anything in the AI space, you need to be internalizing this immediately. The models will get better, smarter, denser, faster, cheaper, multimodular, bigger context, more accurate. The cost per token will drop by 90% next year. There is no possible experience with LLMs made non-viable with cost or capability that will remain so in 12 months. You shouldn't be building with today's capabilities, but with next years.

The crazy part is you don't even need to. The current models are egregiously under-exploited. We're currently experiencing a human creativity deficit in AI development. We're not building wrappers fast enough. User experience, context management, memory integrations, tool use. These are your moats. And I think that's the point. When you start to dig in and you get past the first wave of analysis,

you can also find balance takes like this one from AI for Success, who writes, honest opinion after trying Manus for the last three days. Here's the good and the bad. Good, the research it does on the internet, and the reports it generates are incredible. Its ability to run scripts behind the scenes to execute tasks is impressive. The plans it creates to achieve tasks are well structured,

which is why the end results are so good. But on the bad side, it's slow, but I guess they can scale. it could use a longer context window, which would help a lot. It broke in between due to context issues while working on coding tasks. Sometimes the second iteration doesn't work as well, and it just gets stuck on web searches or certain tasks, making it difficult to control. Finally, the coding capabilities are good, but still behind Sonnet 3.7. Ultimately, none of that matters, though, right? The point is that people are having a mental unlock moment with what an agent assistant, that's not as constrained as something like deep research, is going to feel like having in their

arsenal. Now, when it comes to the viral explosion, Manus is clearly smart enough to understand that DeepSeek is created an environment where people are waiting for the other shoe to drop. In other words, where people are waiting for the quote-unquote next deep seek. And there is probably a lot to unpack there around what that says around global geopolitics and competitiveness discussions. But I think when push comes to shove, the Manus moment really isn't about China. It's not about deepseek 2.0. Instead, it's about a true multi-purpose agent 1.0. This isn't a deep seek moment. It's a chat GPT moment, where people are experiencing what's possible in a way that they hadn't been able to imagine until they saw it. The crazy thing for all of you out there is that this is just

the very beginning. I predict that in just a matter of months, what we're calling an agent now with Manus will seem quaint, barely autonomous, unsophisticated in its planning, and a far cry from what we're using instead. But we always remember those first moments. And for many people, that's exactly what Manus has just given them. If you have used Manus, let us know your take. Spotify and YouTube comments are both open. For now, though, that's going to do it for today's AI Daily Brief. Until next time, peace.