Meta’s Muse AI Agent Cost $0. What’s the Catch?

Sam Altman, Elon Musk, they've had similar thoughts where they aren't necessarily being doomsayers, but they are recognizing that there is this like existential risk of AI and like bio-threats and weapons and whatever. And then Meta comes out and they make their Muse avatar mascot, whatever. It's this cute little like sloth stuffed animal. And basically, while everybody else is saying, 'Oh, AI is like this potential society altering or ending technology,' Meta is like, 'Oh, look, our cute little stuffed sloth will like shop for you and help you save money.'

I'm Jack Raines, an investor at early-stage venture capital fund Slow Ventures, writer of the Substack newsletter Young Money, and author of the new book, Young Money: A Field Guide to Wealth and Purpose in Your 20s. In this show, we take a big story in tech or money and ask what it means for your life and your wallet. Meta just released an AI assistant called Muse. It's free to try and the pitch is that it can do more than just answer your questions. Give it access to the right accounts and it can actually do things for you: find a better price, deal with a subscription, or help sort through administrative tasks that eat up your week. AI agents are still pretty new. They really only became useful around December of last year, as AI models had a noticeable step up in coding capabilities. But until recently, using an AI agent often meant buying a separate computer and spending hours wiring together different apps and workflows or hooking up some sort of cloud server with a database backend. And nobody wanted to do that except a very small group of people in Silicon Valley or addicted to X. Muse has actually abstracted away all the complicated setup steps, giving users a 'do everything on my computer for me' tool wrapped in a bow. Mark Zuckerberg's new big bet is that millions of ordinary people are ready to skip all that and simply ask an assistant to get something done for them. If that bet proves to be correct, it will change how we spend our time and our money.

So, let's get into it. What exactly is Muse? Per Meta's launch announcement a few weeks ago, Muse is a personal AI agent. It doesn't just answer questions. It actually does the work. It helps people stay on top of things, takes tasks and projects off their plate, and turns long-term goals into action plans. So, Meta officially announced the Muse launch on September 8th. And in the last three weeks or so, Muse has had an estimated 3.4 million downloads and over 76,000 ratings on the App Store. For context, ChatGPT, which is still the biggest consumer AI app, has about 11 million ratings. And Claude, Anthropic's AI app, is at around 270,000 ratings. So most of the consumer applications for AI so far have been a chatbot. You open ChatGPT, you ask it for the weather or to break down a spreadsheet or you treat it like Google. And for more than 90% of people, that's kind of been what AI has been since 2022. Meta came out with Muse and is advertising it as an always-on AI assistant that can do work. Of course, Muse can be used as a chatbot, but as soon as you sign into Muse and open your account, they instantly want you to connect your calendar, your email, your iMessage, your notes app. It's very obvious that one, they're trying to aggregate as much data as possible on their users, and two, make it very, very easy for the AI to both read from and use all the tools that you connect to it. Of course, Meta already has a bunch of data on you and me through Facebook, Instagram, WhatsApp, Threads, and just browser cookies. What's been really interesting is how many companies have also partnered with Muse off the launch. We have PayPal, Shopify, Expedia, Stripe, Best Buy, Walmart, Instacart, American Eagle, Dick's Sporting Goods, Fanatics, Gap, Wayfair, and plenty of others. Notably, one company who is not working with Muse is Amazon. We'll touch on that later. They also connect to your contacts, your calendar, your email, your browser, your Google Drive, notetaking apps like Granola. You can connect your Spotify, even Peloton and Open, and most importantly, Plaid for all your financial accounts. Basically, Zuckerberg and his team have made it as easy as possible to connect as much of your life as possible to Muse. So, Muse can see all of your data and it can interact with all of your accounts. But the most interesting thing about Muse to me is that they're giving all this away for free. ChatGPT has a free plan, sure, but it uses weaker models and you hit usage limits quickly. Mark Zuckerberg is giving away a hundred million tokens per week, which is a ton of compute.

For power users, you can pay $20 a month for 500 million tokens a week, or for max users, you can pay $100 a month for three billion tokens per week. But, you know, reading between the lines, he wants as many people as possible using Muse as much as possible. He's willing to subsidize that. Neither Anthropic nor OpenAI publish their free allowance numbers exactly the same way Meta does with the tokens per week. But as a user of both Claude and ChatGPT, I hit usage limits a lot. You know, you send hundreds of messages over the course of one to two to three hours and you get rate limited for the day on both platforms. The reason that Anthropic, OpenAI, and these other AI labs tend to have stricter usage limits for their users is they can't afford to just give away all this compute. Something that Meta has that OpenAI and Anthropic don't have is Meta is a massive advertising business. In the last quarter alone, Meta did $59 billion in advertising revenue. The entire rest of the business did $1.5 billion. Meta is an advertising company. Anthropic and OpenAI are not advertising companies. Yes, OpenAI has started launching an ad business and they are starting to monetize their free user base through ads. They are largely an AI consumption business. You either pay for a subscription plan to get access to higher models or you're paying for an API plan where you're getting charged on your token usage. They only get paid based off of how much you're using the models. Meta is getting paid by how much time you spend scrolling on Facebook and on Instagram. Therefore, they can use all that ad money to subsidize their ability to roll out a new AI agent platform. The result: Meta can afford to subsidize demand while other AI players are forced to rate limit their users. And so far, the stock market seems to love it. Meta's stock is up from $613 a share the day before Muse's launch to around $720 a share today at the time of this recording. My take from this: never bet against Mark Zuckerberg.

Zuck was getting dunked on three years ago for going all in on the metaverse. I mean, he literally changed the name of Facebook to Meta because the metaverse was the next big bet. Everybody's probably seen the picture of the cartoon image of Zuckerberg in the Metaverse or Horizon Worlds or whatever they were calling that platform three years ago and there were all the memes going around of, oh, Meta's stock has collapsed 80% while Zuckerberg is trying to make the Metaverse a real thing. The metaverse didn't really work. Meta's original VR bets like Oculus and Quest never really worked. But it doesn't matter. Zuckerberg keeps going after what he sees as the next big thing. And when AI became the next big thing in 2022, he didn't hold back. He bought a big stake in Scale AI. He made Scale AI CEO Alexander Wang their head of AI. And over the last year and a half, Meta has aggressively gone all in on building out their own AI platform, training their own open source models, now launching Muse as their AI agent. However, as we record this, Meta also just announced a new business line to sell to the enterprise. They'll sell AI tools, including Muse, a coding tool, and access to its top models and business-specific agents to different businesses. Former MongoDB CEO CJ Desai will run it, reporting directly to Mark Zuckerberg. MongoDB also makes database software that companies use to store and manage the data behind their apps, which has been increasingly important as AI usage has exploded. Zuckerberg's statement says the business will draw on strengths. Quote: 'Few other companies have models, agents, infrastructure, and years of relationships with businesses.' As of right now, there are no pricing or operating details yet. For context, Meta is spending more than a hundred billion on AI infrastructure this year. In July, Meta's stock price fell 10% after hours when Meta raised the low end of its spending guidance, and investors didn't like Zuckerberg's answers. Meta's stock price dropped 4% today at the time of this recording. But all this goes to show that when Zuckerberg wants to take a big swing, he's all in. And it's very obvious that the current swing is AI.

But that's enough on the enterprise. Let's go back to focusing on AI agents and the consumer. The last few weeks have been the first period in which AI agents have been released and marketed specifically to the average person. Several months ago, OpenClaw had a moment, as did Claude Code, Codex, and Cursor. But all of these were largely AI coding assistants that were being used by tech insiders, nerds in Silicon Valley, and the folks that are terminally online on X.com. OpenClaw was all the rage back in January and February of this year. Basically, OpenClaw was this non-lab-associated AI assistant. But to make OpenClaw work, you had to install it on a separate computer like a Mac Mini. You had to download dozens of programs. There were a lot of formatting issues with having to go into the settings and making sure it had access to your email and this and that. Users had issues with OpenClaw not actually remembering certain things that they had told it and they had to tweak their databases. Basically, the normal user or the normal just consumer was not going to spend the hours it took to get OpenClaw set up and actually wired in with all their systems. And they also weren't going to trust that it wouldn't get hacked, that they wouldn't suffer from a prompt injection, that it wouldn't accidentally leak their data or buy something that it wasn't supposed to. If you had to spend 10 hours to get your OpenClaw working correctly, most people just weren't going to do that. The thing that Muse got right is the onboarding is so, so simple. Within five minutes, you can have every piece of your life wired into Muse and then you can just start asking it questions and it'll do things for you. Some critics would say, 'No, I don't want to give Mark Zuckerberg or Muse or Meta access to all my data.' Me personally, I love targeted ads. I don't give a damn about data privacy. I would much rather Mark Zuckerberg see all my data and send me like better ads, more useful information through Muse. I would like to think that they can have a much better handle on my data than say me building my own privacy systems and trying to use OpenClaw or working with another like YC-backed startup that can barely spell SOC 2 compliance. One of my hotter takes is I really just don't care about my personal data that much and if I give a much smarter company access to the data and I get better outputs because of that, I'm going to do that all day. And I think most consumers feel pretty similarly given how much time we spend on Instagram and Facebook. It's almost like a very boomer mentality to turn off ad targeting. Like, I love targeted ads.

So what does this broad proliferation of AI agents mean for the AI landscape? For the last three years, AI has always just been code. LLMs are good at two things: writing and coding. Because coding is really just another form of writing. The way that LLMs work is just next token prediction. They make a good guess on what the next word or letter or number should be and keep going from there. Coding adoption took off really quickly with AI because one, we have a ton of verifiable training data needed to train these coding models, and two, there's a lot of economic upside in automating coding with AI because so much of our world runs on software now. If you can make coding faster, better, and more efficient because AI can do it better than humans, there's a lot of money to be made and therefore a lot of reasons to invest heavily in AI coding. That's why we've seen Anthropic double down on Claude Code. That's why we've seen OpenAI start pushing Codex as their biggest thing rather than ChatGPT. Silicon Valley realized that coding agents could basically do anything on a computer and for the last year, basically since Claude Code got good in Q4 2025, coding agents were all the rage. What we have not seen until now is broad consumer use of AI. Only 3% of Bank of America customers, for example, were using paid AI services as of February 2026. There was a huge divergence in what AI was capable of doing at the beginning of this year and what the majority of Americans thought AI could do. Muse is the first real chance for a cutting-edge consumer agent to hit escape velocity. I'd like to note that Muse is not the first AI agent platform for consumers. A new startup called Instinct hit 100,000 users within seven months of entering private beta. Instinct is an iMessage-based AI agent that can book flights, read your email, send follow-ups based on your calendar, and it has a bit of a cult following right now, and they're rumored to be raising more funding at a $10 billion valuation. Grokbot, which was launched by SpaceX AI, Cursor, the whole conglomeration of Elon Musk companies now, hit 418,000 weekly users one month after its August launch. But Grokbot was geared toward the enterprise. Instinct has a cult following, but they don't have the same distribution that Meta has. Muse is smoking both of these apps with 3.4 million downloads in its first couple of weeks. While Muse wasn't the first consumer agent, it is the first agent platform that benefits from both one, broad distribution, and two, Meta's ability to give it away for free thanks to its killer ad business. It's only a matter of time before more and more of the big tech players get involved in the consumer agent battle. Apple will probably have a consumer agent. They're partnering with Gemini for the new Siri AI models. Google should also have their own version of Muse considering they have Gmail, they have your calendar, they have Android. It's actually shocking to me that Google didn't beat Muse to the punch on this. There's also some interesting implications of AI agents both being commoditized and accessible to the public in terms of insane spend on compute. OpenAI and Anthropic need to go public now because the private markets are basically tapped out. They keep needing to raise more and more money to buy more and more compute to keep servicing more and more customers. They aren't profitable. They are in cash burn mode. They've been raising hundreds of billions of dollars to continue to subsidize this demand. They're spending a ton of money on training frontier models that quickly become commoditized as the other model catches up. You'll see OpenAI publish, they had a new model come out that breaks all these benchmarks. Then Anthropic publishes another model that beats them. And then you'll have a Chinese model that's 95% as good for 5% of the price. So whatever edge that these companies get with their new model seems to get fully commoditized within just a few weeks of launch. Meta and Google have massive advantages in that they generate a ton of free cash flow from their advertising businesses. When you're making hundreds of billions of dollars with your main business line, you can then afford to give away billions of dollars of compute for free just to get users to use your AI agent over theirs. If you're using Muse and it works non-stop for you all week and you're using ChatGPT and you keep getting rate limited, sooner or later you're just going to choose the AI tool that doesn't break.

One of the more interesting things about the Muse launch is seeing which companies did partner with Muse and which ones have not yet. So, PayPal, Shopify, Expedia, Stripe, Best Buy, Walmart, big one, Instacart, American Eagle, Dick's Sporting Goods, Fanatics, Gap, and Wayfair have all partnered with Muse. You go on, they have a connector, you can plug it in, and you can access all those platforms through your accounts. Amazon, notably, blocked Muse. A lot of users were using Muse to try to buy stuff on Amazon a couple of weeks ago and Amazon went out and blocked Muse from being able to access the platform, claiming it was violating user terms and conditions. There was something about like safety issues and making sure data privacy and blah blah blah. Like it's not a surprise they don't want Muse coming in and screwing up their advertising business. And they also have their own opportunity to keep building out their own AI agents that have Amazon's inventory, that have user purchase history, where if your Muse is connected to your Amazon and it gets your purchase history, that's a lot of data where they can better target goods or in-agent ads. Not even sure what that would look like yet, but Amazon's big enough, they have leverage. They don't have to partner with Muse. Something that is fun to watch is just which companies are very open to agents being on platform, which companies are absolutely against it. We're about to talk to one of my buddies, JC, who had gotten banned from Resy for having an AI agent try to make a bunch of reservations at Carbone. Meanwhile, OpenTable is integrating with Muse. So, nobody actually knows how this is going to play out, but it'll be fun to watch over the next six months which companies are very pro-agent being on platform and which ones are trying to block it at all costs. Here's JC Bahr-de Stefano, a venture capital investor at Better Tomorrow Ventures.