Meta's Amusing Muse
It happened. Meta finally has a product that I really like using again. Perhaps the first since Paper way back in the day? Certainly since everything they've done to try to ruin Instagram.1 Fine, I also quite like Threads, but more so because it's just a version of what Twitter could have, and probably should have, been years ago. And it just so happens to exist in a time when Twitter itself has been turned into some sort of A/B test for surfacing the worst of humanity. Yes, Threads has social mogged Twitter.
I'm already on a tangent. I was trying to pay Meta a compliment after months of mockery. And it's honestly a big one. Because I must say, Muse is pretty great.
I speak, of course, of the personal agent product, not the actual models that power it.2 Those seem pretty good too, which certainly matters, but there are a number of great models out there in the world at this point. What truly matters, at least to me, is having a great product. And that has historically been very hard to do in the personal assistant space. Which is putting it nicely. Putting it less nicely would be noting that the category is filled with carcasses. But I feel like Meta has a legitimate shot at making such a product work with Muse.
ClosedClaw
This is where people will jump out of various corners of the internet to yell about OpenClaw or Instinct or the like. Those products are certainly interesting and impressive in ways. But at least so far, they lack the consumer chops to truly take off with the masses. I wrote about this last March when no one would shut up about OpenClaw in the aforementioned echo chamber that Twitter has become. Again, there are cool ideas there around plugging and playing with your own models on your own machines, but there is no way that regular users were going to, say, buy a Mac mini to set up such a server.
That plus the notion that OpenClaw's founder apparently rebuffed an offer from Mark Zuckerberg to instead join OpenAI made the prediction that Meta would go directly after the space an easy and obvious one to make. They may have even gotten there earlier had the Manus acquisition not been torpedoed by China. It's unclear what, if anything, Meta was able to keep from Manus, perhaps little more than three letters and some learnings for Muse. But whatever Meta has been cooking up over the past many months under the guise of 'Hatch' clearly worked.
Again, it's really good. I got that vibe immediately after I signed up, but now after some time using it quite a bit, I'm quite sure of how good it is, certainly relative to the efforts by their peers.
The key, again, is the product itself. You might assume that given the success of ChatGPT, this needed to be a fully chat and text-based interface, but actually Muse's UI is full of little tweaks here and there that make this decidedly more user-friendly. Most visible, of course, is the avatar.
Avatars
Others have tried this before, notably Microsoft – no, I don't mean Clippy, but their weird cloud blob cute thing for Copilot. With the company constantly swapping copilots for Copilot, that thing didn't last long. But that doesn't mean it wasn't a good idea, at least to try. Yes, others have had avatars mixed with AI for a long time including Grok's sexy bots and Codex has their "pets" (which are now baked into the unified ChatGPT), but I think the key is tying it to a persistent assistant, not just a chatbot or coding agent.
Naturally, I've named mine Moneypenny.
There's perhaps nothing more controversial than anthropomorphizing AI at the moment, but for this particular use case, giving an AI assistant tasks to complete and going back-and-forth around those tasks, it feels most natural to feel like you're talking to something if not someone. And leveraging the avatar in clever ways to let you know when it's thinking/working feels correct too.
Ideas
The other main obvious problem that many such services run into is twofold: the "blank page" problem mixed with the "boil the ocean" problem. That is, in our current Age of AI, these things can do almost anything – and certainly you can at least ask them anything. So what do you say? And what do you ask them to do? Again, Meta has not only some simple UI here, but an entire main area of the service called 'Ideas' that are decidedly just that: ideas for what you should ask your Muse to do.
This sounds simple and it is simple, which is key. But it's also deceptively more complex because Meta is constantly updating this 'Ideas' tab based on what it knows about you – as best I can tell, based off of general ideas but also what you chat about with your assistant but also what services you connect to Muse.
Connections
The last bit is the real key to the service. One way to overcome the 'cold start' issues mentioned above is to have some layer of content and/or context already baked into the product. And Muse being a Meta product... well, they likely have a shit ton of such data. Facebook. Instagram. Threads. All a tap away thanks to your Meta login. But assuming you're using Muse on your iPhone, Meta also has a streamlined way to bring in a lot of data via Calendar, Contacts, Reminders, Health, and Home. And then the core Google services, Gmail, Google Calendar, Google Contacts, Google Docs, Google Drive, etc, are just an auth away. Outlook too, if you're into that kind of thing. And the list keeps growing.
One thing you won't find: Messages – aka iMessage. Apple clearly doesn't want Meta (or anyone else) tapping into that, quite literally. But that's less of a big deal for Meta when you also happen to own both Messenger and WhatsApp. These are particularly useful integrations if you simply want to use the interface that you use for your other chats to communicate with Muse.
Security
But should you feel secure hooking up such things? There's basically two questions baked into that one question. First, should you feel secure hooking up an AI assistant into your personal data? Second, should you feel secure hooking up a Meta AI assistant into your personal data?
I'll be honest that I was pretty hesitant on both fronts. I started slow, with only the Meta products connected. But reading up on how Meta architected Muse a bit, I started to feel more confident. In fact, I think it's fair to give them credit here for the way they're seemingly handling such security with information such as passwords sequestered away in a way that Meta never sees them, and all of your information sandboxed in your own computer instance in the cloud.
Yes, it's their cloud. And yes, you are opted-in to help improve Muse with your usage. But you can opt-out of that, and given Meta's overall scale and expertise, I have more confidence in my data there than with, say, a new startup. Famous last words, I know. But at some point, you either trust the system they've built or you don't. And again, they've put a lot of legwork in here, clearly. Because they knew that if they didn't focus on such security, no one would use Muse, clearly.
I do wish they would do what they're doing with their Muse models where if you opt to help train their service, you get a massive discount on usage. That said, the usage rates right now for the free tier seem pretty generous. And I suspect that will continue because of another natural advantage Meta has here...
Feed
Yes, they have their other core businesses which bring in billions of dollars in profit each quarter. But as you're probably well aware, those billions are now going straight back out the door to pay for the AI build-out. Which is to say, Meta can't just give a product like Muse fully away for free. Plus, they need to diversify their business model away from being so dominated by social feed-based advertising. And so some level of paid upsell for AI power users makes sense, even if the overall premium offering is pretty convoluted at this point.
But there's one more element here which I suspect will truly be the key to making this all work for Meta. Aside from the main 'Chat' feed with your assistant and the aforementioned 'Ideas' feed, there's also an actual 'Feed'. That is, a feed of constantly updating information that you might find useful. For example, right now mine is showing a couple news items around the first benchmarks for Apple's M6 chip and the news that the Cleveland Browns (my hometown NFL team – sigh) have had to put a key player on injured reserve after just one game (a very Browns development – sigh). But beyond these topical items based on your interests, Muse also injects elements from your calendar and email (assuming you hooked them up) that seem timely. Here it notes that I have a package being delivered today which will require a signature. Nothing groundbreaking, but nonetheless useful!
Why not just put these in the main chat? Because all of this would obviously clutter up that key interface. And it would naturally make you lose your place in any current ongoing conversation with your assistant. In a way, this is yet another solve for that "cold start" or "blank page" problem, as you can choose to riff off of any of these topics with your assistant; to get more info or to, say, set a reminder about that package to get pinged about later.
Others, like Google, have tried to do this type of feed in their Google app (and when swiping left on Android devices), but Muse's feed is so much better in just a few days using it, in my experience. And it can easily be tailored with simple natural language prompts.
Advertising
But I've buried the lede. The real key to this feed may be the fact that it's a feed. Who manages and scales such feeds better than anyone? Meta. Yes, historically that's been around social algorithms, but now there's a new type of algorithm in this type of product. And given this type of product, it may be even more lucrative when it comes to Meta's true area of expertise: advertising.
Injecting ads into the current paradigm with AI chatbots runs a pretty high risk of backfiring. While chats that replace some search queries will work for obvious reasons, not all chats have that same intent-based discovery mechanism. Nor is that area the traditional feed which has worked so well for Meta historically. But this new 'Feed' is certainly that, and again, arguably better given all of the information it can, yes, feed off of. This sounds gross, but I'm not sure it has to be. In particular if it keeps your main chat ad-free. And I suspect such items surfaced in this other feed could actually be useful quite often. It could be an actual win/win. And it's so obvious that everyone else is likely to copy such a model.
Goals
Another main area of Muse is 'Goals', ongoing tasks and high-level agenda items that your assistant is working on and/or keeping track of for you. Currently, I have an ongoing fare watch for a flight I'm looking into, a ticket sale watch for Dune 3, a monitor for tax documents coming into my inbox, an ongoing calorie tracker, and more. I suspect this list is going to get quite long and that actually makes me feel quite good, because it's all stuff that, while perhaps not fully off my plate, can be off my mind until such a time that I need to deal with them.
Okay, can't a reminder app just do this for you? Some of it, sure. But the key again is the product here. The way you interact with it and the ability to take the open-ended nature of the AI chat paradigm and squeeze it into something that will make sense for regular users. There's no need for special incantations or Terminal code here. You can just type – or talk – to your bot.3
Mainstream
To that end, I suspect Meta's real rival here isn't OpenClaw but Apple. To be clear, Muse seems about 1,000x more powerful than pretty much anything even the newly upgraded Siri AI can do at the moment. But I'm not sure how much that matters for everyday users. Certainly not more than Siri simply being there by default, connected to all of your data on your iPhone.
If you just need a way to better interface with your phone to naturally dig through email and messages and surface information, clearly Apple is going to have an advantage with Siri baked into every device. But as workflows move more agentic – i.e. you want to use AI to take more actual actions – the advantage quickly shifts towards Meta. Though that assumes the data access is equal, which it won't be. But it also assumes Apple doesn't continue to improve Siri AI with more capabilities, which they will – but undoubtedly far more slowly than Meta does.
Then it becomes a question of who does a better job stitching all of this work together in a cohesive product experience? Both have historically been good at this from a consumer-perspective, but in very different ways! And you can already see how the strengths and weaknesses of each will impact such a battle. Can Meta convince people to use their devices more than the iPhone for such use cases? Can Apple keep Siri relevant and useful in such a fast-moving field of technology?
Other Competition
OpenAI remains a bit of a wildcard here because they've historically done well on the consumer side of AI, from ChatGPT on down. They make compelling products that people naturally gravitate towards, even if they have to shut them down. But clearly the focus is on the business and the coding use case in particular as they race against Anthropic. Does that shift again in the near future? Can OpenAI find their way back to consumer? Might their own devices lead that charge? And will they have an offering that truly competes with Muse?
Amazon, Google, Microsoft, and SpaceXAI are all going after this space too to varying degrees. Muse has perhaps shown them the correct path here; can they take a page out of Meta's own playbook and fast-follow? And the startups are coming too led by the aforementioned Instinct. Will being nimble, not weighed down by the big company baggage help? Perhaps being model agnostic is a key play as it is starting to be on the enterprise front?
The Path Forward
This can still go any number of ways and the key for Meta will be user retention (as getting people to sign up clearly won't be an issue given all their surfaces). Can they make Muse truly sticky? Not just a daily habit, but an hourly one? I'm seeing signs of how this could work, but it's easier said than done as the key unlock may be slightly different for everyone given just how much you can do with such a tool.
And that (good problem to have) will grow as the capabilities of Muse continue to grow. It's honestly impressive that Meta is doing all of this with the Muse Spark model, what will 'Watermelon' – their coming soon flagship model – bring?!
Frankly, I still need more time to trust Meta with, say, signing into my bank account! Even if just to pull statements. Again, it feels like and sounds like Meta has taken all the right steps here. But it's still Meta! If there's a worse tech brand when it comes to user trust, I don't know what that is. Maybe Theranos?
And with that we're back, dunking on Meta again. Just like old times. But seriously, I'm impressed by Muse. To the point that I'm actually using it. A lot.
I'm going to ask Moneypenny to proofread this now.
1 I realize that I'm no longer the target user nor demographic for Instagram. Which is fine. I just wish that was a different product, not something crammed into the carcass of the old product that I loved so dearly. But I also know that ship has long since sailed... ↩ 2 Yes, this is confusing branding-wise, which is pretty par for the AI course. Muse, the product, is powered by Muse Spark, the model. There's also Muse Glimmer (which is "open") and Muse Image. Presumably the forthcoming "Watermelon" will also get a Muse name. ↩ 3 It would be nice to have a proper voice mode here – as in a back-and-forth with voice. When I asked Moneypenny to speak back to me, "she" recorded a voice response for me, which was sort of fun, but a bit cumbersome. I'd bet that's coming soon given that this is all undoubtedly coming to Meta's smartglasses. ↩
