I literally groaned when I saw that Meta released their own AI assistant called "Muse" that will basically do everything for you. Just to jog your memory, the CEO of Meta is Mark Zuckerberg, who absolutely cannot be trusted with all of the data that Muse actually wants to do the stuff it advertises.

The intro video is particularly funny because it shows the same tired use cases that all of the out-of-touch tech executives like to approve in marketing videos. Stuff like blindly buying presents for people and scheduling travel. You know, things that a sane human would want to participate in? The video shows Muse telling the user that their flight is delayed for two hours, then offers to rebook it seemingly a day later. Like yeah... maybe extending a two hour delay by 22 more hours wasn't what the user wanted?

But I figured I'd sign up anyway if only to burn some tokens and make Meta pay for some electricity out of their own pockets.

The onboarding was fairly quick, I just gave it my email address, then told it to ABSOLUTELY DO NOT LINK ANY OF MY OTHER META ACCOUNTS TO THIS ACCOUNT. I am not connecting anything to this AI if I can help it. Then I just started messing with the agent. I saw that I could change my agent's name and avatar, so of course I changed it to Mark Zuckerberg riding a hydrofoil with a bunch of sunscreen on his face.

As I said before, I absolutely do not trust Meta with any of the built-in connectors like email, calendar, payments, health, etc. If you do then I'm happy for you. So I limited my testing to things that don't require my personal data. I had Muse try shopping for a 4TB SSD for me since I've been running out of storage. It was actually able to browse through a bunch of stores and even Facebook Marketplace for used ones. I had it look for trending dog related Reels on Instagram to give me ideas for my own dog's account. And I also had it do other research for me, like shopping for powered USB hubs.

I am sad to say that Muse is actually pretty competent and fast when it comes to this stuff. I think the main advantage it has over other things that I've been using like OpenClaw and Hermes Agent is that it has the infrastructure and expertise to circumvent the typical bot-blocking stuff (I saw it click a captcha checkmark) that I see when I'm trying to run my own agents. Interestingly, I had it hit one of my sites on Cloudflare and it looks like they're not identifying as a Meta bot, but are using Fastly with a regular browser user agent. I guess that's somewhat legit as they're doing it on behalf of a user, but still seems kinda sketchy to me.

For a normal user who isn't going to install Hermes or OpenClaw on their computer because all they do is scroll Facebook all day, this would actually seem quite useful. I can see this being a compelling product, especially since it'll obviously slurp up all of the purchasing intent, personal information and everything else it can get from its users. I mean, Meta already knows I need a new SSD and USB hub now (and so do you, I guess).

If you're curious how I have my stuff set up, I'm currently running Hermes Agent on an isolated VM. It does have access to my email but I only have it running a local LLM when querying it (Qwen 3.8 27B at the moment). I do run web queries on it using my ChatGPT subscription simply because the local LLM I'm running is a bit slow for things that I don't care too much about in terms of privacy.

As much as I am morally against Meta releasing this and believe it'll be a terrible thing if people start using it, I am also going to admit that this is a pretty good product. It hides the complexity and only surfaces the answers. I don't like that but a normal user would. I think ChatGPT has a product somewhat similar to this, but it isn't nearly as simple or intuitive as Muse. The model that Muse uses isn't as smart or persistent as ChatGPT's are, but they probably don't need to be, either.

So overall, good job, Meta, you bastards!