I've used it for a few weeks. It's good
I've been using Muse for the last few weeks, and I can confidently say it's a good product. I have it hooked up to my personal email, I use it to plan my days and my schedule occasionally, and to take care of the stuff that shouldn't take up my time in the first place (like reading through all emails, refunds or making sure I get invoices sent to me).
It's also really simple. If you're someone who is a non-technical founder who doesn't know code and gets nervous opening an IDE (though, I don't know why, but some people do), it bridges that gap really well. It's sort of like Claude Cowork for people who would never open a terminal.
My one gripe is Meta itself
Meta isn't exactly the most trustworthy company when it comes to your data. Doesn't help that Muse makes planning and hand-off easy. Almost too easy, to the point where you should be asking some questions. You can give it your calendar, your notes and your call follow-ups, you can let it buy stuff for you, but who owns that data, and how is Meta using it?
If you're a builder like me, I'm sure that you're already thinking about this for your own product and/or service. You've got your Supabase (or any database for that matter), a table somewhere with client data in it, a privacy policy; the whole nine yards. I think you should DEFINITELY turn that same eye on the tools you use yourself.
When you begin to peel back Meta's history with data, it gets a little bit murky. Cambridge Analytica, Russia-linked accounts reaching about 126 million Americans, a business built on ad targeting. It does not mean that Muse is doing anything wrong today, but it does mean you should definitely read the settings and privacy policy (boring I know) before you connect your inbox.
Muse's personal agent acts for you, so whatever mistakes it makes are yours
A personal agent is, by definition, something that acts for you. You as the human person give it a task and it does it, on its own or close to it. When a chatbot gets something wrong, at worst, you get a bad answer. When an agent gets something wrong, it does the exact thing you do not want it to do (like tell a random person your address, and to come pick up something when you're not home).
This is exactly what happened in the Marketplace story. It's not like anyone was hacked or anything. The setting in Muse said always, and the agent took it at its word. It's pretty important to remember that AI and LLMs struggle to think in nuance, and thrive more on objective or logical tasks and ways of thinking.
Some questions I had, and what I found from articles
Can you export your data or delete it? You should definitely check this specifically out in your own settings before you connect anything. I couldn't find much on this.
Does Meta use your data to train its models? Yes, by default it does. This is however, a setting you can turn off. Meta does say the training data is always anonymized first.
Does Meta use your data for ads? Meta says Muse conversations don't go to its ads systems. That's a policy, so keep an eye on it.
My rules, as a builder
1. Don't put client data into a consumer agent. Ever. EVER!
2. Keep a human in the loop. A person approves each output before anything is done.
3. Know where your data goes, how it's stored and who it's shared with, and know which privacy rules apply to you. You definitely don't want a lawsuit, and you definitely don't want a fine.
4. When in doubt, leave it out. If you catch yourself second-guessing whether something should go in, it shouldn't.
What we do at Stack & Lever, is build our own tools, including our own AI operating system, the AIOS. It wouldn't be fair to say we don't use a third party, because we use Claude Code, and coding agents a ton, but it runs in our IDE and our terminal, so we see every step, and we approve what goes out. Client and lead data stays out of consumer apps.
Where do you stop?
Muse is a good personal productivity tool. It isn't a tool for business data, especially if you've signed anything that governs how you handle people's information. I'm not saying it's bad. Use it. Just don't trust it with everything.
There is one thing that worries me about our whole industry, and I'm definitely, to some degree, in the same boat here. AI makes things so easy that your brain kind of switches off, and you forget the boring and hard stuff that matters. The second you slip up, something will happen, even if it doesn't happen right away.
So here's my question for you. Where do you stop trusting AI with your data? And if you're building something right now, do you actually know where it's stored?
