What I Built With Muse, Meta's New AI Agent

Last week, Meta launched Muse, their new personal AI agent. I don’t have a strong opinion on it yet, but I have been using it a bit.

I’m not an AI expert, nor am I an expert on Muse. But I’m also not completely ignorant of AI, and I’m constantly looking for new tools to try out to see what I can build.

In this post, I will share the details behind what this tool is and some specific examples of how I might use it. At the end, I’ll share exactly what I built with Muse, and you can check it out for yourself.

What is Muse?

Muse lives at muse.ai, or you can install the iOS, Android, and WhatsApp apps. While I did add the iOS app, I find it much more useful on desktop.

You can name your agent. I named mine Frank.

As an agent, it’s similar to a chatbot but can perform specific tasks for you. You can tell it to research something, build something, or watch something, and it goes and does it. Frank runs on its own virtual machine, so it keeps working after I close the app. It also remembers context between conversations.

What I Might Use it For

I put Muse in a similar category as Claude Code and Codex, but there are surely some differences.

I asked Muse for a list of use cases, and I won’t share that full list with you. But you can build landing pages, have it run weekly tasks, and surely help with ad copy and creative. Feel free to ask Muse yourself about how you should use it.

But, when considering ways that I can use it, there are a couple of areas that make the most sense:

1. Building tools like calculators, diagnostics, and decision trees. These are things that otherwise require a developer.

2. Watching Meta’s docs. Keeping up with Meta’s changes is never easy. I’ve searched for ways to automate this, and I’m hoping Muse is my solution. I’m having Frank check the Help Center and developer pages every Monday morning for changes and additions.

There’s plenty more that you can do with it, but this is where I’ve started.

What I Built

Once I knew that Frank could build a tool for me, I asked it to research my content from the current year and come up with some recommendations. It came up with three ideas, and I tested out some initial versions. But two of them felt like they were trying to accomplish too much with simple questions, and I abandoned them.

What I settled on: Value Rules Decision Tool.

I went through many iterations of this one, too. Its foundation is entirely my content. In addition to my blog posts, video posts, and podcast scripts, I gave it my Master Brief to build a knowledge base. I discuss value rules a lot, so it wasn’t difficult to give it the info it needed.

Frank and I fought about it. There was logic that didn’t work the way I wanted and steps that felt unnecessary. I didn’t just give it a task and take the finished product. It required a ton of my own detailed thoughts on how this could work.

I had the option of embedding an iframe or adding HTML to a page and updating custom CSS. I went with the more custom approach.

I’m not saying it’s a finished product, but It mostly does what I want. I created it in one evening and wrote zero code.

It’s live here: Value Rules Decision Tool.

The thought process behind this tool is that I firmly believe that most advertisers overuse restrictions by age, gender, and placement. So it forces you to walk through the restrictions you use and the conditions behind them. By the end, you’ll get a final verdict on whether you should use value rules to manage that restriction.

I’m still early with Muse. I’ll share what’s actually useful as I learn it.