I tried Muse with a job I’d already built an agent to do

Like a lot of people, I'm uncomfortable with Meta's approach to data, and giving it another way into my life took some thinking. I mean it's not just Meta, it's big tech. Alas, as I build AI products and my own agents, I want to understand where these tools are going, including the ones I might not otherwise choose to use. Basically do my job.

As I already use Instagram, I'm already giving Meta some of my data so I figured a small, real use case to test Muse. Even better is it's something I personally would use. Before all these agents started appearing from the big tech companies, I'd used AI to help me build my own book automation with scripts. I could work through books with my agent, give it criteria for what I liked, and have it add books to Hardcover, the app I use to keep track my "want to read" list.

There are tons of book reels on Instagram. Currently, if I see a book I want to save, I take a screenshot, send it to my agent in Telegram, and have the agent pick out the title and author and send the details over to Hardcover. While useful, it's still moving information from one app to another so the automation could begin.

The question I had was when using Muse, can I start from where I find the book (Instagram) and do less of the moving things around myself?

I signed up for Muse and explained in natural language how I want to save books and send them to Hardcover. Muse described to me that I start with posts I save on Instagram. Better yet, make a collection called "Books" so it wouldn't have to go through everything I'd saved. I also told it what type of books I like and asked it to search for books based on that criteria.

This flow is easy to understand. Saving a post is something I already know how to do. A Books collection gives those saves a purpose. I didn't have to begin by designing another little system for getting a screenshot into the right place. We were up and running in about five minutes.

Connecting Hardcover was where I started paying closer attention to the product experience. Muse explained that it had its own computer and could browse the web, and that Hardcover had an API it could use. An API is a way for one piece of software to work with another. In this case, I needed an API key from Hardcover so Muse could connect to my account.

It gave me a connectbutton in the chat and a place to enter that key. Muse said this kept the key out of what the agent could see. What I could see was that it gave me somewhere specific to put the key. If you don't build this stuff, how would you know whether to paste it into the conversation, put it into a terminal, or do something else? Even those words can make a fairly simple task feel like you've wandered into something you aren't qualified to touch. Here, the product was guiding me through the connection as part of the conversation I was already having. It did a good job of abstracting away the complexity.

I noticed the same attention to getting someone started in the onboarding. Muse introduces itself and asks you to name it. There's a cute avatar, and you can see it doing fun things like typing on a keyboard while it works. It walks you through what you can do, and you can ask questions as you go.

As a product person, I can see the thinking there. Naming it gets you involved immediately. Watching it work gives you something to follow. There's an effort to make this feel approachable and give you a reason to come back, without needing to learn what an agent is or how one works. I can't tell from my own use whether that will retain users but I could see what the experience was trying to do.

I did run into some bugs. I used Muse on my phone, and several times I had to back out and come back in. My first attempt to connect the API key failed. I kept trying, but it's unclear if the patience and tenacity is a result of the fact that I build these things . Someone trying an agent for the first time might assume they'd done something wrong, especially at the point where they've just handed over a key to another account. They might not want to try again. The connection flow can explain the task really well and still lose someone if they don't understand what happened when it fails.

I had a smoother experience with Grok Bot. What I did with Grok Bot are different tasks, so this isn't a controlled comparison, but Grok Bot felt less buggy to me. Both experiences made me think about how much of the work around agents can be made easier for someone who doesn't want to build their own setup.

That's what interested me about trying Muse with books. I already knew the task was something I could automate. Muse got a book into Hardcover with a flow that began where I was already finding books, and helped me make the connection along the way. There's a lot to like there. These are just going to get more user friendly as time marches on. Next up is OpenAI's Dots.