Muse gets AI agent UX right

1. 0:00 – 4:41

   What Muse is and who it’s actually built for

   1. SPSpeaker0:00

      It's the weekend, the kids are out of the house, and that means one thing. It's time to test a new AI agent. Today, I'm gonna give you my first pass opinion about Muse, Meta's new personal agent. We're gonna talk about what Muse is, what I had it do for me, and what it didn't do quite well, and why I think it might be my favorite designed agent I've played with in a long time. Let's get to it. Optimizely Agent Platform for marketers is built on a bigger idea, a whole directory of agents now joined by a newly revealed squad of virtual teammates, each with a defined role and a personality of their own. This isn't about faster drafts, it's about handling the coordination, too. The research, the approvals, the handoffs, the routing, the back and forth that quietly eats your week and never makes it into anyone's highlight reel. Get the story behind it all. Hear the confessions from marketers who have handed off the busy work. Meet the squad and understand what virtual teammates can do for you at optimizely.com/howiaiai. First, before I get into what I loved about Muse, let's talk about what Muse is. So Muse is from Meta. Yes, Instagram, Facebook Meta, that Meta. And they're positioning it as your personal AI agent that gets things done. Don't we all want this? Now, let me tell you what I can for sure verify having looked at the landing page for Muse. This is for me. This is for millennials with kids, maybe Gen X with kids. This is for the olds. So if you are a young person, um, I'm sure you can find many use cases for Muse. It's actually pretty great, but if you look at what they're targeting in terms of use cases and ideas, it is 100% for parents. You see here they are going to highlight things like filling out permission slips, shopping for strollers, all those sorts of things. So this is gonna be a, um, AI that is really well targeted, and spoiler alert, I actually really like it, and I'm gonna go through why I really like it and what I would use it for and where it surprised me in a really delightful way. Now, I wanna call out a couple things that make Muse the same as other agents and a couple things that make Muse different than ano- other agents. First of all, why is it the same as other agents? Well, it is a agent that has a computer. Everything is computer. The first thing they put on this landing page is, "It has a dedicated, secure computer with its own browser." What I think is so funny is everything on this landing page for Muse is very consumer friendly except for this one paragraph that says, "Muse secure VMs run on a persistent, isolated Linux virtual machine equipped with full browser and enough storage, CPU, and memory to do the kinds of things you'd do on your own computer." That is clearly, like, not what my mom is gonna care about. This is clearly not what the, like, middle of the road consumer's gonna care about. I, as an early adopter who was probably the target audience for Muse, cares about this, but I don't think anybody else does. The other thing is Muse has desktop and mobile. You can also communicate with it through WhatsApp, so again, like, ease of communication using the Meta ecosystem, really useful. It really has good controls, and so this is something that I've observed through this entire onboarding and my initial experience with Muse is it's very good at asking for permission, telling you what it's going to do, and moving through that process nicely. Again, they're reinforcing security credentials. It can store credentials. It's calling out a future 1Password integration, and then, like all of these agents that people wanna have shop for you, Stripe's Link as a payment method and payment protections, the first for AI agents, is highlighted front and center. Finally, you know, just going to, like, the consumer friendliness of this, they don't call it plug-ins. They don't faul- call it connectors. They say apps. You can connect your apps, and you can see the first ones here, the ones listed here, are very consumer friendly. Facebook, Instagram, Gmail, Health, Shopify, all those sorts of things. So again, this is gonna be a really consumer oriented, not a, um, work focused agent, but that's kind of why I love it. And then, um, again, they put a lot of content together about why-- how they designed Muse and how they built safety into Muse. I will just shout out preemptively to the Meta team, it's designed really well. I actually really love this agent. I think it's delightful. So we're gonna go through my experience setting up this agent, a couple things that I had it do, and why I think you might use Muse as your personal

2. 4:41 – 7:16

   Signing in and the onboarding flow

   1. SPSpeaker4:41

      agent. Okay, what does Muse benefit from? Well, benefits from the fact that when I went to the Muse landing page, it had my Facebook account already signed in, and I just had to click one button to sign as a, in as Claire. The first screen they show you is just this few things to know, and again, this is kind of like the meta commentary on this product. They really want you to feel safe. They want you to feel confident, and they want to engender trust, so it tells you upfront what Muse can do for you and why it's great for you. It can take actions. It can work around the clock, but you are in control. Once you get started, you get to dive right in with your Muse bot. Now, there's a couple things I really love about Muse, but the first thing is it lets you pick an identity. I love an identity. I want my agents to be cute and named, so right off the bat, I was asked, "What do you wanna name your agent?" And of course, I named my agent Polly. Up in the top right, it used to be this little, like, yeti style animated fuzzy guy, and I will show you the process by which we updated the avatar. It's really cute. We'll give Polly a little makeover, but I named her Polly, and why did I name her Polly? Polly is my beloved original OpenClaw, recently replaced by Grok Bot, but to be replaced by Muse. I think my personal agent, Polly, is gonna be a Muse bot. Now, a couple things you're gonna notice when you've logged into Muse. This is OpenClaw-y. It's got OpenClaw tells to it. What does it have that I really love from OpenClaw? Well, over here in identity, it actually has an identity MD, a soul MD, and a memory. And you know what? I love you Grok Bot, but I want to craft your soul. And so the fact that I can click in here and open soul MD and edit it. Again, if you read the soul MD and you've ever looked at the OpenClaw kinda out of the box soul MD, this is very familiar. It looks a lot like OpenClaw. Um, but I do love the fact that you can actually manually edit the identity and memory and make these changes. Now, I've used this for a couple hours and it hasn't actually updated the identity or soul that much. So I'm curious how integrated this, like, updates are. Does it do it overnight like OpenClaw sometimes did? We'll see. Um, but it's kind of interesting.

3. 7:16 – 8:24

   The activity feed and its task lineage

   1. SPSpeaker7:16

      There's a couple other primitives of Muse that I think are fascinating that we're gonna go through. First is this activity feed. So it will actually trace for you every task that it's done for you over time, and I love this feature. If you are an agent designer, pay attention because this lineage and this history of what it's done for me over time is really nice. I can go back, I can understand what it's worked on. And so we're gonna go into the design of this particular feature a little bit later in the episode, but I really, really do like this. I mean, the other thing is, is it will list all the approvals that you've given it, so if you have to explicitly approve an action, it has that there. Again, like most tasks, it has its heartbeat. We got a little bit of OpenClaw. What's it doing in its heartbeat? Um, I don't know, but we're gonna find out. Um, so it has its heartbeat and then it has a couple tasks. Again, very user-friendly. We will go through the items on the left, which again are very interesting, but I wanna get into just the first chat I did with Polly.

4. 8:24 – 9:48

   First real task: managing the family calendar and deleting soccer practice

   1. SPSpeaker8:24

      So past naming Polly, I, you know, asked it for one thing. I said, "I need to stay on top of my kids' schedule. They have so much going on. My middle kid quit soccer. Like, help." And it went through this very, very nice onboarding experience for me. It asked me for a Google Calendar. I connected. It was super easy, not scary. Then it asked for my kids' names, which we won't show you. And then it said, you know, "Are you ready to delete soccer practice since your middle kid quit soccer?" And I said, "Yep, go for it." And it did it super easily. It asked for permission. It left the right things. That just first experience was so nice. It was really easy to go through, and it just set me up on a good foundation to feel like, yes, this agent could be really useful for me. Then it said, "Hey, do you want me to pull an agenda together?" And this is where we get into so- to testing some of the capabilities of Muse against these other agents. So I said, "Yes," and I asked, "Can you make it really easy to read?" And my hope was it was gonna make an artifact because apparently Muse can make artifacts, HTML, PDF, et cetera. So I was hoping, like, maybe I could incant it into making some, like, cool agenda. It did not do that. It gave me a bullet point list of things, but then I decided that I would ask it for the Trad Bot favorite

5. 9:48 – 14:29

   Requesting a morning newsletter PDF

   1. SPSpeaker9:48

      feature, which is the morning newsletter brief. This is the thing that I think is so cool. Every morning we print out a morning brief with a schedule for the kids, what the kids need to know, what we need to know, and then news for the day. And so I prompted it. I said, "What I really want is a PDF agenda for tomorrow morning, kind of like a newsletter one-page style that I can print out and put on the kitchen table. I want it to have not only our schedule for a week, but highlights of each kid's call-outs for the parents. You might need some information from my email about what's going on, and then recent news, recent events, things in San Francisco, the weather. Um, we can make this kind of a fun read for the whole family that we can talk about, not just the logistics, but also things that are going around in the world, in our city, in our lives." And I, y'all, I really like this. So it did have me connect my email address. Then it ingested some things that it knows about me and my family from my email, and again, this was a really nice experience. It found out my kids' names. It found out my husband's name. It found the kids' activities, like they do jujitsu and basketball and piano and all this kind of stuff. It inferred that we're civically engaged in San Francisco. It inferred that we work in tech, that we care about tech and AI, that we like to do personal projects. And then it says, like, you do a lot of, like, food delivery and shopping, which is correct. And I said, "Yep, that sounds like us. Please build this, um, this PDF newsletter." So it started to work on it, and it gave me this beautiful thing, and I just cannot tell you, for those that are not watching, we have been making these since OpenClaw came out. This is an idea that my husband had, and we have been printing these out, and this is the first one that has been one shot in a design I actually like. So this has made me want to test some of the Facebook models because I have hated the front end design of a lot of these PDFs out of Claude and out of Codex. They've just been, like, kind of boring, and this is so easy to read, so beautifully laid out, and, you know, it has a couple sections. It has a beautiful agenda for the week ahead. It has three call-outs for each of the kids. It actually does call out, I didn't have to ask it to, where, like, Br- Brazilian jujitsu slams up against piano class. If you're a parent, you really appreciate that. And then I really love this news to know section. Because it has news that's local, relevant, and kid appropriate, and encourages us to have these conversations at the kitchen table. And so I said, "Hey, I want news locally." It gave us news about safer e-bike rules, that there's a new telescope observation in the Tarantula Nebula, and that the Russia-Ukraine, um, conflict continues. And then it has these great questions. "Talk at the table. If you could volunteer for one school event, what job would you pick? What would you name a brand new star-forming nebula?" Just fun things that we can have conversations with our kid. And then for San Francisco, I love it, it calls out that the Autumn Moon Festival is this weekend, is this upcoming weekend, and, and that's happening around town. So a couple things I loved about this. The content, great. The design, beautiful. It's actually better than the one that Grok Bot or OpenClaw does. And again, this is, like, for family agents, for personal agents. I don't want agents that get me on my computer. I don't want agents where my face is in the phone. I want agents that get me with my family into my life in an offline way, and I think it just did this really lovely job. This episode is brought to you by OpenArt, where you can manage your entire creative workflow from idea to final output with top models for images, video, and music. I create a lot of assets with AI. The problem is figuring out prompts, models, and settings before you can make anything good. OpenArt's new chat mode changes that. Chat with it like you would with ChatGPT. Whether you're creating an ad, a short film, or a promo video, start with a rough idea and it can develop the concept and script with you. Chat mode chooses the right models and settings based on what you describe, or you can specify exactly what you want. You can create multiple images and videos and handle entire creative workflows through simple chat. Then keep iterating by simply typing what you want changed. Whether you know AI or not, you only need an idea. If you can say it, OpenArt can create it. Learn more at openart.ai. That's O-P-E-N-A-R-T.A-I.

6. 14:29 – 16:05

   The personalized news feed and how I set it up

   1. SPSpeaker14:30

      So since we're on kind of, like, daily newspaper, daily feed, I do wanna call out some other kind of components of the Muse experience. So in addition to the main chat, you do have this feed, and this feed is basically a personalized news and information feed for you. And all I had to do is actually write this prompt and click set. It says, "Make me a feed about my interests. Keep the tone clear and direct. Ensure it is quick to skim. Avoid clickbait." That's it. That was pre-populated. I click go. And now this, today it has already pushed me two news items. One, um, totally targeted at me, as somebody who is sleep training, poorly, her baby, that says, "Sleep quality is more important than screen time limits for school success." So sleep is the number one thing for success. This is really super targeted to me. And then the second thing, which again is targeted because, uh, later I will show you the goal I set up, is, "Salt in drinking water is linked to higher blood pressure." I mean, you should know this. This is pretty obvious. Um, I do like to put a sprinkle of salt in my water, um, because I have been electrolyte pilled. But this little feed is nice. It's much healthier, I will tell you, Muse engineers, and much healthier than my Facebook feed, which is truly nightmare fuel. So I do like the idea of building my own personal feed. I'm curious if this will pull in from Instagram or from Facebook. I'm just unsure. I haven't seen it yet, but that's really interesting. So feed is one really unique part of

7. 16:05 – 17:10

   The “Ideas” feature as an out-of-the-box prompt library

   1. SPSpeaker16:05

      Muse. The other really unique part of Muse is ideas. So instead of you having to come up with a blank slate or name an agent and give it a job, which is what I do with OpenClaw, it's what I do with Grok Bot, Muse just comes out of the box with /ideas, and ideas has great ideas for what it can help you do. Every, you know, it said that I'm civically, uh, engaged, so it's like if there's a pothole or busted streetlight, I can try to get it fixed. It has obviously the productivity things like managing your inbox, managing returns, but also has awesome things like relationships, where it can host a trivia night for you or set up a dinner with a friend that you haven't spent time with in a long time. I think this is gonna be a killer app for health and fitness. I think we're gonna see a lot of wellness Muses. That's probably what I'm gonna use it for, but I think that is really amazing. Personal financial management and shopping, as well of a bunch of other ideas. And if we look at the design here, so beautiful, so cute, so delightful, out of the box. Just really love how they're presenting these ideas that you can then kick off.

8. 17:10 – 21:40

   Setting up personal goals (water, shoes, and sleep training)

   1. SPSpeaker17:10

      The other primitive that I love is goals. So it will track your goals. And again, it's not a to-do list. It's not like this, like, developery task list. It is like, what are your personal goals? And they help you create these goals from health, relationships, finance, career, interests, productivity, something else. Um, so it's tracking two very important goals that I set up prior to recording this. One is drinking enough water every day. One is I'm buying shoes and I want them on sale. But I wanna show you what it's like to set up a goal and how nice of an experience it is to do this, and why I think goals is gonna be one of these, like, great unlocks for personal use cases and something genius that I think the Muse team did. So we're gonna make a something else goal, and we're gonna say, let's do this. This is danger, danger zone. And then we're gonna go all the way down and we're gonna say, I wanna start a goal about sleep training my eight and a half month old. Gently, we are room sharing because Our house is small, but mom has to get some sleep, please. Okay, so let's see what it does here. With my water goal, it was very kind to me. It gave me one step that I needed to do. Let's see what it does when I have a goal to get my beloved darling child to sleep a little bit through the night. Not even through the night, just sleep a little bit. So it's thinking, and then what it's gonna do is it's gonna propose a plan to me and then say, "I'm gonna track that goal for you, and I'm gonna do reminders for you of this goal." And it, it just is a really nice experience. And my reflection on Muse, overall using it, is that this is just a really well-designed personal agent. It's not that it's the most intelligent, it's not that it has the most powerful, um, it is just so well designed. So my goal is gentle sleep training. The baby is eight and a half months old, and we room share in a small house. Claire is running on very little sleep. It's true, and I want a gentle plan. And it's saying, "Sleep depri- deprivation with a baby is its own special kind of hard. We'll get you some rest." Like, my heart grows 100 times. This is what I need to hear from a personal agent. It's not that sycophantic like a GPT-4.0 voice, but it is gentle, kind, human first. I really do like the voice of Muse. I find myself 0% annoyed with Muse, which is a miracle. It speaks to me like a human. Um, so it's gonna say, "What's a typical first step right now?" I'm gonna say, "He goes to bed between 7:30 and 8:00, and usually wakes up between 12:30 and 1:30. Recently, he's been wanting to party for an hour before he goes to bed. If I can get him back in his crib, he wants to be patted on the back or hold my hand. It's very sweet. I like doing it, but mom needs to sleep. Then he sleeps till, like, 4:30 or 5:00. Again, needs the same treatment. Um, and then ideally we can get him back to sleep till 7:00 or beyond. Sometimes the 4:30 or 5:00, my husband comes and gets him and carries him till he sleeps. Help." Okay. Again, I, you know, this is a goal, and I have done this goal with OpenClaw. I have done this goal with ChatGPT. I have just liked the experience of setting personal goals with Muse a lot better. The tone is a lot gentler. The path towards the goal is nice. The fact that it's actually tracking it as a goal as opposed to a chat is really helpful. And, um, it's telling me the two things that are going on, the 1:00 AM party, and try trimming down 30 minutes off the second nap, great, and the patting and hand-holding are sleep associations. So what does this nap look like? I'll do this. We'll skip this in the episode. You guys don't have to listen to my full baby schedule, but let's end up with what the final goal and recommendation is. Okay. So once Muse has your goal set up, it s- sets up a reminder to nudge you every morning on your goal, um, or n- every night on your goal. And so I really like this. It's a combination of this new primitive, which is a personal goal, with the reminders that we've n- come to know and love from, like, Grok Bot or OpenClaw or any of these agents, and it's just a really nice way to track personal things. And then the last

9. 21:40 – 23:11

   Library: documents, websites, images, videos, and podcasts

   1. SPSpeaker21:40

      piece I wanna go to, like, kind of unique navigation of Muse is Library. And so Library are all the things that Muse has made for you, basically. And what's really cool is the first time I came into the Library, of course, I was expecting this artifact of, um, the PDF that I made. It can make you documents, like PDFs. It can make you websites, which I haven't tested yet. But given the front end design of the artifact, I really like. But what's really cool is it can make you images, videos, and podcasts. Podcasts. I did not know it could generate podcasts for you, sort of like NotebookLM style podcasts. And so what I prompted this podcast to do is just make me a podcast about the AI news this weekend. And it, it didn't take very long to do this. Um, and it generated this podcast, and just listen to what we got.

   2. SPSpeaker22:32

      Welcome to The Weekend Wire. I'm Alex.

   3. SPSpeaker22:34

      And I'm Jordan. It's Sunday, September 13th, 2026, and honestly, this might be the most dramatic weekend AI has had all year.

   4. SPSpeaker22:43

      It's really a tale of two moods. On one side, the biggest labs in the world are suddenly talking about hitting the brakes. On the other, the president is saying those fears are wildly exaggerated.

   5. SPSpeaker22:54

      I was pretty impressed. It's a six-minute podcast. You can actually publish it to Spotify or Apple Podcasts if you want. I'm presuming you can share it with people. This was a real surprise to me in Muse that I was not expecting that I really liked. And so

10. 23:11 – 23:56

    Quick recap and what I love

    1. SPSpeaker23:11

       here's my experience with Muse. There are just these moments of delight that are surprising and nice, these design affordances that are really cool that really make it feel like the personal agent you want. You know, there's been a lot of talk about Instinct, um, especially on the timeline and, like, I had my own kind of bad data experience with Instinct. Maybe I'll try it again. I probably won't. It was nice to be able to iMessage an assistant, but I'm telling you, this is what consumers want. They want beautiful things. They want trustworthy, like, progressive disclosure. They want, um, permissions to be asked for. They want these, like, nice affordances in UI, and this just feels more comprehensive, more complete, more consumer-friendly

11. 23:56 – 25:18

    Activity feed design deep dive: tool calls and step-by-step lineage

    1. SPSpeaker23:56

       to me. I wanna talk about a couple things that I really love in the design. So yes, of course, I said I loved identity. Second, I really love this activity feed, but if we go back to creating that podcast, let me just find it. What's really cool about the activity feed is it's not just the list of what it did for me. It is actually a list of every tool call, script, et cetera. Now, most consumers are not gonna care about this, but it's really kind of nice as a developer to see the step-by-step that it went through on every task. And as I was watching this progress I was thinking, "Man, I really want this in Codex. I want this in Claude Code. I want this in GroqBot." I love this lineage of the tasks that it's done for me, how it's thought, but it's in just, like, such a nice UI. It's so delightful. It's really cool. And so I think just the design thinking is very, very strong here, and not just the fact that it will track everything it's done for you, which shows the work, gives transparency. It's not just that it exposes the tasks that it's done and how it's-- and the tools that it's used and how it's done that work for you, but it's presented in a really easy to understand way, where depending on the elevation you're comfortable with,

12. 25:18 – 26:12

    How Muse handles permissions

    1. SPSpeaker25:18

       you can really dive in. You know, the second thing is permission granting. It does a really nice job of asking for permissions, especially at the appropriate level for consumers. So you don't see down here, like, auto-approve. You don't see, like, YOLO mode. You don't see ask me every time. It just asks you the right time for permission. So when I connected my email, I think a GroqBot would've said, "I have your email. I'm gonna go read it." Muse said, "I have your email. May I go find out some information about you?" And I said, "Yes." And then it said, "This is the information I found out about you. Can I use this information, please?" And I said, "Yes." And so it was just this, like, really lovely way to give trust to this agent that didn't presume that I wanted to go YOLO mode, but also was not annoying, like, just was not annoying. It, it really wasn't. And so I really liked that experience.

13. 26:12 – 29:34

    The animated avatar: Polly becomes Slime, the teal dragon

    1. SPSpeaker26:12

       And then I gotta tell you all, my favorite part of the UI is the avatar. Now, look, it animates. It's so cute. Um, I wanna show you a couple things. One, when I ask it to do something, it gets a little laptop, and it starts working. So if I say, "Hey, my parents are coming into town the first week of October. Can you find some activities in San Francisco that they might wanna do with a kid? It's, like, the first week in October to the second week in Oc- October." And let's see if-- Ugh, look, Polly's got a little fuzzy laptop that she's working on, and also these, like, activity feeds are so cute. Perfect emoji, perfect task name, just, like, great design here. I really do love it. Um, and so I think this is extremely cute. Extremely cute. The other thing is editing. Editing the avatar is adorable. So if I want to change the avatar, it gives me a prompt here that says, "Change your avatar to," and I'm gonna say instead of a red lobster, lo- Polly, I love you. You know, you are no, are no longer an OpenClaw. "Change your avatar to a teal, adorable dragon. I want your name to be Slime because that's the name of my kid's favorite stuffy." And what's really fun about this, okay, again, Polly's going to work on her laptop, but it's gonna generate an image here for the new avatar. And look again at how adorable this is. Whenever it's doing image gen or media gen, it pulls up, like, an orb that it's working on. It's so cute. I love it so much. Again, when I talk to designers about operating at the top level of your craft, I do not mean rounding buttons. I do not mean making strokes. I do not mean running impeccable. What I mean is this. I mean you take a static avatar, and you use image gen to animate it for an on-demand loading screen that you could have never done before. Now, look, it's giving me options for who Slime could be. Oh my God, he looks like this. It's so cute. Um, so I'm gonna pick this guy, and I select it. Again, question and answer tool, selection approval tool, so delightful. And Slime is now working. This is his new look, but what's great about this new look is it is not just fixed. This is not just going to be a fixed image, although it's adorable. You will see if I say, um, "Awesome, Slime. Can you look up when The Odyssey is playing on IMAX in Daly City this week? Preferably during the day when I have a nanny." Um, ultimate AI use case, finding IMAX tickets to The Odyssey. You will see Slime should be animating soon. So look, he's moving. Again, this is not something that I told it to do. This is not something that I asked for. It just presumed and did it out of the box. I

14. 29:34 – 31:15

    Browser use test: shopping for New Balance 9060s (not great)

    1. SPSpeaker29:34

       love it. Now, this is the last thing, is I have had Muse try to shop for me. I had it sh- try to shop for New Balance 9060s because I am a basic B. Um, I'm a basic millennial mom. I need my basic millennial mom shoes. My experience shopping was not great because it did not find the right colorway, the voice mode, like, it, it didn't get the right number right. It wasn't really great. I kinda got stuck on browser use. But what I do like about Muse is instead of exposing its full vir- virtual machine to an end user who probably really doesn't care about that, it exposes a subset of the virtual machine, which is the browser, which is the thing I care about. I think consumers would be very confused and scared to see this, like, weird looking computer, um, especially with the errors that you would see with, like, a GroqBot or something like that. It just, like, doesn't give confidence. But being able to see, um, the browser and just use the browser actually is an, a more accessible Kind of like way to interact with the Muse Bot's virtual machine. So I've actually really liked that experience. Now, it's browsing. We'll see if it does a good job. Again, it's looking at Practical Magic 2, which I do wanna see, but not what I asked for. This is where I found Muse do less of a good job. So you all know this, I am a Codex browser use, screen use girl. Codex would crush this in a hot minute. For some reason, Muse just really struggles with finding the right thing on the internet to buy, which stinks because I do want it to buy things for me. Now,

15. 31:15 – 33:34

    Browser use test 2: buying IMAX tickets for The Odyssey (much better)

    1. SPSpeaker31:15

       what's weird about this is it says it's playing at IMAX this weekend, Monday, Wednesday, blah, blah, blah. Do I want it to get me tickets for one of these things? Now, what's concerning, and again does not engender trust, and this has been the only part, the browser use has been the only part where I did not get trust, is like it didn't seem to pull up the right movie. So I don't know where it's getting that information. But let's say I wanna go, let's say I wanna go Monday at 12:15. I do wanna see what it does here, because what I suspect it does, it's gonna like start to check out, um, with the tickets. I'm gonna say one 'cause I'm getting one ticket 'cause I, we can't afford a babysitter, so we just have to go to movies by ourself, especially a three-hour movie. I can't have a $300 babysitter/movie date night. Too much. Okay. So on it, it's grabbing one ticket. Let's see what it does. It's gonna come back with the total, and then I bet it's gonna ask me for a Link invite. I'm super curious how this works. Let's go to the next tab. Um, and loading the browser task. Okay, it's finding The Odyssey. It's clicking through. It's gonna buy the ticket. Yep, it's gonna definitely go through and just use browser use for this, and then I bet it asks for Stripe Link. I can't actually go to this movie, so we're going to stop, um, and, and not actually buy this ticket, but this was much more successful than my shoe s- shoe shopping experience. And I think with the integration with Link, we can have a really nice, easy, you know, AI shopping experience with Muse that I honestly haven't tried much in Grok Bot, and then never did with OpenClaw. But this feels like the right agent for the job, which is this personal agent. Okay, $20.99. I'm gonna have to buy... It's going through guest checkout, which I suppose is fine. I don't need a Cinemark, uh, account. And then I suppose I could give it Stripe and pay for this movie. All right, it did it. Now it's asking for a link by Stripe. I've-- If I connect it, I could issue it a card. It could buy me these tickets, and I would be on my way. Unfortunately, um, I actually don't have three hours to go see this movie. I will never see a movie until my children are in college. Um, but what a nice experience. I don't have to browse. I don't have to do any of this, and it did a much better job than shopping for

16. 33:34 – 37:09

    TL;DR and what I’ll actually use Muse for going forward

    1. SPSpeaker33:34

       shoes. Okay, so that is my whirlwind tour of Muse. TLDR, I love it. I like it. It's great. The onboarding experience is fantastic. We've got these top-tier consumer designers from Facebook and Instagram and Meta in general. You can tell they put great kind of like user experience into this. I was delighted at every moment. I was never annoyed. The, um, authorization permission method, you know, traceability, the soul, all of it was what I wanted from personal agent. Now I'm gonna toss it on my phone and probably carry it around and use it, I think, for personal use cases. So probably not for my chief of staff stuff. I-- my life is just too complicated for that. But I will definitely use it for wellness tracking, kid activity stuff. I'll probably move my morning newspaper to it. It's actually, the design is just too good. Shopping, for sure, and then maybe sort of like home relative use cases, right? Like chasing the city to fix something in a house or chasing the city to fix something in the street, signing up for a workout class. Those kinds of things I think I will use Muse for. What I think people should really pay attention to, though, is product managers, designers, we have not had to design agents for most of our career. This is a really new thing to try to design, and I do think you should spend a lot of time with Muse figuring out what they got right, because I think they got a lot right on the design here. How much they expose, how much they don't, how much they make transparent, how much they don't, um, and the new primitives of kind of like tasks, ideas, goals, library, the items inside a library. We're not using words like crons. We're not using words like to-dos. We're not using words like tools or artifacts. This is a very consumer-friendly app that I think I'm gonna use. One thing I don't think did particularly well is browser use. Had some failures with shopping for shoes. I don't know if the vision's quite there. But it did a pretty good job with finding tickets for The Odyssey, so a girl can't complain. That is my whirlwind tour of Muse. I like it. I, I think it's adorable. I'm gonna install it on my phone. And again, for those that are worried about meta privacy, honestly, I've been a Meta, Facebook, whatever user for 20-ish years now. They've got my photos. They've got my mom. They've got like pictures of my kids on Instagram. They've got everything about me, so I guess the devil you know. Um, I also am gonna try the Facebook models because I was impressed by the artifacts that were generated, and I think that's good indication that there might be really strong models on the Meta side, which I, I haven't played with. So we'll probably do a, uh, review of models there. In the meantime, I would love for you to test Muse. I think it's really interesting. I'm curious what use cases that work well for you, what you think of the design, and what you would like to see me do next with Slime the Muse Bot. This has been a How I AI AI agent review. It was so nice to walk through this with you all, and I can't wait to talk to you soon. [upbeat music] Thanks so much for watching. If you enjoyed the show, please like and subscribe here on YouTube. Or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com. See you next time.

Episode duration: 37:11