← Back to search

Best AI Personal Assistants (Claude Duet Won Me Over)

AI Rabbit Holes · 2026-11-14 · 53 min
relevance 59 10621 words Episode page ↗ Audio ↗
Show full episode description
Get My Free AI Marketing Guide ⁠⁠⁠⁠ ⁠⁠⁠⁠ Join My AI Community ⁠⁠⁠ In this video I interview Aaron Makelky about AI personal assistants like Claude Duet, Meta Muse, Grok Bot, and ChatGPT Dots, and why Claude Duet became my daily driver as someone who already lives in Claude Code. We cover which AI personal assistant you should actually use, plus the real jobs I hand off to Claude Duet. 🔗 Video References: - ⁠ Best Personal AI Agent Article - Claude Duet - Meta Muse - Grok Bot - ChatGPT Dots - Hermes Agent ⏰ Time Stamps:
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Which cloud-based AI personal assistant (Claude Duet, Grok Bot, Meta Muse, ChatGPT Dots) is worth running your work through, and how to use Claude Duet day to day.
Benefits
  • Cloud assistants work from your phone while you sleep, unlike local Claude Code
  • Proactive 'dreamer agent' nudges via Apple Watch after quiet hours
  • Claude Duet barely counts against Max-plan rate limits
  • Meta Muse free tier generous; $16/mo web plan beats $200 Grok Bot usage
  • Keep work outside the apps so you can switch when one dies
Use cases
  • Automated podcast prep and X/LinkedIn posts generated from transcripts via Claude Duet
  • Digital PR outreach run through Claude Duet scheduled tasks
  • Aaron used Grok Bot as main assistant for a month before rate-limit cuts drove him off
  • Meta Muse checking inboxes and coding automations; only 9% usage halfway through the week
KPIs / results
  • Grok Bot rate limits cut to 25-30% of prior levels around November 3
  • Claude usage split: Claude Code 60-70%, Duet 30-40%
  • Pricing: Duet needs $20 Claude plan; Dots needs $100 Pro; Grok Bot $30; Meta Muse free or $16 web
  • Anthropic reportedly lost over $100 billion this year ahead of IPO
Tools / build
0:00 / 0:00
📑 Chapters — tap a time to jump there
00:00
Introduction
  • Cold open: Hermes is for in-the-weeds tinkerers; cloud is the big differentiator
  • AI Rabbit Holes with Aaron Makelky; topic is AI personal assistants
  • Aaron moved from Grok Bot to Claude Duet and Meta Muse
02:55
Is Anthropic Nerfing Opus 5.5?
  • Claude Fable 5.2 and Mythos 5.2 released; critics dub Opus 5.5 'Slopis'
  • Aaron got three Fable resets on Max plan, good until March
  • Aaron uses only Opus 5.5 now, dropped Haiku and Sonnet
05:35
The AI Personal Assistant Battle
06:29
How AI Personal Assistants Differ From Claude Code
  • Assistants run in the cloud; Claude Code runs locally on device
  • Duet is its own phone app; can connect remotely to a local machine
07:23
Proactive AI: It Pings You First
  • 'Dreamer agent' scans email, calendar, notes to nudge you first
  • Duet pings Apple Watch after quiet time with overnight task updates
  • Plain Claude needs cron jobs or webhooks to act proactively
08:59
The Big 4: Claude Duet, Grok Bot, Meta Muse & Dots
  • Claude Duet, Grok Bot (trendy on X), Meta Muse (ad-pushed), ChatGPT Dots
  • Duet needs $20 plan; Dots $100 Pro; Grok Bot $30; Meta Muse free to start
  • Ryan pairs Duet with always-on local Claude Code
10:27
Rate Limits and What Happened to Grok Bot
  • Dots currently uncounted usage behind $100 minimum
  • Duet barely dents Max limits; Claude Code is 60-70% of usage
  • Grok Bot limits cut to 25-30% after Elon tweeted they lose money on it
  • Meta Muse $16 web plan: 9% used halfway through heavy week
13:12
Which AI Personal Assistant Should You Use?
16:02
Where ChatGPT Dots Falls Short
  • Where ChatGPT Dots falls short versus Duet
19:00
Other AI Agents: OpenClaw, Manus, Instinct & Hermes
  • OpenClaw, Manus, Instinct and Hermes as alternatives
  • Hermes suits tinkerers; Aaron no longer wants to touch it
21:50
Why Keep a Backup AI Assistant?
  • Keep work outside apps so you can switch when one dies
22:59
How I Use Claude Duet as My AI Personal Assistant
  • Ryan's daily Claude Duet workflow on the $200 plan
26:14
Scheduled Tasks With Claude Duet
  • Scheduled tasks in Claude Duet with completion notifications
27:06
Automated Podcast Prep
  • Automated podcast prep through Claude Duet
29:10
X & LinkedIn Posts From My Transcripts
  • X and LinkedIn posts generated from podcast transcripts
30:40
Digital PR Outreach With Claude Duet
  • Digital PR outreach handled by Claude Duet
33:44
Why Use Duet Instead of Claude Code?
  • Duet vs Claude Code: cloud, mobile, proactive scheduling
38:30
Where Claude Duet Falls Short
  • Claude Duet limitations, including no model selection
40:34
Meta Muse Demo: Spaces (Personal Mini-Apps)
  • Meta Muse Spaces demo: personal mini-apps
50:41
Final Thoughts
  • Final thoughts on choosing an AI personal assistant
Hermes is for those in the weeds tinkerers like Aaron Makelky's. Even I don't even want to touch Hermes anymore. It's this cliche AI assistant that works while you sleep type situation, but I think the cloud is really the biggest differentiator and advantage here of these assistants. I love Grokbot as a product, but the problem is Elon tweeted, we're losing money on it and you could tell they nerfed the models it could use. They cut the rate limits to about 25 to 30% of what they were before. And if you're an active Claude Code or Claude AI user like I am, highly recommend giving Duet a shot. If you're not on OpenAI's ecosystem, like if you're not going to get into Claude Duet, I would actually take Aaron's advice here and probably go with Metamuse as kind of that secondary option. Welcome back to another edition of AI Rabbit Holes. This is a podcast where we go down the endless amount of rabbit holes in artificial intelligence. And today I'm happy to have back on once again, AI consultant, AI expert, Aaron Makelky. Now for those tuning in live on X, appreciate your viewership. Those tuning in through the podcast on YouTube, Spotify, Apple, appreciate your viewership as well. Be sure to subscribe to all those platforms. You don't miss out on one of the favorite rabbit holes that we're going to go down today, which is AI personal assistance. So before we get into our topic today, Aaron, which I know many people have requested this from us, what is new in your world when it comes to just everything going on in this crazy environment of artificial intelligence? This is the perfect topic because this is exactly what I've been doing. And a lot of us have been moving from threads and projects and terminals up to something like a personal assistant app that can control those things. And for me, Grokbot was my main source of that and my main app that I was using for more than a month. But I don't know if they changed the models. It got dumber. And then they changed the rate limits. So it got really hard to use. And I was using Claude Duet because I got access to it a week and a half ago. And the last four or five days, I've been using Metamuse and just seeing what it can do with more serious work and the more powerful model and finding ways to keep all of my work outside of those apps. So I can like plug in anything when the next one dies, I can switch to a different one. And I know you've been kind of going through the same evolution with your own use cases. Yeah, I'm excited to dive into Claude Duet today. So that's I've been like hammering Claude Duet lately, and it's pretty cracked now with what you can do. And I'm excited to share some real world use cases later on. One thing that's top of mind for me, though, Aaron, which is kind of related, kind of not. I saw Anthropic just released Claude Fable 5.2 and then Mythos 5.2. Now, I haven't gone in and explored this yet, but I've seen some commentary from people that I trust in this space. One guy in particular called it Slopis 5.5. So like in my head, Anthropic got the nerf gun out and already implemented this on Opus 5.5 like we saw with Opus 5. Because Opus 5 at a certain point was almost unusable for many use cases. I haven't seen the decline. I mean, maybe minorly I've seen a decline, but nothing drastically with Opus 5.5 like I saw with Opus 5. But have you tested Fable 5.2, Aaron? And are you seeing these kind of nerf or declines with Opus 5.5 yet? I don't see a reason to. And here's the problem with what they're doing. Putting Fable on its own rate limits and then giving people resets. That sounded like an awesome thing. Did you get some resets for being on a max plan? Yeah, I did. I got three resets. Yeah. Yeah. And they're good until March. So first of all, kudos to them because they don't historically do that. They're trying to do something else. I appreciate it. But I go, Opus is so good and so fast and so efficient. Why do I care about my little Fable resets and my Fable-only bar? In fact, I quit using Haiku and Sonnet for anything. I just use Opus 5.5. So to me, I don't see anything from the benchmarks and examples I've seen and people I've talked to, including some who work at Anthropic. I'm like, what good is this? What would I actually use it for that's worth its own rate limits and switching things? And I haven't found any. They're like, oh, refactoring a code base. And I was like, yeah, most people can't relate to that. Like, that's not on my to-do list today. And we'll see. I'll eventually test Fable 5.2. Maybe it is worth those extra usage limits. And, you know, eventually this is kind of like on your iPhone when the software conveniently gets worse and worse over time, where they get you to upgrade to a new phone. I think we're seeing a very similar thing play out in this AI model thing where slowly but surely, once the new models come out, they're just going to keep nerfing them and making them worse and worse. And like, oh, look at Fable 5.2. And they're going to drive more subscriptions, more people use the API. Like, I don't know. I think that's just all part of the play. Yeah. What's interesting is if they're already ahead of OpenAI and Google and Meta, why are they releasing new stuff when they don't have to? Because I think 99% of people agree. Opus 5.5 is so good at almost everything. Revenue-wise, we don't know, though. Right? I mean, it's good for us and our use cases. We don't know what's going on behind the scenes in terms of, like, revenue and how it's doing for them internally on a business perspective. Like, they've lost, I think, over $100 billion this year, Anthropic has. And so it'll be very interesting to see how this plays out with their IPO coming soon. I don't know exactly when it's going to happen. But, yeah, there's a lot of that stuff going on behind the scenes, too. But let's actually transition into our topic today. I'm going to get the slides going here of the AI personal assistant battle. So before we get into this, I'm going to add some context for the audience listening. So I have used ChatGPT dots, and I'm using Claude Duet a lot now. That's where I'm coming from. Aaron, on the other hand, has used or has used Metamuse, Hermes, and then also GrokBot, I would say, and probably some others, like Instinct maybe, Aaron. And correct me if I'm wrong here, but you've used those AI personal assistants where I've kind of used the open AI and Anthropic version. So it's going to be interesting to kind of hear our perspectives differ or clash as we go through this. Yeah, and usually you're the more Claude-centric person, but you had been using GPT dot more than me, even though I've been using the Codex desktop app since like February. We've had different experiences which we can talk about. But, yeah, I've used Duet. I've had it for like a week and a half. Yeah. And just for like quick clarifying, for those who don't fully understand, like what is the difference between ChatGPT dots and ChatGPT, Claude Duet versus Claude or Claude Code, right? And so I think the biggest takeaway here, Aaron, and feel free to hop in at any point, is they run in the cloud, right? And that is the big differentiator here. And so when I have tasks, like I'm literally on my phone. Duet's its own separate app, by the way. So when I'm on my phone on Claude Duet, everything that I'm doing is through the cloud versus Claude Code. It is local on my device. And that's a big difference there. When you're on the go and you're trying to get things done, you can't really do things locally on your phone. Yes, you can connect remotely to a local device, which is one of my favorite features that Claude Duet recently added. And I'll talk about that later. But it's this cliche AI assistant that works while you sleep type situation. But I think the cloud is really the biggest differentiator and advantage here of these assistants. And I would add one thing. That's exactly right. The other thing that a lot of them have been doing is proactivity. How can we get the AI app to engage the user when they wake up in the morning drinking their coffee or before they go to bed? Before, you had to go to a chat box and say like, hey, do this for me. And now all of them in their system prompt, and I've read their system prompts. They use what's called like a dreamer agent. And they basically say, look through all of Ryan's emails, calendar, meeting notes, whatever context, and find things that sound important that we can nudge him on instead of waiting for him to open the app. So I'm sure you have this. But do you have quiet time set on your Duet app? I don't know if you can see it, but I have it connected to my Apple Watch and I get notifications from Duet whenever it's done working on like my scheduled tasks now. So that literally just went off right as you were talking. That's beautiful. I mean, here's what mine said. When I got up this morning, two minutes after my quiet time ends, it says, hi, good morning, Aaron. Updates on your current tasks. It tells me two things it did while I was sleeping. I didn't ask it to. And it says, here's a task I could help with. Want me to start on this third one. Okay, Claude would never do that on its own. You would have to have a cron job or a web hook or like set up your own system for it to proactively work. One of the things they're trying to do is like, how can we get you sucked in and ding your Apple Watch without waiting for you to open Claude or whatever app? They all do it the same. Yeah, it's crazy, man. And before I actually get into this, let's talk about the big four here. Okay, so I talked about how I use Claude Duet now for everything. GrokBot, I would say, is still kind of the we're on X, right? So it makes sense that it's the trendy one because they're incentivized and the algorithm to make it the trendy one. And Meta Muse is a very popular one. That's on mainstream advertising with Meta's ad network. Obviously, that makes sense. Chat GPT Dots, I would say they're not super popular right now. They were big, obviously, when it was released. But in my opinion, for what I'm doing, Claude Duet has really improved. When it first came out, it was, I don't want to say it was unusable, but it's not something I would be like, oh, let me drop what I'm doing and start working on Claude Duet. Where now, it's actually the opposite. I'm doing a lot of stuff on Claude Duet paired with my local version of Claude Code running at all times. And so, yeah, so my daily driver is Claude Duet. I'm actually on the $200 a month plan. But I think what this is saying here is that you have to be at least on that $20 a month plan to use Claude Duet. And then Chat GPT Dots, I would say this is a disadvantage. You have to be on that $100 a month pro plan. Maybe they've changed that since then, Aaron. But GrokBot looks like it's being the $30 a month. And then Metamuse, there's no actual cost or subscription required to start using it. And so I can leave these slides in the description for those who want more details on this. Of course, you can just ask AI for more details on this. But Aaron, what are your thoughts on kind of the big four here? Yeah. Let's talk about usage. If I'm going to pay for these, how much I get. Yeah. Dot is weird right now because it doesn't count. So you basically have this $100 a month minimum to access it. But you can use it like crazy for now. They haven't done anything where it's like taking your tokens or you have certain limits. So you kind of have this free period. With Claude Duet, Claude's so efficient at everything. It's not counting against your rate limits. When I look at my Claude stuff, it's like Claude code is about 60% to 70% of my usage. And then Duet's the other 30% to 40%. It's really hard to tell because you can't change the model. But if you're on a max plan, unless you're going crazy already, Duet's pretty usable. Like you're not going to get hit with rate limits. The two that are the opposite are GrokBot. And GrokBot, I'm on the highest paid plan. So people who have opinions, I'm sorry. You can use it on the free, but it's so limited. You can't do anything. I'm on the $200 a month plan. I did not unify my SpaceX, Grok, all those cursor plans into one. I had a pre-existing one. They drastically cut them back about a week ago. So let's say like November 3rd, somewhere around there. And they said it was using Opus 5.5. So like it was smart. It had great rate limits. The app, the voice calling, I love GrokBot as a product. But the problem is, Elon tweeted, we're losing money on it. And you could tell they nerfed the models it could use. They cut the rate limits to about 25% to 30% of what they were before. There's no way you could run your business or do all of your work through GrokBot on the highest paid plan. I know because I have been. What's really interesting with Metamuse is the free plan is actually pretty generous. Like you can hit limits if you're doing real work. But for $16 a month, don't subscribe in the Apple App Store. Go in the web on Chrome and you'll save $4 a month. Yeah. I didn't even need to. Like I had a ton of usage, but you do get a higher model. You get the full wonder model instead of Spark. But for $16 a month, based on how this week has been going for four and a half days now, I've been going crazy with it. I have automations. It was checking inboxes. I had to code stuff. I was at like 9% halfway through my week. So the usage of it is insanely high. It's very generous on the free. But if you pay $16 a month, you get more Metamuse usage for $16 a month than I get for $200 a month on GrokBot. And it's not even close. That's crazy. That's crazy. Well, Metamuse, you also have to understand that I talked about this on a previous episode, the ad network. Right? Why do you think there's a generous free plan? Think about all the data they're collecting and feeding to their ad network for Instagram and Facebook and whatever other properties that they own. But I'm going to talk about for those like watching or listening, they're probably asking right now, Ryan or Aaron, what should I use? Right? That's the question that probably a lot of people have. I'm using Claude Duet and that's the only one I'm using mostly because I'm using Claude Code as my daily driver for almost everything right now. Now, I do use Codex. So I was using ChatGPT dots before Duet was released and polished around the edges, but I'm no longer using dots. And I would say the only advantage as of this video, because things always change so quickly by the time this is going to be released. I would say as of this video, one of the main use cases of dots, I would say the Slack, I would say the Slack integration works extremely well. And I would also say voice mode. They definitely have an advantage on the voice mode side versus Anthropic and Claude Duet. That they have an advantage, I would say, almost everywhere else over ChatGPT dots. Opus 5.5 is a much better model than whatever the 6.1 version of GPT that you're using. You can use Claude Code on your phone through remote sessions. And I think they just added this recently. I saw ChatGPT dots. Now they're copying Anthropic with this thing. Shocker, right? There's a wait list. So, I mean, if you're on the wait list, I think you have to be on the max plan to maybe get off that wait list. I've seen a few comments about that. But I'm going to dive in later, Aaron. I don't think now is the appropriate time to get into all my Claude Duet setups. But what about, like, your opinion here? What should someone use, you know, if they're just kind of listening to this and thinking, like, well, I've dabbled in ChatGPT. I've dabbled in Claude. I've seen GrokBot everywhere. I see Metamuse everywhere. Like, where should someone just get started here? Okay. For now, I would say not GrokBot. Because their models aren't there. And I'm sorry. I have no confidence that Grok48 or whatever's next is going to be anywhere close to Opus 5.5 or Sol 6.1. If they change the rate limits, maybe. But, like, for now, I would just cross it off the list and go, okay, here's what Metamuse is good for. If you're on Instagram, Facebook, WhatsApp, or if you have the glasses, I have them. Beta access only right now. You can get Muse through your Meta glasses. That's still pretty niche. The good thing is it's free. So, like, if you just want to try it, it has amazing mobile apps. It has a really good desktop app that's on Mac only. And the models are pretty good. It's really fast. It doesn't have a ton of connectors. But, like, if you live in Meta's ecosystem, surprise, surprise, you should use it. Because it's got a really low barrier to entry. It's simple. The issue that I have with GPT. It's super slow and it's unreliable. So I tell people all the time, on paper, it's powerful. It can do way more than do wet. But when I call the thing and it takes 45 seconds to answer me, I'm not doing that. And I was using it last week. And it just ended the call. It hung up on me like it failed. I've never had duet do that. Now, duet's voice mode has a long ways to go. But guess what? It always works. I've never had duet fail at something. And this is what it looks like behind the scenes quickly is I call it Bodhi. It was my Bodhi bot where I still have an automation running, which I'll probably delete that. But you'll see here. You can call it, text it. You know, you can start Slack messaging, add it into different Slack automations and whatnot. But I don't use this anymore. And another disadvantage here is that you'll see there's just one ongoing chat. There's no way to organize this into channels. I think that's what duet calls it or groups or just a way to make your life and brain more organized versus if I pull up Claude duet here. Notice on the left-hand side how much more organized this is through different client work, personal brand work, digital product work, just personal life. I'm having it automate my bookkeeping. I'm having it do investing things for me now behind the scenes, right? So this is just such a better and like the scheduled task. Like this whole interface, just everything about duet is so much better than chat GPT dots right now across the board. I would say except for voice mode. That's really the only thing that OpenAI has going for it right now. And they have a lot of work to do if they want to keep up in this AI personal assistant race. I just can't believe that they acqui-hired the OpenClaw guy. Yeah, that's the other thing I don't understand is OpenClaw is still running. And I was going to talk about that here next on these other popular personal assistants. But like whatever happened to that? Why didn't they just take it and maybe like reintegrate or rebrand the exact same technology into OpenAI? Like I just don't understand it. Yeah, they were slow to get their agent thing out. And then when it came out, completely broken for me. Still, it's been over a month now. It's the most unreliable. Like GrokBot works. It's just dumb and doesn't have great rate limits. But it doesn't break for me. It doesn't not answer the phone or hang up on me or fail. I just can't. I can't see how they have an unpolished, unreliable product. And they either need to take it seriously and fix it or just kill it. And they've had this problem before where they try to do too much. Yeah, exactly. The OpenAI agent builder is my favorite example or the Sora app. Remember the Sora app? We can go down the list here. They just have way too much shit going on. Yeah. So Claude is living up to their reputation of like they were the last one to release it. And it wasn't very good at first. But yeah, in the week and a half, you and I have had it. The voice mode went from broken to usable. It can now see things locally on your machine where it couldn't do that before. So it's gotten better fast and Dot has not. Yeah. And some of these other ones, like for people wondering, they probably heard OpenClaw. They probably heard of Hermes. Manus is a big one. I think here's another question, Aaron. I think Meta still owns Manus. Do they not? Or did they get rid of it? They spun it back out on its own. Okay. And it like restarted just last month. Because that would be like, why would they come out with Muse and they own Manus? Like, so, okay. Thanks for clarifying that. But should anyone mess around? I know you're a Hermes fan. I see you in that merch that you get all the time. Should anyone mess around with any of these other personal assistants over the big four? There's no reason to use OpenClaw unless you want to spend your whole life fixing it. I'm sorry. I agree. It's not better. I killed mine in May. It's I love the idea. It's a beautiful thing, but it's not worth using. It was a pioneer. I don't know if a lot of this stuff would be here if OpenClaw didn't come out. Yeah. And I mean this with like all respect to the people of the idea and this Peter Steinberger guy. Like he's a visionary. Amazing just to have the idea that inspired all these other. But OpenClaw itself is a mess. Yeah. Manus I haven't used since it spawned back out from Meta, but I also don't see a reason to. Like it's a one-off thing and you're probably already on an open AI or an Anthropic subscription. So I don't see the need. Instinct I killed just because it scared me privacy-wise. It was connecting to stuff that it never asked for. It was just shady. The reason why I like Hermes is it's the only one on here besides OpenClaw that you can change the model. You can host it where you want. You can copy and paste it to a Raspberry Pi or put it in the cloud. I want control over things, even if it's a backup system. So like my GrokBot once a week backs itself up to GitHub and my Hermes has a bot interface that just mirrors it. So if I killed my GrokBot subscription, I could basically pick up where I left off in Hermes and then I could use DeepSeq, GLM, OpenAI models, Grok models, whatever I want. So to me, it's an insurance policy. If you're an open source purist and you just believe in what they're building and want to give them their $20 a month subscription, you can host it on their computer. I don't do that. I have enough subscriptions to run myself. But that's where I use it. And there's lots of new iMessage ones like Instinct. I think that's consumer-based. It's like for busy moms. It's not for people building websites. I don't want to be an iMessage to do my work. Yeah, I would not waste your time with Instinct, OpenClaw, Manus. I would argue most people should not touch Hermes either. Hermes is for those in-the-weeds tinkerers like Aaron McCulkey's. Even I don't even want to touch Hermes anymore. I did before, you know, the Duet and the Dots and all those came out. But like now I don't see the need to touch Hermes for my use cases. Okay, I agree. But I'm going to add one thing and take all the shots you want on me because that made me laugh. That was fair. I earned that one. When the subsidy era ends, revisit this. I'm going to say it again. It's my insurance policy because what GrokBot did to all of us last week pushed me to Hermes and was like, hey, I already have this ready. It's already set up. If I didn't have Duet and Metamuse, I could have just gone all in on Hermes and used whatever models I wanted. So for now, you're right. It's like buying a video card to host your own stuff locally. Do the math, people. It doesn't make sense today. In the future, maybe. Yeah, we'll see. Because a lot of this too, like we could keep talking about this because a lot of this, people will connect their subscriptions, right? They'll connect the open AI and the anthropic subscriptions to Hermes. Well, what's the point of that if the subsidies run out, right? Because now we're going to APIs. Of course, you can hook up OpenRouter and use open source models. Like, again, it's the tinkers. It's the more customization. But I agree with you on the insurance policy. I think it's something to keep in your back pocket if we get rug pulled sooner than we think. When we get rug pulled. Yeah, when. When. That's a good question. When. Well, cool. Let's actually get into real world use cases now, Aaron. So I'm going to talk about Claude Duet. Okay. So how am I using Claude Duet? What are some things that I'm doing? Okay. So these are just a few examples. And I'm going to share some of these right here on the screen of what's going on. So anything from various SEO and content tasks, daily competitor scrapers. Here's the cool one that I just did the other day, Aaron. Podcast guests and invites. Managing my inbox and calendar. Managing my bookkeeping through the QuickBooks MCP. Um, it's actually automating my content creation on X and LinkedIn for various things. Digital PR outreach. This one's a little more advanced, and I'm going to briefly talk about this here in a little bit. So if I pull up Claude Duet, which by the way, you have to be on a paid plan in order to access this. The first thing I want to talk about are the connectors. So you can also use skills and plugins. This isn't any different from traditional Claude, but this is way more important now when we have Duet in the mix. And so for connectors, you'll see for my connectors here, if I scroll down, I don't know what just happened. Um, I wanted to get on here. It is. Here's my connectors. So I have about 17 connectors. I think you had like 50 or something crazy, Aaron, when I saw you screenshot it up one day, but like, here's all the connectors, right? I have Apify, Alertmouse, Blotato, Calendly, Cloudflare, Data for SEO, GitHub, Gmail, all the Google drives kit for my email marketing, Riverside for the podcast stuff, open router. And I could do hyperframe Slack. Like there's a million more others that I could do here, but it's important that I mentioned these connections first, because what I'm about to show you would not be possible without these connections. And if you, if they don't have an official integration, let's talk about that quickly. If they don't have an official integration inside Anthropics library, well, you can add a custom connector. So you can name it whatever you want. And then you can add that MCP server URL, which if you don't know how to find this, just go to Google, type in your software MCP server. And if they have one, it'll be there. If they don't, then it's not going to work typically. Um, so that's what I've done for some of these connectors here that have, you'll see the, the custom where it says custom right here. You have to go through that custom connection because it's not in the official, uh, Anthropic library yet. Um, any, any questions about connectors or comments about connectors, Aaron, before I dive into these use cases. It's probably the biggest thing people aren't using that they should be. And the way I think of it is if I'm paying for a Claude max, uh, I don't want to switch to Gmail on a tab and then switch to Calendly in a tab. Like I want to do my work through Claude or whatever platform. So search for apps you use in their connectors. And if they don't have them, learn how to make your own. Yeah, exactly. So let's actually see what this looks like in practice. Okay. And so let's see here. One use case. If I actually go to, well, you'll see, I have on the left-hand side, how I'm organizing this. I think these are called channels and groups. So I have client folders or client channels, personal brand folder, digital products folder, personal life folder. And on the right-hand side, you'll see here, I have various tasks. So active up next. So this is where it takes initiative, where it's finding things that it could be doing and then scheduled tasks. Think about this. People need to realize gone are the days of me ever thinking about N8N Zapier or make for my use cases. I'm not saying that's the case for enterprises or larger organizations that still rely on N8N automations at scale. But for me, why would I ever touch those ever again? Now that I have those connectors and the integrations, that's the beauty of this. And so every day it's doing a duet log where it's, you know, keeping logs every day of what it did, what it learned through a GitHub repository. Morning brief. That's an easy one, right? MedX report check. So every single week, this is, this is a cool one. Every single week, it sends an automated report to my client and then it updates that report using an artifact. And it has all those connectors in the background. I can't, I don't know if I'm able to pull it up right here in this example. Oh, here's a cool one. AI rabbit holes podcast prep. So if I go to my calendar, here's one thing that it helps me with Aaron is podcast preparation, including ours. And so what it can do is basically it looks at my calendar because remember it's integrated to my calendar and you'll see, I have two podcast episodes today with Sam Rexford and Nick Leroy. Okay. And so what it did is it saw that on my calendar, it's connected to Riverside. It went ahead and created a Riverside podcast invite, and then it uses a podcast research skill in the background to do research on this guest because it has their email. It has all the other stuff about them through, you know, fire crawl through Appify through web fetch and all those scrapers. And it creates these podcast questions that I instructed it to do in that skill. Did the exact same thing for my other guests today with the Riverside link and making sure these topic titles and questions are good to go. And this runs once a, once a week. So it scans my entire week to see what podcasts I have. And then it takes initiative to create those guest invite links, add the guest email, go through and actually create all these questions. So I'm ready to go because manually this actually used to take a decent amount of time going into Riverside, creating the thing, coming up with the questions. Like it's an actual manual process. If you want to have successful podcast episodes, we'll just put it that way. And then once you get it going, you just let it run. That's the beauty of this. Every Monday at 1 13 PM, it's going to run every week now. I don't even have to think about it. All I have to do is just say, Hey man, you want to join my podcast? Here's the Calendly link to book. And that's it. Um, Medx report check. I talked about this. Um, where is it? Oh, this is kind of a cool one, Aaron, uh, Tuesday content run before. So I call it my token max or automation. So every single Tuesday, I think at 10 PM is when my usage resets. Well, I have an automation that goes out every Tuesday at almost 5 PM to make sure I get the most out of that usage, doing various things behind the scenes. So I don't leave 50% untapped usage is what that's for. Um, this is cool, right? X drafts run. I need to rename this, but basically what this does, and we laughed about this before the show, this automates my X posts and my LinkedIn posts. Now I'm not going overboard. I'm not auto replying to those stupid AI generated comments or anything like that. But what this is doing is it's pulling from my Riverside connector, my Calendly connector. It has the transcripts of the meetings. It has the transcripts of the podcasts that Aaron and I are doing in Riverside. And it's also pulling from my Google drive folder of the weekly school calls that I do. So it's pulling from those transcripts and it's creating posts in my style. I gave it a skill on how to do this, um, using those transcripts. So what did we talk about last week, Aaron, your website stack can cost 25 bucks a month. Claude Astro, GitHub cloud flare still on WordPress or lovable move one site. This was all automated. I did not even touch this. And you know, 13, 1400 views for just being hands off is pretty damn good at scale. I would say, at least for an account like mine. Yeah. And it's, that's the power of connectors because it's not making up something. That's almost verbatim what we said in the transcript of the podcast. Yeah. And we have an episode coming out soon on how to avoid AI slop. This is a very good way to do this because it's pulling from my actual transcripts. This was all automated. You know, all, almost all of these are automated, except when I come in here and like add videos, images, or like quote people. But, um, that's another good use case. Um, here's another use case I'm going to get into this digital PR. What's called reporter request scan is the automation, but the use case is this digital PR outreach. So you could do this for any client, you know, anyone else here, but I'm going to just going to use Ryan Dozer, my personal brand site as an example. Okay. So from an SEO perspective, the goal here is to always get highly quality, high quality, highly relevant, authoritative, do follow backlinks. That is the pain point of every SEO. So it's the hardest part about SEO is to get featured and mentioned from other authoritative and highly relevant quality sites. And my DR, which stands for domain rating has actually gone up quite a bit lately, but basically I'm on H refs here and I'm just going to look at my backlink profile quickly and filter by do follow domains since those are more valuable. But basically what I'm trying to explain here is that I've now built an outreach system that isn't spam. And how I've done this is if I go to Claude Duet, I've set up this reporter request scan. So in my email, I have my email, Ryan, Ryan Dozer.com doing all of this behind the scenes. And so how this works is it says weekdays. So every day what it's going to do, and I might change it every day might be overkill. I might do every other day, but what it's doing is it's scraping all of this content on Ryan Dozer.com. I'm like, I put together all this research articles about the dead internet theory and AI layoffs and you know, the lovable, the AI website builder, idiot tax and poison Google serves like all these cool little research articles along with all the content that I've created and all the tools that I've created, like this AI agent readiness checker, all of these, in my opinion, of course, I'm biased would be good stories for a lot of journalists in the AI and marketing space. But they're not going to find this, Aaron. And we need to actually take initiative and outreach to these journalists. Okay. And so what it did with this automation is you'll see, again, it's every day I might change this, but what it did is it's connected to my email and it took initiative to create a Google sheet because it's remember it's connected to my drive. So it created a Google sheet called digital PR journalist tracker. Now I've hid the names and the emails in here because I don't want to show that publicly on this podcast, but Claude Duet did all of this. It found the names, the emails of the outlets, found the articles, the article URLs, personal hook to mention, because remember, I need to provide value and make my pitches intriguing or else it's going to go into this endless abyss, abyss of inbox of thousands of AI generated emails that these reporters get constantly. So you have to be very careful in how you're crafting these pitches with your subject lines, ensuring you're providing value to the reporters. I did the same thing for podcast pitching. So podcast pitching is a whole other beast here, um, general pitch ideas. And then once this is all done, I'm going to have stats. So then it's going to keep track of who did I pitch to? What was the pitch? Was it successful? So now it's keeping track of all of this. So it knows I'm not double dipping or I'm not bombarding a reporter. If I followed up or reached out five times, it created this whole system on a Google sheet. And while I'm going through this, I'm going to go into my artifacts here. Um, Aaron, any other thoughts on this while I try to find some of these things that it created, um, behind the scenes, I wanted to show some of these artifacts. Yeah. I think the missing part is, okay, that's really useful. And you could have done that with Claude code or Claude coworker before for somebody who's just in the Claude desktop app or web app. Why would they use Duet to build something like this? What's the advantage of that interface or standalone app that you couldn't get if you were just in Claude code? You could, but now this is all automated. That's the beauty of this. And you don't have to go, you don't have to go through your folders and threads in Claude code to find where's that thing with that cron job or that trigger. It's to me, it's like on your phone or with your voice, you can work across all of the stuff that you, you can do the same thing. It's just a different way to interface with it. It's like a higher level approach to your work. Yeah. And so speaking of podcasts, well, here's something that is nice for you in the audience to know. One thing I've actually learned doing this, um, is that having a, um, podcast pitch sheet actually helps you quite a bit. And so I told it to put together a pitch sheet to basically send to these podcasts where it says who I am, you know, here's my data, who I am. What am I good at? Here's my numbers. You know, here's who I've been featured on. Here's some suggested podcast topics that we could talk about, or then you put this actually in the pitch. So it's just like a one pager for these podcast shows to review. Um, and then speaking of this, our automated report, I know I'm all over the place, but this medics report check, I think this is really cool. And so every week this goes out, remember it's pulling from, I didn't even talk about this, but it's connected to the Google search council MCP. It's connected to being webmaster through API. This is data for SEO pulling in this AI search data. This is Apify pulling in the LinkedIn and Facebook data. Um, and this is, I think it's WordPress MCP pulling stuff from his WordPress site. But like, I know this is a little in the weeds. It's on brand. It looks nice, but like, this is just like, I'm just scratching the surface here. Like this is the most recent one. I think I'm just scratching the surface here of what you can do, you know, with Claude Duet through just reports and outreach and literally everything here. I think this is the actual report I was trying to show. Yeah. October 31, November 6th. And so every week now, because it's automated, Aaron, it updates this report and then it sends it out every week to the respective emails that I wanted to send out to and CCs me to ensure that it went out correctly. Yeah. And I think the vision with like a duet is, could you do all this stuff by talking to your phone while you're jogging on a treadmill or out for a walk, pushing the stroller or driving around buying groceries? And it's like real Claude code on your local device or cloud environment, whatever you want. Yeah. And now you just have the one app and the one main feed to do all of your work. That's what they're trying to build. One, cause it'll burn more tokens. But two, it's like, it's way easier to keep track of stuff. How many times have we lost a Claude code thread that we were running in a terminal and you're like, where's that session? How do I bring that back up? Well, here's another one, Aaron. Okay. So I can't tell you, this is like a, anyone could relate to this, especially in the business world. I can't tell you how annoying it is as a business owner trying to keep track of receipts and save them. Well, now I have a receipt saver automation that scans my email. Every time I get an invoice, every time I get a receipt, it automatically dumps it and labels it correctly with the name and the date in a specific Google drive folder. So now I never have to think about collecting these receipts ever again. I just have to make sure that it all processes through my email. Yeah. And I got to give Mercury a shout out. Cause yeah, Mercury's good too. Yeah. If you want an AI native way to do your books and your invoicing and your banking as someone who's been stuck in Wells Fargo since I was in fifth grade, I just started moving some of my business stuff into Mercury and it's, it's like going through a portal into the future. So, and they have connectors you could use do that or dot or anything with it. And here's the ultimate sauce here. Okay. I haven't had the time to fully set this up yet. Um, I did and I deleted it cause it wasn't, it wasn't great, but this remote sessions, right? And so what I'm basically going to do once I get a better system here, I have a three month old child and I'm like trying to run around, get reorganized and this and that, right? You know how it goes, Aaron, but what I'm going to do is I'm going to eventually set up this bad boy, this beautiful Mac mini, uh, as a remote session for Claude Duet to always call on where I can start using the Claude desktop app with Claude code built into it and keep always have these sessions on at all times. Now, my wife's probably going to hate me cause I'm going to be on my phone probably more checking these sessions to make sure that things are running properly. Um, but that's the beauty of this. Now you can connect remotely to an always on 24 seven Mac mini running Claude code at all times. It's truly a beautiful thing. Yeah. I just hope they get their voice models. Cause to me, the thing that's lacking is you can't start a new local thread. So let's talk about some of the shortcomings of Duet, even though I've used it for a week and a half. It's great. I think the voice model is the biggest one. It's, it's not as powerful as smart as GPT voice is the best, but now that Meta Muse got voice, I can tell you Meta is ahead of Anthropic too, with their voice model. So they have a ways to go, um, with that. If you're not on the a hundred dollar or $200 a month plan, Duet's going to be very tough to use at scale. Yeah. And to me, it's best for people that have a bunch of different environments and Claude code things spinning around them. And you wish there was just one focused way to orchestrate all of that. Yeah. So if you're using Claude as like a chat, you know, you're in the chat interface a lot. I don't see an advantage to Duet or an advantage to subbing to a higher plan. Cause there's no way you're hitting your rate limits if you're just using chat. Well, and I mentioned the integrations. Remember I had GitHub as one of my integrations. We and you both love GitHub. We've talked about this a lot on this podcast. Um, but let's say here, here's a quick tip for you guys, right? So what I've done is I've have a whole repo with my Claude code backup, all the skills, all the context, all, you know, everything else that I have going on here. Well, guess what? I can seamlessly integrate that to Claude and have Duet learn from my Claude code backup. See what I'm doing. What's my chat history. What are the things that I'm constantly working on that it could take initiative and start to automate and create scheduled tasks. It's done this before. Um, same with any other repository. If I wanted to study any repo extract skills from any repo, it can do that. Like the beauty of having all these GitHub repos seamlessly integrate with AI is truly like a power move, bro. Like I can't tell you just like how blown my mind has been lately. Just the more I've been into Duet and the connectors and figuring out what the light bulb going off of why I can actually automate that with Opus five, five running at all times. I'm just like, holy shit, man. Yeah. Let's do, let's do some Metamuse because you haven't used it at all. Is that correct? Correct. I'm going to share. I have the desktop app running on my Mac and it's Mac only. So if you're Windows, sorry. If you're Linux, that's not a thing for you yet. Let's go. So here's my biggest sell on Metamuse besides that the rate limits are very generous, even on the free plan. They have a feature called spaces. So yes, you can go in a chat just like everything else. There are side chats just like Duet. But with the space, think about all the one app, one off dashboards and vibe coded apps and like little things that you just wish you could have. For me, I always have a backlog of video clips that I'm like, I can cut them out and edit them faster than I can review them and post them. So I told my Metamuse, I want this space. It's like a personal app. It works on the mobile version or on your desktop. And pull these clips and let me just approve them, cancel them or go in. And I use this app called Flask to like give creative edits and say, change this or add that and basically do a little dashboard for my content. I have one that has all of my business stuff connected and my goals. I don't want to show that one. But here's another one that's really interesting. Because it's personal. The first thing I started using Muse with about a month ago was fitness, nutrition. I weigh myself once a morning back to March. You can see I was like over 210 pounds at one point. Now, I'm not saying Muse helped me get healthy. Right. But it would ping me in the morning and say, did you weigh yourself? If so, what's your weight? In the evenings, it would check in on my nutrition. And then here's one of the best things it can do. And Claude can do this too. All the Apple native apps, calendar, mail, messages. Those ones aren't really attractive to me. But Apple Health. You can see my sleep. You can see my workouts. You can see my heart rate. And people, I know what they're going to say in the comments. Like you trust Meta to have that? Well, do I trust Apple? Apple. And Apple's had it since I had an Apple Watch. So I have it pull that and it basically makes this little dashboard and I can see my weight. I can see my goals. I can see whatever I want to put in there and I can even log it. Like this morning, this is what I weighed. And then I say, did I meet my goals with my nutrition or not? And I just do that once a day. It lives in my sidebar. So if you haven't used spaces, you can think of it like an artifact. It's not just HTML. Like you can have buttons. You can approve stuff. You can have it send something through a connector. It's great. It also has a library. So anything that you've made, like I have way more spaces than fit on my sidebar. They had a Halloween game. I was like, okay, I got to try this just to see what a space is good for. I had it prep me for this episode with Ryan. And it basically put a whiteboard up of all the things we talked about via text message and what I've tweeted about to find some good examples. I have a family dashboard of everything this week that my kids and my family have on the calendar. Who's picking who up? Who's dropping them off? And it's shareable so I can send that to my spouse. I have all sorts of little dashboards and web apps and things. So people who say that Metamuse is just like chat for normies, I had it by a domain, set up a new website, and then publish stuff there with a bunch of connectors last night all from my phone. And it took it about five minutes and one prompt to fix how it looked. And it was very usable. It's probably not. I mean, it's definitely not on Opus 5.5 level in terms of coding and design. But it does a lot of things better than the best GPT models right now. So my biggest takeaway is like don't think it's just for normies and you can't do real work because you have a ton of usage. If you want some other features, there's a feed. So you can say like go out on Instagram and find stuff I'm interested in. Here's news articles about AI and education. So you can build kind of any vibe-coded thing within the Muse app. Ideas is kind of the same idea. It's like this proactive stuff that it can do. And it's like, hey, want me to do this? I could do that. I could build you this. I think people don't appreciate what all is there. Yeah. And if you haven't used it on a free plan, it's awesome. The other thing that I wanted to show is they have some moats that other people don't. And the biggest one right now compared to Claude and OpenAI is hardware. So one, they open-sourced all of this. They have projects where you can build Muse onto a Raspberry Pi. If you're a nerd like me, I think I have two Raspberry Pis in the cabinet behind me that I'm not using. A lot of people have been into these little screen devices, whether it's like an e-ink display or the little watch face style, like 450 by 450 screens. You can build your own little Metamuse assistant and plug it in on your desk. So that's cool. I know hardware is getting very expensive. Then you got these expensive AI subscriptions on top of it. So if you really wanted to play this, how much is a Raspberry Pi or like what's a cheaper, I guess, decent local hardware that you could connect Metamuse to and run it 24-7 autonomously? Okay, here's the cheapest one. And I had a tweet kind of blow up last night. They are sending people in the U.S. a free Muse home link. And I know everybody's going to say, oh, now Meta is going to spy on your house. Yeah, first it's the Amazon Echo or whatever it's called. And now here's the reality. You're walking around with a cell phone. You know what I mean? Like they've been spying on you for years at this point. And well, here's what's cool about it. They shipped mine, so I don't have it yet. But it's free for U.S. people anyways. It's just a little USB-C device that has a board in it, like a little developer tool. Well, you can plug that in anywhere in your house, put it on your network. It doesn't have to be on your main home network if you don't want to. And say, turn off my lights every night at 10 p.m. Keep a log of who comes to my doorbell. Control my thermostat. Like it's a way to have an always-on device where your Muse is now in your house if you closed your laptop or you turned off your computer or you're not on your phone. Wow. The cheapest ones of these, I have a digital photo frame. It was $40. It's like an Android tablet that has crappy software. You can just plug it in and hack your own. The little ones with like a watch face are $40 to $50. The ones with no display are like less than $20. Raspberry Pi 5, they've all gotten more expensive because this is such a popular thing. I think I have a 4GB RAM version and I paid $80 for it. It's probably double the price now. It's like $150 for the lowest end. Yeah, I mean still, a Mac Mini is what, $600 or $700 now at this point? Yeah, I mean I ran an open claw on mine for nine months and I had another one running Hermes. What I think is cool is if you're a nerd, you probably have something lying in your house or your garage that you could run this stuff on. You can't do that with Claude. You can't do that with ChatGPT. You need an open source tool like Hermes. But Meta is kind of leaning into this. And the other one is they're the first one of these to have an Apple Watch app. So just yesterday, November 9th, your Metamuse, you can dictate to it on your Apple Watch and your little custom avatar shows up on your watch face. They're good at consumer products. And they're able to open source and do all these things and have cost efficiencies and all this because of their ad network. They make so much money. I don't even want, I'll look it up. We can do this after the show, but it's unfathomable how much money they make in ad revenue from Instagram and Facebook alone. Not to mention all the other shit that they have going on behind the scenes. Yeah. And they have had their glasses for a long time. But what a lot of people don't know is they have some new two different frames that are audio only. So I have Meta Ray-Bans with the little camera. I have barely used them for months. Their AI models sucked. The camera's weird. The glasses are big. It's a long story why I even have them. But what people have been saying is if I could get my Muse agent on my glasses and say, hey, Muse, you know, do this thing on my website, answer this email and not have the camera on the front. So they're smaller, but they're also less socially awkward. Like you're going to walk in a bathroom with Meta Ray-Bans on. That's just weird. Talking to people when they see a camera, they always get weirded out and asked. Yeah, it's weird. This is a hardware remote that no one else can touch for years. Like I know OpenAI is trying to come up with their own device. They don't have anything. They say they're going to put Muse as an option for your AI assistant on all of their older frames, plus these new ones that are audio only. It's kind of niche. I know most people don't have them. But if you have a pair that you haven't been using, it's beta access right now to switch it to your Muse agent from your glasses. I think that's amazing. And that's a huge moat that nobody else is going to be able to touch. So they're smart. They're leaning into their strengths. They're letting you burn billions of tokens for almost nothing. I'm very happy with it. I think it's super underrated. And they can take that burning of billions of tokens to fuel their ad network. That's where this whole thing is going. All right, Aaron. So with all of that being said, what are just some final thoughts that you have to conclude here about this AI personal assistant race and everything else that we discussed here today? I'd say two things. In the subsidized era, you should use the one that you're already in. So don't go get a Claude subscription because of Duet. Use Duet if you're using Claude code and already on a max plan. The other one is you should keep a backup for when the tokens are going to get more expensive or the models get nerfed. I don't see a better option than Hermes. So think of all of these as experiments right now. Everybody's moving towards an always-on assistant that's proactive. It's a single interface. It's got to have voice. It's got to have mobile apps. That's going to be how you run all of your coding sessions and stuff in the future. But you should use GitHub to back it up. You should keep track of it. Do some sort of a weekly export so you can build it somewhere else because they're also changing like crazy. Yeah. And if you're an active Claude code or Claude AI user like I am, highly recommend giving Duet a shot. If you're not and you're not on OpenAI's ecosystem, like if you're not going to get into Claude Duet, I would actually take Aaron's advice here and probably go with Metamuse as kind of that secondary option. And for the tinkers like Aaron, Hermes definitely still has a moat as your insurance plan, I would say. Yeah. And not just for privacy reasons. It's a control thing because the GrokBot people have felt this in last week. Somebody hits one button and you get a worse model and you're stuck. It makes the whole thing worthless. So have an insurance plan. And I think this Metamuse is probably the lowest barrier to entry. So if you're showing this to a spouse, a parent, a coworker who's not techie, good luck explaining Duet and Claude Code to someone who's not already in the Anthropic ecosystem. But Metamuse, you can be like, here's one chat. It has an app. Here's what you can connect it to. I think it's going to be the most broadly used. And like in the long run, think about how fast they've caught up the last two months. They were a joke. A joke. Yeah. A complete joke. And now they're like, some of their models are up there in the top 10. They're not the best, but I'm bullish on them in the future. For sure. Well, awesome. We'll end it there. And with all that being said, appreciate everyone who's made it this far for watching or listening to this edition of AI Rabbit Holes. Be sure to tune in again next time and also be sure to subscribe on whatever platform that you're listening to this and also reach out to us on if you have any topic ideas in this ever-changing world of artificial intelligence. We'd love to go down an endless amount of rabbit holes, hence why we're doing it and why this podcast is named that. So with all that being said, appreciate you all tuning in and be sure to tune again next time.