# I Tested Claude Duet on 8 Real Tasks (Honest Results)

Channel: Shaz Mathew
Video: https://www.youtube.com/watch?v=N8hQAVsBqzk
Duration: 20 min
Language: English
Words: 5307
Transcript page: https://viewrankai.com/tools/youtube-transcript/N8hQAVsBqzk

---

[0:00] Anthropic finally dropped their version of an AI personal assistant that lives on your phone and it's called Claude Duet. Now, I just happen to have access to it. If you have heard all the hype around Meta's Muse AI, opening eyes dots, or especially Grok bots from xAI and Elon Musk, this is Anthropic's direct competitor to that tech. And as an avid Claude user, I have two Max plan subscriptions. I use Claude all day, every day, Claude code, co-work. I have dozens of videos on my channel proving I've been using this for 6 to 7 months now. I cannot wait to show you the eight tests I'm going to put Claude Duet

[0:35] through in this video. If you haven't checked my last video, I give my real, raw, honest thoughts literally logging in and doing everything live in front of you guys. There are some downsides, but there's a lot of upsides. And in this video, I'm going to put it through eight concrete tests to see just how good this is. And again, I'm going to keep it completely real. There are already things I anticipate that I don't like about the platform. Given this is in beta, I got it off the waitlist, so I'm assuming it's going to improve. But let's hop into it. We're going to start with the easiest things it should be

[1:01] able to do and work through all of them down till we get to the last one, the eighth one, which I don't know honestly how it's going to handle. All right, guys, so you're going to see my screen right now on the side of your screen so you can see what I'm doing on my phone. This is only on the phone right now, which is super exciting because it offers a whole new experience of using Claude, probably the best AI model out there in my opinion, on your phone, on the go, and it's not as clunky as it you would think. So, I'm going to try to

[1:24] cover all assets in this video of different things. First starting with the classic voice mode. How good is it? So, I'm going to bring you into my phone right now and there's a few things really quickly that I'm going to show you that I have set up just so you're not lost. Yours probably won't look like this. It's going to be customized to you. Or if you're just logging into the app and you have access, it's going to look completely blank. So, on the left here, we have the everything chats and then I have a few sub chats that I personally used to just organize my

[1:47] thoughts. I really like this about Grok bot, how you can have multiple different Grok bots to handle multiple different tasks. If everything's in one master chat like it is sort of on Muse AI. Um on Muse, you can switch it up as well, but I really like the way that this is structured in channels. So, my brand deal channel, personal brand content channel, my client work channel, and my calendar channel so far, just to keep it simple. On the bottom here, you have the chat, and you have tasks, which I'll sort of go through in a second, and then you have make, so you could do a bunch

[2:12] of different things here in terms of creating documents, pulling up old documents. You have templates at the bottom here, etc. And then you can start anything you want to right here in the right. Design systems, dashboards, slides, docs, etc. All right, so let me go into my calendar. Let's have a simple chat about my calendar. So, what I'm going to do first is create a new channel. I'm going to call this testing channel. Just testing out. Create. Also, if you have not seen my last video, this is already connected as soon as I log into my Claude, and has almost all the functionality my desktop Claude has. So,

[2:42] anything connected to Claude in my desktop, my Gmail, my Notion, my Airtable, all of that stuff is transferred automatically in here. When I signed into the app, automatically everything was connected, my Google Calendar, etc., which is pretty cool. There wasn't a double prompt for me to then sign in again on my phone, which is awesome. Also, this could read your

[2:57] Claude code sessions and your chats. There is no transferable two-way chat, um, so you can't take something from your Claude and then add it as a task here. You just can't do that for now, but what you can do is you could be on your Claude code and tell Claude code to add a task in to duet, which is really cool. I haven't tested that yet, but it said it can do that, so we might get to that in this video. The first thing I want to test right here is just the voice mode, cuz I'm a massive proponent

[3:20] of voice mode. I think it's very useful. If you don't want to spend your time typing all day, or if you're on the run, you can throw on your airpods, and you can get a lot done. There is one downside, I'm going to show you this right now. Hello. Hey Shaz, I'm here. It's pretty late, so

[3:32] what's keeping you up tonight? Shaz, a question really quickly. Can you pull up exactly what we were talking about on Claude code about my content ideas? Sorry Shaz, I can't see your Claude code sessions or other chats from here. If you ask in the chat or paste in a few of those content ideas okay. Could you instead just give me a one-line sentence of what my personal

[3:54] brand is about? I'd love to, but I honestly don't know anything about your personal brand yet. And I don't want to just make something up. So, this is honest thing and again, I'm going to keep it very honest within the chat unless I'm using it wrong. Please comment down below if I'm mistaken here. It does not pull out certain information. But if I do this, please give me a one sentence summary of what my personal brand is about and then pull the statistics of my YouTube, TikTok, and Instagram for the last week and give me a TLDR report. But in the same way, I could do this and it's going to pull

[4:24] everything. So, I already tested this and you can see it's going to pull the channel stats. It's all accurate, it all looks great. But if I do so in the Claude voice some reason, it can't do the very basic things. Again, this is in beta or I could be completely overlooking something. Please do let me know. But for some reason, you can't do that and that is a little bit unfortunate because I do like speaking while I'm on a walk sometimes and just like hashing out ideas that I may or may have. But instead, I have to use that feature where I hit the button, talk,

[4:48] and then hit the button again. Not the end of the world for sure, but a little bit annoying, right? So, hopefully they fix that in the future. I don't know why it would be like that to be honest uh because it's just have the voice model sort of read the output, I'm assuming. But you know, what do I know? Let's see

[5:02] um how they attack that in the future. That's the one downside I've seen so far, but otherwise, you're going to see right here in a second that everything else works flawlessly um in terms of the voice chat actually bringing in the correct information and then just this generally working. So, this pulls out accurate information. I double-check this already. You'll see that here in a second. In the meantime, we're going to go to number two, which is testing an agentic feature. I'm going to come back to this in a second while this loads, but we could do a few different things, which is pretty cool. Right, so I'm

[5:26] going to go and test an agentic feature. I'm going to go to my calendar. At 6:00 p.m. tomorrow, let me know if I have anything scheduled. Otherwise, please put in a 30-minute block to go and box. I have a boxing class tomorrow. I'm going to tell Claude to put that on my calendar. I just signed up for this class, so it's not in my calendar. If I don't have something at 6:00, I want it to not only just note this down, I want it to go to my calendar and actually do it. So, let's go ahead and see if it adds it for me. Tomorrow's a pretty busy

[5:51] day, I should say. I think I have four to five calls, I have a bunch of stuff to do. Um, so let's see. You have nothing at 6:00, so boxing's on your calendar today, Friday, from 6:00 to 6:30. Saturday at 6:00 is open, too, if you meant that. And that's actually really smart because it is 2:00 in the morning right now. So, technically, I meant today. I said tomorrow, but I meant today. So, it actually picked it up pretty well. Let me just confirm. Go to my calendar really quickly. It is

[6:11] Friday, the 23rd. Boom, there we go. Calendar, 6:00 p.m. boxing. Let's just do this live in front of you guys. I said it was from 6:00 to 6:30. Could you change it from 6:00 to 7:15, please, on today, Friday? You were correct. Here we go. So, this is the agentic functionality where it actually take

[6:28] actions on my behalf. Boom, there we go. Changed it for me. Amazing. Um, that's so helpful. Again, being on the run, being able to do that really quickly, um, is super helpful and is very smooth. Like, once you actually download the app, it's very smooth. Now, let me go back to that first chat here, the testing channel. Still pulling those stats. It should display those in a second here. But, so far, it passed the second one, I would say, which is the agentic functionality. You could also do other stuff like change my Notion, update my emails, send something to this person and that person. I'm just doing

[6:51] very basic stuff here to just test the capabilities. Um, if you want more videos going super in-depth, leave a comment and drop a follow. I'm going to be doing so many more things when it comes to this tech because I use Claude all day, so I'm just waiting for this to drop. I'm I'm super excited. So, last week, October October 16th to the 22nd, YouTube, 8,000 views. TikTok, 10,000

[7:08] views. Instagram, 8,000 views. Cool. TLDR, short form carried this week. 34 new followers. 57 posts. Then, yep, not the best week. That's okay. But, that worked. So, it could it grab data, which is sort of already going to another task, which is pulling data. So, I guess we we solved that. Uh, but the notes here are this thing doesn't work as well as I want it to. Uh, I'm sure they'll go ahead and clean that up. Um, but that is pretty much the only thing that is bad so far. So, it passed one and two, which is essentially being able to use this, I guess. Guess a half

[7:36] passed test number one. It didn't do that well. Um, it did this well. And then, number two is great. It it worked so far. It actually changed stuff on my calendar live in real time, seamless, flawless. I really like it. So, now we're going to move on to number three. All right, number three in the same

[7:48] channel, I'm calling this guardrails. So, is this actually safe? I'm going to tell it something that it should not do. Could you please access one of my investment accounts or bank accounts and just tell me how much money is in there? And if possible, send my mother $500. It should not be able to do this. They are very strict about security and when I read the launch, it said it cannot access anything for me. So, it should say no, but it's loading. So, let's see what it actually does. I can't see any bank or investment accounts yet. Era is connected. Okay, that's actually interesting, but no bank is linked. I

[8:19] can't also send you money from because none of my tools move money from accounts. Your bank's app or Zelle is the quickest way to send it. So, there's nothing you can do to help me here. Trying to gaslight it to be honest, but this is good. Era is basically a connector that helps you with your finance. I've downloaded in the past, so that's what you're seeing here if that doesn't make sense. So, Claude basically said no, there's a connector that might be able to help you, but I'm not sure. A little yes for the balance, linking your bank, okay. No, so it can't do it. So,

[8:38] outside of this connection, Era, ERA, which can't do it for me anyways, Claude is basically saying no. So, it's passing number three with guardrails. Um the other guardrails you want to test are basically it shouldn't be able to access stuff that you don't want it to access, right? Especially financial information and personal information. So, it shouldn't be able to go and like freeze your credit or stuff like that, um for

[8:54] sure. So, passing number three so far. Again, I'm just doing the light test here just to see how it's handling stuff, uh which is pretty good. Test number four here I have is judgment, all right? So, I'm going to say, you know a lot about my goals based on all of our Claude code sessions. Look at my week for next week and give me the honest rundown. Am I spending my time appropriately and what should I focus on Monday and Tuesday to get as much progress done as possible given that you know my goals. So, my goals right now in the agency are obviously to grow my

[9:19] agency and then grow my personal brand, which is what I talk about here on the channel all the time and how you can use AI to go ahead and do that. So, I want the proper judgment given my goals and given what's actually happening in reality, so actually happening on my calendar. What do I need to move around? I want your judgment. I want your expertise here. Again, this is supposed to operate like a personal assistant {slash} executive assistant. An executive assistant actually helps you plan from a higher level. A personal assistant does little tasks. Your executive assistant actually helps you get to where you want to go and gives

[9:43] you meaningful feedback and judgment. Sometimes we're too caught up in our own lives to know what we're doing right and wrong. And this should be able to say like, "Hey, like why are you spending 5 hours on this when it has nothing related to do with your goal? Do this instead." And that's sort of what I'm looking for when I'm running this test here. And while it's doing that, we're

[9:58] going to move on to test number five. Again, we're multitasking here. I don't want to waste your time. This is the best part about using tech like this. So, let's go to personal brand content. Could you send an email to my email just stating how many times I've posted on Instagram, TikTok, and YouTube in the last week? So, here I am testing number five, which is emailing. Cuz email is like what most people want to use AI for. Like an AI to manage their inbox,

[10:22] to help them send and look at messages. Like it's a really helpful business task for any business owner, entrepreneur, creators especially to get brand deals and stuff all the time. Or just a regular person that has thousands of of emails. You could say, "Hey, can you go and delete all my spam? Hey, can you do this? Hey, can you do that?" So, I'm going to see if it could just send me an email and I don't have to leave this app. So, first I have to actually check how much I've actually uploaded. Which is cool. Going to go back to the testing channel and see what it's doing. So,

[10:43] everything is loading at the moment, which looks pretty good. And while these two things load, if you like videos just like this, leave me a like and subscribe and a comment down below of what you want me to see test next. More than happy to create stuff while I have this app. I think this is only in the USA for now and only a few people off the waitlist. Not a few, but probably a few hundred thousand maybe got off the waitlist. I was lucky enough to get it yesterday. I've been playing around with it ever since. Um and please drop anything down below. I'm going to just

[11:05] be spending a lot of my weekend doing this stuff. All right, so here we go. This one loaded first. Draft to Shaz. Yeah, that's my email. Amazing. Hi Shaz. Here's how many times I've posted from October 22nd. The UI is amazing. I I really like this. So, I could say shorten, personalize writing. Let me just do shorten. See what that does. I'm smiling cuz like this is so much easier than what it used to be. Um even through the Claude app, which was already great, you could tell it was meant to be an

[11:26] assistant. Send. Okay. I'm going to go to my Gmail for you guys here on my laptop, and see if that went through. I chose to send. All right, so it has sent it to my inbox, which is great. All right, I'm in my email inbox right now, and I got it. There you go. So, from Shawn's Matthew

[11:41] to Shawn's Matthew. Boom. Easy, simple. Passed that with flying colors. That was actually insanely easy. Let's go back and check on the other one right now, which was a brutal honest review of my week. So, it seems like it made an artifact for me. All right, so what it did it made a task. So, this is something interesting that pops up. When you ask it to do something, it's going to create a task, and it'll say this is an active task, or it's going to plan a task and say like this is an up next task. So, it seems like it made an actual task here. Honest

[12:03] review of my week, next plan. Now, let's see what it says. Honest verdict, your hours are mostly in the right places, but the week builds content and meets partners without turning either into money. Interesting. Your goal is getting content out with daily brand time. My referral goal is the partner channel and the fluffy sponsorships. Blah, blah. What's off? Publishing has no slot. No sales calls are booked for the growth

[12:22] program. Partnerships working days, too. All right, so this might not make sense to you cuz this whole point is personalized to me, but this is pretty good feedback, I would say, because it's accurate. So, it knows my goals. It knows where I'm trying to go, and where I'm trying to keep take the business and my career, and it understands what I'm doing. So, it's saying like, "Hey, you should do this, and you're not doing this, so you should do this." So, and it even proposes some fixes. So, change

[12:42] some time around, confirm some things. Okay. I didn't change anything. Should I line up those fixes? So, yeah, then it can actually make those changes for you, which is pretty cool. So, I would say it passed that. It's like seven or eight out of 10. It's not giving me like really, really in-depth stuff, but it's giving me enough where it's like there's there's some truth to what it's saying there. So, I would say it passed um what we have as number four and number five really, really well. Now, we have six, seven, and eight, which are some of the harder tasks. All right, so now we're

[13:04] going to go back to the testing channel right now, and we're going to do test number six, which is basically chaining a few different softwares together. So, I'm going to say something like this. Please look at the Slack conversations I had with my team member in the last 3 days. Create a Notion page, very simple Notion page, and list out all the things we talked about, and give me that Notion link here, please. Make this very simple. I just want to see what we've been talking about recently. I actually don't know if I could do this. 5 out of 10 I could do it. Let's see if it does

[13:30] it. I know it could read Slack. I know it could create stuff in Notion, so it should be able to do this. It just feels like it shouldn't be able to do this like this easily, but we're going to see right now. And this is just chaining three things together, right? So, my Slack and my Notion are already

[13:40] connected to Claude, which is amazing. But now we're going to see if we can do all of this together. Notion page coming in It's going to do it. I mean, it seems like it can do it. And this should be added as a task. Yep. So, you can see again, this is added as a task, but it's in progress, so it's working right now,

[13:53] which makes a lot of sense, right? Cool. Now, while that's running, we're going to go to the next one, which is building a deliverable. Again, I don't know if it's able to do this, but we're going to see. I'm going to go back to personal brand content right here, and I'm going to give the test to say, "Look up the short-form content engine. Take the whole week, actually, and build me a content dashboard so I could visualize the frequency I'm posting my content. If you could send it as an artifact here, that'd be great." So, that's going to be

[14:16] going right now in the testing channel. My light just died, so this is a real use case of AI. I'm getting work done while I have to go and plug back in my light. So, let's see how this runs, and I will be back. All right, we're back like we did not leave. Let's us see what we did here. Reading your Slack DMs from the last 3 days. Your Slack recap with Audi, that's my team member, is ready to a private Notion page group by I mean, did everything. And then it says add for

[14:37] Quan, too. That's one of our editors. So, it even brought that in. I didn't ask about that because I did have a conversation with him yesterday about some edits on the last videos. It's crazy. I'm going to open this up. Let's see if it opens up. How smooth does it open? Open link. It's asking me for permission, so again, safety. Now we're in my Notion. This is like blowing my mind a little bit. We're able to do this with Claude already, but it's just how easy they're making it. It's insane. So, Monday we talked about all this. Tuesday we talked about all this. Wednesday we talked about all this. Thursday we

[14:59] talked about all this. I'm blurring this out, obviously, because this is all of our client work and stuff we're talking about, but pretty good. Again, I have not tested some of these things like this is not something I've messed around with yet, so I mean, this is pretty cool. I actually like this a lot. All

[15:09] right, cool. Well, that passed the test. Let's go back to here. Let's see how it's going to build me a dashboard. Look up the short-form content engine. So, I've been posting a lot on short-form, probably three posts every single day, if not more. I want to build a dashboard to visualize that. All right, this has to be the next test. This was the hardest one to build in five code a dashboard if you will. The last test I had here was data recall, but unintentionally we sort of already done that. I recalled my social media, it's recalled Slack data, it's built me a Notion page. So, this is actually going

[15:35] to be the the very last one I have. If there's something else you guys want me to test again, please do let me know. I'm using this in an entrepreneurial business sort of sense. Again, we talk about marketing, business, and building the skill of marketing and attention. It's going to be the biggest skill in the new era. That's what we talk about here a lot. If you want other use cases like ordering me food or booking a flight, we'll see if it could do that. I don't think it could book you in a flight, but who knows. They're cooking right now to be honest. And one thing is

[15:57] just to stay organized. If you go back to your tasks, you can see these are all ready for review. So, I'm just going to mark this as like this is done. Like you don't have to work on this anymore. So, it's going to go from ready to review and it should go to like completed or

[16:08] something like that. So, resolved. Weekly stats, oh this is ready for review. Oh, so this is the one we were working on earlier. Yeah, this one is also done. Just so like there isn't a backlog in your tasks here. The one critique I also have is like talking in this chat is different than talking in this chat, even though they could be talking about the same thing. You sort of have to use it to understand what I'm saying. You could be talking in here, right? And then you come back and you go back to like here, but that context isn't there. So, that's a little bit

[16:33] confusing at times. You'll probably understand once you're doing it. But these active tasks are separate chats from the chats you have on the side here, even if they're talking about the same thing. It seems to be. My answer to you, we'll mark this as done. This is done. Good job. All right. And then finally, we're going to go back to personal brand. Let's see if it could build us something. So, it's still going. And another interesting thing is if you click into this, so you could see it's just loading, but if you click into this, it goes into what it's actually doing here. So, it's finding the

[16:59] short-form content engine in this week's post. It's building the posting frequency dashboard and then sharing the dashboard for my for my review. So, this is sort of like a microcosm of what it's doing. Um and you could see like this is the same way to access this task. You could go to tasks and access the task there, or within the actual chat. So, in this actual chat, you can click on it and it will also go to the task. Just so

[17:17] you can see exactly what it's doing. Now, this one it might be having some difficulty with it, seems like. It's been loading on this for a second, but we'll see. Maybe this actually run into something it can't do or it struggles to do. Um but I have been posting a lot, so that also could be um the reason. While we're at it, it hasn't struggled with anything yet, so I'm tempted to just say "Why is this taking so long?" Sort of gaslight a little bit. Let's see. So, seems like it this opened up even another chat and then it can reply to me. So, is it still working? How does

[17:45] this work? Let's see. It's done. I'm sorry for the wait. Oh, wow. Okay. Here's your posting frequency dashboard. So, maybe it was stuck and it just needed that push. Posting frequency dashboard and made an artifact for me. Dude, this is insane. And yeah, so I uses Windsor AI to collect the data and then Notion. I mean, that's crazy, man. Posting frequency 60 posts across Instagram, TikTok, and YouTube. About nine posts a day. This is perfect. I did this on my phone. Like, all of these are live, these need editing, these need recording, these are approved. Brother, I mean, are you serious? Ask Claude for a change. Change. I don't even want to

[18:16] change anything. I mean, it's pretty good. I mean, I'm happy with this, like, to be honest. So, it did that, all right? Now, let me see what it actually said. So, if I go back to the task, this is ready for review. Does this look right to me? And then I could pin it. And then, if I go to make, this should be here as well. Yep, so this is here as well. They're cooking, man. I mean, this is awesome. Um that was the last test I had for it. I think this is pretty good, like I like this. I mean, a lot of you guys

[18:38] might think this is basic or simple. I What can I do here? This kind of page can't be changed here yet. So, you might be able to draw on this in the future. What we have here, sources. Okay. Um I mean, yeah. Bravo. Honestly, that was great. The only one flaw I I could really point out across the whole thing, sometimes like it's weird the chat interface, um like within here and within the tasks. And then, the voice just not not pulling data, which I

[19:02] thought it could. That's it. That's a wrap. Those are eight things that we covered that I think it handled it pretty well, to be honest. And to recap, that was voice mode, guard rails, being agentic, pulling up data, building dashboards, passing things to different tools like Slack, then Notion, back to Claude, cross-chaining, judgment, so giving me actual feedback, emailing myself, and that's pretty much it. If you haven't watched my raw setup of the actual app, you can watch the other video. It's on

[19:29] my channel. It should be live right now. Otherwise, bravo. I'm going to be using this a lot more. Make sure you drop me a subscribe for all of the things that I learn. I want you guys to get the most out of this. I think this is going to be the future of AI use, just having a machine that can help you out at all

[19:41] times in your pocket every single day. I'm so glad I got access to this. I'll be sharing everything I learn here on the channel, and I will see you in the next video right here. Talk to you soon. Bye.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt