26:19The Next New Thing · Reaction
Make Money, Cut Costs, and Run Your Life With Meta's Muse
A creator reacts clip by clip to Meta's free AI agent, grading the calls, purchases, and file reads it actually completed on camera.
September 29thThe argument in one line.
Claude Duet's task view and Claude underneath make it the best-organized personal agent yet, but with no computer, no multiplayer and only one agent, it is Claude Projects with a new name rather than a full agent replacement.
Read if. Skip if.
READ IF YOU ARE…
- You already pay for Claude and want to know whether Duet is worth requesting beta access to before you rearrange your workflow around it.
- You run a personal agent like GrokBot or Muse and want a feature-by-feature read on where Duet beats it and where it falls short.
- You are a solo founder or busy professional who wants an agent to turn brain dumps into tracked tasks and pull business data into dashboards.
- You use apps with no native Claude connector and want to see the Zapier MCP workaround applied to a real email marketing tool.
SKIP IF…
- You want a step-by-step setup tutorial; this is a commentary on other creators' demos, not a walkthrough you can follow from start to finish.
- You need an agent that controls a browser or a computer today; the video's main conclusion is that Duet cannot do that yet.
- You want a team agent that lives in Slack with your coworkers; Duet is single-player and the hosts point you elsewhere for that.
The full version, fast.
Claude Duet is Anthropic's phone-first personal agent, aimed at small business owners and busy professionals rather than hobbyists or engineers running fleets of agents. Its strongest idea is a task view: brain-dump a list, Duet splits it into tasks, and you track active, upcoming, scheduled and resolved work instead of scrolling chat. Claude Code underneath makes it a strong coding agent, and it builds polished dashboards from public data or, via Zapier MCP, from apps with no connector. The gaps are serious: no desktop app, no computer or visible browser (an Uber booking died on a human check), no multiplayer, one agent instead of many, and a confusing stack of main chat, channels and task chats. The hosts call it a promising beta that is not ready yet.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:54
01 · Is this the agent to replace all agents?
Borrowed hype clip, the question the episode will answer, sponsor mention.
00:54 – 02:24
02 · What Claude Duet is
Claire Vo frames Duet as a phone-first, proactive agent synced to your Claude tools and aimed at small business owners and professionals.
02:24 – 03:09
03 · Channels
Duet borrows Slack's channel model to split work off the main chat, such as a dedicated sales channel.
03:09 – 04:21
04 · One agent vs many agents
The hosts argue threads only work with multiple agents, each with its own email, personality, skills and memory.
04:21 – 06:09
05 · The multiplayer problem
Duet is single-player; GrokBot and HyperAgent can be brought into Slack for whole teams to use.
06:09 – 07:03
06 · Brain dump to tasks
Alex Finn dumps a messy list of goals and Duet splits it into separate tasks it can start in parallel.
07:03 – 08:06
07 · Task management beats chat
The hosts call the task view the feature every agent should copy: checklists to act, chat to think.
08:06 – 09:00
08 · Proactive recommendations
Duet suggests what to build next with one-tap buttons that kick off Claude Code work and pull requests.
09:00 – 10:03
09 · Too much proactivity
The hosts want action, not a stream of nudges, and suspect agent makers use nudges to stay top of mind.
10:03 – 11:24
10 · No desktop, no computer
Duet is mobile and web only and has no computer of its own, so it cannot shop, browse blocked sites or manage WhatsApp.
11:24 – 13:39
11 · The Uber test
Peter Yang's Uber booking fails on a human check he cannot see or reach; the hosts explain why errands like this are the point of an agent.
13:39 – 15:27
12 · YouTube analytics dashboard
From a channel link alone, Duet builds a performance dashboard with breakout topics, solo vs guest results and length analysis.
15:27 – 17:06
13 · Skills and the morning brief
Typing slash reveals built-in skills like a styled morning brief; the hosts debate visible vs hidden skills.
17:06 – 18:18
14 · Scheduled check-ins
A default check-in scans email, calendar and apps every three hours; the cadence is adjustable by asking.
18:18 – 21:36
15 · Email dashboard via Zapier MCP
Brock Mesarich connects Drip through Zapier MCP and gets an email campaign dashboard from his phone.
21:36 – 22:57
16 · The task folder
Peter Yang tours active, upcoming, scheduled and resolved tasks, each a sub-chat off the main thread.
22:57 – 24:27
17 · Artifact collaboration
Riley Brown comments directly on a slide deck and Duet applies the edits, or he edits the artifact by hand.
24:27 – 27:18
18 · Voice editing
Riley dictates a batch of slide changes in one voice note; the hosts compare it to real-time voice coding.
27:18 – 28:12
19 · Channels vs tasks vs threads
Three nested chat types make Duet hard to explain and easy to get lost in.
28:12 – 29:24
20 · Artifact recaps
A scheduled channel resurfaces unfinished decks and docs so long-running work does not get forgotten.
29:24 – 31:03
21 · The verdict
One host is bullish thanks to Opus underneath; the other calls it Claude Projects rebranded and not ready.
Lines worth screenshotting.
- Duet is positioned for small business owners and busy professionals, not casual users and not engineers running fifty thousand agents.
- Threads only make sense when each thread is a different agent; one agent spread across many threads takes on too much and has no clear identity.
- A separate agent can have its own email address, personality, skills and memories, which a separate chat thread with the same agent never can.
- Neither Anthropic nor OpenAI shipped multiplayer in their personal agents, which leaves team use to tools that live inside Slack.
- The best agent interface is a task list for doing and a chat for thinking, not chat as the only way to steer background work.
- Proactive suggestions are only useful when they arrive as one-tap buttons that start real work, like opening a pull request.
- Duet ships with a scheduled check-in that runs every three hours from 9 a.m. to 10 p.m., and you can ask it to run less often.
- An agent without its own computer cannot finish real-world errands: Duet's Uber booking stalled on a verify-you're-human check the user could not see.
- Asking Claude to build a dashboard from only a public YouTube channel link produced topic, format and length breakdowns with no API access.
- When an app has no Claude connector, Zapier MCP bridges it; Drip exposed seven tools including stats, tagging and adding to campaigns.
- Paying a SaaS an extra $100 to $200 a month just to unlock MCP access is avoidable when a Zapier MCP bridge does the same job.
- Hidden skills trap your work habits inside one vendor; skills you can see and export can move with you to the next agent.
- Telling an agent to log every task in Notion tends to fail over time, so a built-in task system is more reliable than an instruction.
- Three nested chat types (main chat, channel chat, task chat) make a product harder to explain with every level added.
- A daily artifacts-recap channel that resurfaces unfinished decks and docs solves the problem of losing work across long-running agent jobs.
- Without computer access, desktop software or a team of agents, Duet reads as Claude Projects repackaged under a new name.
Task views beat chat, but agents need computers.
WHAT TO LEARN
Duet gets the interface right by turning agent work into a visible task list, but an agent with no computer, no teammates and one identity cannot finish the work that matters most.
02What Claude Duet is
- Duet is built for small business owners and busy professionals, connected to the same tools and plugins your Claude account already uses.
- It is mobile and web first, so judge it as a phone assistant rather than a desktop workstation.
03Channels
- Channels split work off the main chat by area, like a sales channel, using the Slack model most teams already know.
- Treat channels as filing, not as separate workers; the same single agent still handles every channel.
04One agent vs many agents
- Give each distinct job its own agent with its own email, skills and memory instead of another thread on one overloaded agent.
- An agent with a clear role can act in the world on your behalf; a thread cannot carry a separate identity.
05The multiplayer problem
- Duet is single-player, so if your team needs to share and direct agents, look for one that lives inside Slack.
- Personal and work use blur together quickly, which makes it harder to bring a personal agent into team workflows later.
06Brain dump to tasks
- Dump every open loop into one message and let the agent split it into separate tasks you approve with a tap.
07Task management beats chat
- Use chat to think and a task list to act; the list shows what needs input and what is already running.
08Proactive recommendations
- Proactive suggestions earn their place when they come with buttons that start real work, such as building a feature and opening a pull request.
- Duet runs coding work through Claude Code, which makes it a strong mobile front end for shipping code.
09Too much proactivity
- Too many unsolicited ideas create agent fatigue; tune proactivity down until suggestions are rare and worth acting on.
- Judge an agent by the actions it finishes, not by how often it pings you with something new to consider.
10No desktop, no computer
- Without its own computer, an agent cannot buy from Amazon, browse sites that block bots, or manage WhatsApp for you.
- If you live at a desk, expect friction until a desktop app ships; for now Duet is a phone and web tool.
11The Uber test
- Agents that book rides or food need a browser you can see and step into when a human check appears.
- The real payoff of errand automation is timing: an agent that knows your calendar can order the car before you remember to.
12YouTube analytics dashboard
- Ask Claude to build a dashboard from any data you own; a public channel link alone produced topic and format performance insights.
- Point the same request at bank or subscription data to see where money goes, a job that once needed a bookkeeper.
13Skills and the morning brief
- Type a slash in Duet to list built-in skills, including a morning brief that renders as a styled page.
- Prefer agents that expose their skills so your work habits stay portable instead of locked inside one vendor.
14Scheduled check-ins
- Duet's default check-in scans email, calendar and apps every three hours from 9 a.m. to 10 p.m.
- Five check-ins a day is a lot; ask Duet in plain language to cut it to a morning and afternoon pass.
15Email dashboard via Zapier MCP
- When an app has no Claude connector, create a Zapier MCP server, connect the app, and the agent gains its tools.
- Once connected, a single question from your phone can return campaign stats plus a dashboard of opens, clicks, unsubscribes and drop-off.
16The task folder
- Duet tracks work as active, up next, scheduled and resolved tasks, each a sub-chat off the main thread.
- Built-in task tracking beats telling an agent to log tasks in Notion, an instruction agents tend to drop over time.
17Artifact collaboration
- Comment directly on a slide or document and Duet applies the edit, or edit the artifact by hand without chatting at all.
- Artifacts like decks and design mockups live in the web and iOS app, so review and revision happen where they were created.
18Voice editing
- Batch several edits into one voice note while looking at the artifact, then send them as a single request.
- Voice works best when changes apply in real time; one dictated block that runs later is slower than live voice coding.
19Channels vs tasks vs threads
- Duet has three chat types, main, channel and task, and it is easy to lose track of where a conversation lives.
- When a product needs a diagram to explain where conversations go, simplify the structure before adding features.
20Artifact recaps
- Set up a scheduled channel that reminds you of unfinished decks and docs so long agent jobs do not get forgotten.
- Better still, turn every important chat into a task so unfinished tasks become the reminder.
21The verdict
- Duet's Claude and Opus backbone is its biggest strength, and the hosts see it as a strong start.
- Missing computer access, desktop software and multiple agents make it feel like Claude Projects rebranded, so treat it as a beta.
Terms worth knowing.
- Personal agent
- An AI assistant that runs tasks on your behalf across your apps, often proactively, instead of only answering questions in a single chat window.
- Channels
- In Claude Duet, separate named workspaces (for example a sales channel) that branch off the main chat, borrowed from the Slack model of organizing conversations.
- MCP (Model Context Protocol)
- An open standard that lets AI models call external tools and data sources through a common interface, so one connection works across many AI apps.
- Zapier MCP
- A Zapier service that exposes its library of app integrations as MCP tools, letting Claude or other agents act in apps that have no native connector.
- Artifact
- A standalone output an AI produces, such as a slide deck, document or dashboard, that you can open, comment on and edit outside the chat stream.
- Multiplayer
- An agent setup where several people on a team can talk to, manage and direct the same agents, typically inside a shared tool like Slack.
- Human verification check
- A website test meant to block bots, such as a click-to-confirm box, which an agent's browser often cannot pass without a person stepping in.
Things they pointed at.
00:56channelClaire Vo (How I AI)
01:30productMuse
02:00productGrokBot
04:57productHyperAgent
06:10channelAlex Finn
08:50toolClaude Code
11:30channelPeter Yang
18:20channelBrock Mesarich
18:35toolDrip
18:55toolZapier MCP
22:40toolNotion
23:00channelRiley Brown
25:00toolChatGPT voice mode
Lines you could clip.
03:31
“Once it's just a different conversation with one agent, it takes on too much and it can't have a personality that interacts with the world in a clear, meaningful way.”
05:28
“The fact that both Anthropic and OpenAI missed multiplayer mode is a huge, huge miss.”
07:45
“The more we get task-oriented, the more I want to see checklists and tasks and project management being the way that I interact with my agent, and chat the way that I think it through.”
09:13
“I don't want too many suggestions. I want action.”
10:50
“It doesn't have its own computer. Big miss, don't you think?”
15:15
“I connected my finances in. I got a really nice dashboard of where my money was going. It's better than a bookkeeper.”
16:20
“More and more of these agents are hiding their skills away, which means that I have all my work habits trapped in their systems.”
19:10
“I have to pay an extra $100, $200 a month just to get the level that includes MCP access? Come on. I'm not doing that.”
30:10
“How about if we take our Claude Projects, which is essentially what this is, and we make it into its own name with its own little twist?”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
Anthropic just released their Grokbot and ChatGPT .killer. It's called Claude Duet, and it might just be the best personal AI agent out there. In this video, I'm covering everything Claude Duet, what it is, how it works, how to get stuff, how to get access to it, because not everyone get access to it, and which AI agents you should be using.
Every new agent for him is the best possible changing of based agent in the world. Is this one it? We're going to talk about it.
I've watched endless videos about this. I've looked at how this works and we're going to talk about whether this really is the agent to replace all agents or if it's flawed and where it's good, where it helps your business, where it won't. And Andrew, you are a fanatic like me, right?
So you want to see this and we're going to talk about all the uses. Presented by Zapier, the AI automation company. Let's go before all the hype.
Let's go into just some basics. This is Claire Vo. She's talking about what this new thing is.
phone. iPhone Android based personal agent that is synced up with Claude and does all the things that personal agents do. So it is proactive, although I will go through how proactive it really is.
It's proactive, it can use all the tools and plugins that your Claude has access to. And they really position this as being a personal agent for small business owners or busy professionals. So none of this like your mom's going to use it with Muse and none of this like you have to be Lauren at Cursor, who's going to run like 50 ,000 agents and ship a gajillion PRs.
Okay, I'm good with that. I do find that Muse is really not made for business. It's made more for fun.
You too? I fully agree with Claire. I think this is more for business than Grogbots or Muse.
And it's very proactive when it has a taste of anthropic that is tailored mostly for businesses. Interesting. Do you actually think this is better than Grok?
It depends from the use case. I'm a huge GrokPod fan, but I do like how they connect all the channels and the discussions you already have. So I will not pick one.
I will pick one. By the end of this video, you're going to know exactly which one I think, but let's look at the way that it's structured. And this is one of my problems with it.
I really love about Duet is the idea of channels. So we're all trying to decide what the primitives of are for agents. In Codex, you have all these threads in chat.
You have all these chats in Cloud. You have threads and chats and projects, all sorts of things. In Muse, you have your main chat and side chats.
Everybody's trying to figure this out. And Anthropic borrowed a little bit from Slack, which I feel like is everybody's favorite collaboration service, which positions these subchats, these subthreads, these subplaces that you're going to work in as channels.
So now you can kind of think of your duet as a main thread, a main chat, but then you can break off other chats into channels and have specific channels. Like here, I created a sales channel for working on sales projects. This is my main problem.
I think the idea of threads is nice as long as you have multiple agents. Once you have one agent and multiple threads, and threads are how you keep track of agent conversations, it's not good. And we'll get into more of it as this goes on, but that's my big frustration here.
Right? I think what you need is a different persona. I need an agent that works with my clients.
I need an agent that goes and gets me new clients, an agent that helps me make sense of my inbox, not a different thread. And the difference is if you have a whole agent, you can give it an email address. You can give it a personality.
You can give it a set of skills and a set of memories. Once it's just a different conversation with one agent, it takes on too much and it can't have a personality that interacts with the world in a clear, meaningful way. And that's the big disconnect here.
This is more like your personal agent for work where Grokbot is multiple agents using for whatever you want. Okay, I'm seeing some nodding. I'm going to show you one other problem.
So she was very high on this. She did say that one problem she had was that there's agent fatigue. There's so many different agents.
She doesn't want to try too many. And then she had another issue. And I'll just keep showing people both wins and frustrations with this.
The final thing that's really missing is, look, we don't work. alone and duet right now is clearly a agent for single player use in fact like most of the clod projects feel like very good products for single player use i just don't think even with clod tag clod and slack all these things i don't think they've quite cracked multiplayer and i think that there have been some so grokbot to some degree did get into multiplayer.
You can actually bring it into Slack with other people. But so has HyperAgent. Do you use HyperAgent internally at your company?
I'm going to tell you about HyperAgent. It is a set of agents that you can build on the web. You can interact with them on the web.
You can give multiple people on your team the ability to manage and lead them. And you can bring them into Slack. And so we have them in our Slack.
And people who are not agent -obsessed like you and I are using it. They're interacting with it. They're asking it to create landing pages.
They're asking it to create descriptions and briefs and this and that. And it's just all happening in Slack. I think the fact that both um anthropic and open ai missed multiplayer mode is a huge huge miss and it's a frustration point they do have some benefits that i want to show but that's that's my big frustration here Let's go into another one.
And a question that comes to my mind is, are they going to be work -related or are they going to stay on a personal level too? Because it might not be multiplayer, but now they're branded as your personal assistant for not so much work -related connections and apps. Yeah, she was talking about using it to get warrior tickets.
And so the more it gets... personal and business, the more complicated it gets to disambiguate them and bring them into work. Okay, let's look at what does work really well about it.
Here's Alex Finn, our man. So let me just show you what an idea of like a brain dump would be right here that it breaks down into tasks. And this is one of the things that he liked.
You could just brain dump and let the agent figure out what to do with what. So I just put in everything that's on my mind into this task creator. I said, I want to write a newsletter on Herder.
I also want to start planning out multiplayer in Knoll District and how that would work. I also just got a second DGX Spark. I want to see what the best model are for that cluster.
And I'm just going to hit send. And what now Duet's going to do is it's going to break those down into tasks. Either save it for later.
I got to say start, start, start. This is great. And now let's look at my task list in my active.
And you can see all the tasks that are waiting for my review. And if I scroll down, you can see which ones need my input and which ones are doing work. And now it's doing the multitasking of a Grok bot, but it has the simplicity of a chat GPT dots God thread.
This is my favorite feature. And I think that every agent needs to come. Tell me about this.
What do you like about it? I really love how Duet and also Muse introduce new UIs that is not just chat. You have a view for your tasks.
You have a view for your goals on Muse. And I think that gives you a better control on what the agents are doing in the background versus Grogpot that it goes away and comes back with the result. That's exactly it.
I wish everyone would copy this. You're using chat as the interface for agents, but the truth is the more we get task -oriented, the more I want to see checklists and tasks and project management being the way that I interact with my agent and chat the way that I think it through. I would love for every agent model to steal this, and you're right.
This is one of the best parts of Muse. Muse will give you a place to say what your goals are, and then you work together towards them. Okay, another benefit, the proactivity.
making recommendations on what you should build next without me even asking it just, Hey, by the way, I think we should work on these five things next. Do you give me the go ahead? Right?
So I wake up today. I say, Hey, let's work on Henry. He goes, okay.
Here's what we should build out next. It gives me really good recommendations of what to build. A lot of times I'll go to AI agents, I'll say, hey, what should we work on next?
And it gives me like total slop. These are actually really good recommendations on what I should be implementing next. Then you can't see here, but it has like buttons it automatically creates on with each one of these features in them.
I just hit the button and it starts working on it. One of the buttons I hit was start the pipeline victim. Click the button, boom, it goes, it starts working.
It creates the PR. Then it comes back a couple minutes later. It says, should I open the PR for you to review?
Then it has a yes button or no button. I hit the yes button. By the way, it's using Claude Code for all this.
Right now, Claude Code with Opus 55 is the best coder out there, hands down. No question about it. So instantly, Claude Duet becomes the best coding personal agent out there.
What do you think of proactivity with agents? I'm not sold yet to it because similar to Claire, I have a fatigue of too many agents coming to me with suggestions. And I don't want too many suggestions.
I want action. You know what? You're right.
I do like that he showed an example of how it was proactive in a smart way. But in general, I don't want them pinging me with, I've got another idea for what I could do. I've got another suggestion.
Exactly, yeah. For the most part, they're completely off base. I don't need them.
And what they're trying to do is the makers of them, I feel, are trying to worm their ways into my life and to remind people why they should be using an agent. And you and I, we're not in the world of needing to be reminded. So I like this level of proactivity and the intelligence of it.
I don't need it to be so nudgy. But you know what? Let me jump over for a second here to Peter Yang.
Actually, we'll get to Peter Yang because he also talked about proactivity in a way that I think is helpful. But let's talk about a weakness from Alex. There are still some weaknesses to Claude Duet.
Right now, it's just on mobile. There's no desktop component. There's no desktop app.
I assume this is because it's in super, super early access. There is a web version and we're going to see people show that, but there's not a desktop app. I assume a desktop app will come soon.
So that kind of stinks. I wish I could kind of lock it on my computer and talk to my agent. It is what it is.
Two, and this might be the biggest weakness. This might be a reason why I still use Grokbot a ton at the moment, is that it doesn't really have a computer of its own. It can do cloud code, cloud agents, where it spins up cloud agents to write code for you, but it doesn't have its own box.
It doesn't have its own computer. Big miss, don't you think? Yeah, I think that could be a deal breaker for many because it doesn't have its own computer.
It cannot access your computer. Then what, only cloud agents? So basic things like needing to buy from Amazon, it can't do.
Needing to go and do research by clicking on sites that often won't allow agents. Managing my WhatsApp for me, really important. It's just such a disappointment that they miss so much with this.
You know what? I'm going to jump to Peter Yang for a second because he has an example of where having a computer helps. One shortcoming is, for example, I asked it to book an Uber ride.
because there's actually a default skill called uber ride in duet so if uber ride there's also skills book door dash and so on so when i type in the skill it asked me you know where should the car pick you up and is it right now and where is it going and i said palace of fine arts and unfortunately it was not able to do it because us browser required verify your human check from uber and it was able to click through this so that required me to basically like the other personal agents, actually open Duet's browser and navigate this human check for it.
But unfortunately, at least right now, Duet doesn't actually let me see its browser or let me interact with this browser. I couldn't find a single place in this app where I can actually engage. Now talking to the Duet PM, I know that he's hard at work on this feature.
So I think it'll be coming eminently. I do like that he has access. Deal breaker, right?
Yeah. And when I talked about how I use an agent to order from Uber Eats or Uber, people laughed at me like I was overkilling. And I'll tell you why.
When it comes to Uber Eats. If I'm having people come over, like I had somebody come over for dinner last Wednesday, I needed vegetarian food for three adults and three kids. Here's the type of food that we eat.
My wife needs it healthy and not not oily, etc. I could just dump that into a chat and let it go find the places. It says, here's what I have.
It creates a beautiful menu for me to share with my wife. So she could say, this is what I want. This is what I don't want.
And then meaning like menu of catering almost. And then I give it to the agent. Similarly with Uber, I know that I will not, five minutes before leaving, remember to go call an Uber.
I'm just in the moment. But if my agent knows, Andrew has a meeting in a little bit. He's not going to order an Uber.
I'm going to have an automation set up to go and order him an Uber so that whenever he needs to leave, the car is there. That needs a browser, and that is one of the beauties of having an agent that actually accesses this stuff. This is why AI is made for.
Right. Call me pampered. I want that.
Okay. All right. Let's get away from the pampering and look at some business performance.
It basically looked at a year of my published videos and their public data to come up with this report. So we have subscribers and a number of videos published up here. And what's more interesting is down here, we're sharing insights about my videos.
So for example, it found that personal AI agent videos are my breakout topic. And actually my solo videos are doing better than the guest. unless the guest has a practical demo or something to show if i scroll down even more there is some pretty amazing charts so here is a chart that shows every video by publish date and how they performed and you can see here that my solo videos started doing better this year compared to my podcast videos and if i scroll down more there is also keywords that perform personal ai agents model showdowns cloud videos perform and also how long each video is by length And by the way, to make this awesome dashboard, all I said was like, hey, can you make a dashboard and insights on my YouTube performance data?
And I linked my YouTube channel. I love this use case. What do you think?
And that's the killer benefit of DoIt. It comes with Cloud. The Cloud we love and we know.
And you get that into the app. I still have Claude even with Grokbot and the other agents. As long as they have a computer, you can connect them to Claude and Codex and the terminal.
Having said that, This is a really beautiful dashboard. It's great research done.
And he didn't even give it access to YouTube Studio or API access. It was just straight up public information. Go figure it out.
And anyone who has any kind of data in their lives should ask Claude to create a dashboard like this. I connected my finances in. I got a really nice dashboard of where my money was going, what subscriptions I'm paying for, what categories.
It's better than a bookkeeper. This is an incredible use case. I agree.
He did mention skills earlier. I think that this is interesting. My hidden feature in Duet is that if you type slash, you can see all kinds of skills that you can run inside this app.
And down here, there is a morning skill, which basically renders the user's morning brief as a styled HTML artifact. So that's basically what I did. I said, morning, let's try out the skill.
And then it rendered a morning brief. Let's take a look. And here we have it.
As you can see, it also has that cloud aesthetic. I love how it says that I have two short calls for nine o 'clock and then the day is wide open. Now that I'm independent, I try to avoid meetings as much as possible.
You know, I know I have a lot more in this clip, but I'm going to cut it right there. It's an interesting use of the morning brief. I don't like morning briefs in general, but this is pretty.
I think the bigger point here is it shows the skills. I don't like how more and more of these agents are hiding their skills away, which means that I have all my work habits trapped in their systems. I want skills that I could then take to another agent if I want to.
I agree. That's also a personal taste, to be honest. Some people don't really like to.
know the skills, they trust the agents to update them, use them, maintain them. Some other people like to have access to it. But I will double down to what I said earlier that this is the beauty of having cloud within your personal app just because maybe we have get used to it from cloud desktop or the work we do every day and it feels familiar.
Remember, we were talking earlier about proactivity. I think it's important to say that you decide how proactive this is. So I mentioned how Duet is the most proactive assistant that I've tried.
And the reason is actually pretty simple. It has a scheduled checking task that comes by default. If I go here and say view history, you can see that this task basically runs every three hours from 9 a .m.
to 10 p .m. And I'm sure it checks my email calendar and other connected apps to find work. that they can do for me.
So here's an example. It says, morning to you, Peter. Here's a sponsorship deal with Diana that you're working on.
And there's also a Mercury payroll video that you should launch. By the way, Mercury payroll is a potential sponsor for him. Lovable Diana is from, or Diana, I think is from Lovable.
Nice work. And you got to prepare the Friday podcast recording with Nate. So basically, you know, , it brings things to me.
But honestly, I think five check -ins a day is a little bit too much. I prefer maybe once in the morning and once in the afternoon. But you can also just ask Claude and do it to change the schedule.
All right. This is what you and I talked about before. Too much.
We don't need all that information. Let's get to Brock. He has a couple of things that I think that are interesting.
The first is the report on email campaigns. So with YouTube. Peter was able to go in, get public data to analyze his YouTube account.
Brock wanted something else. Now, one thing I wanted Claude Duet to be able to do is pull my email marketing stats from Drip, which is the app that I use in order to run my email marketing campaigns. So for example, I want to be able to ask how my last email marketing campaign did right from my phone.
But the problem is if I come over to the connectors inside of Claude and search for Drip, there's no Drip connector. Yeah. So you know what he did.
I mean, you're from Zapier. So in order to get around this, I'm going to use the Zapier MCP, which basically lets Claude or any of your other AI agents connect to over 9 ,000 app connections. I'm going to pause here for a second before he shows how to do it.
This is the unlock. Dude, there's still software that charges you. I'm using one.
I'm not going to call it out because I don't want to get into a fight with the founder of it. He's loving fighting everybody. But I have to pay an extra, what, $100, $200 a month just to get the level that includes MCP access?
Come on. I'm not doing that. It's just not – it makes me angry not from the cost but from the – Just the justice of it.
Okay, so you can always use Zapier to get around that. In order to get around this, I'm going to use the Zapier MCP, which basically lets Claude or any of your other AI agents connect to over 9 ,000 app connection. So in order to do this, I'm going to come over to Zapier MCP.
And if you want to use this, there will be a link in the description to use it yourself. Then on the left -hand side, once I'm inside of my account, I'm going to click on MCP servers, which is where we set these up. And I'm going to click create server.
Next, it's going to ask which AI tool we want to connect. So you've got options like Clod, Clod Code, ChachiBT, Cursor, and even Grokbot. So I'm just going to click on Clod Code for this instance.
And then I'm going to click on Finish Connecting. Then all I need to do is copy this prompt, paste it back into Clod Code, and it's going to set up the entire thing for me. Now, while that's running, I'm going to come over to Apps, which is where we pick the applications that we want Zapier to use.
And I'm going to search for Drip. And as you can see right here, Drip has seven different tools that Claude can use. So it can pull my stats and even do things like tag subscribers or add someone to a campaign.
So I'm just going to click on connect, sign into my Drip account, and then that is it. And it's pretty much just as easy to do this with any of the other apps that you use on a day -to -day basis. So now if I come over to Claude Duet on my phone, I'm just going to ask it how my last email marketing campaign did.
And I'm even going to ask it to generate a dashboard to visualize all this data for me as well. About like a minute later, we now have this message here. It pulled up the statistics from our email marketing campaign in Drip via the Zapier MCP.
If I click on this, it brings me to the dashboard that it created. All right. And just like that, we now have this dashboard with all the statistics pulled from our Drip account.
Shows the open rate, shows the click -through rate, shows the unsubscribe rate, delivery rate. It even shows where the viewers, the readers drop off. To be able to take data and then have it organized in a dashboard that's useful is incredibly helpful for me.
I do this now all the time. I agree. Let's go back into Peter Yang because we need to see tasks.
Tasks, you and I said, are a killer feature here. He goes deep in it, deeper than anyone else. Let's take a look at the task folder right now.
So a task is a thread inside the main cloth thread that you see here. And here's a bunch of tasks that I have going on. Prepare Tuesday call with Diana for sponsors.
Plan tutorial about Claude. And find show times for you can see everything, the movie, and so on and so forth. And you can see here that here are my active tasks.
There's a bunch of tasks coming up next. There's a bunch of scheduled tasks, which are basically like scheduled routines and scheduled jobs that run every day or every week. And then there's also a resolve tasks folder.
So in many ways, this works a lot like cloud projects. You have a main chat here with sub chats for each task on the right.
Beautifully done. And that's a real problem. And I'm happy that they solved it this way because many times you have tasks here and there or discussions here and there, especially let's say on Grog bot with different bots and then you get lose track of them.
I keep telling my agents. put all tasks into notion and inevitably they stop doing it i don't know why and so This is a surefire way to see it.
And the fact that it's a built -in ensures that it gets it done. Okay. Now let's look at Riley Brown for collaboration.
We talked about getting artifacts out. Let's look at how to collaborate on them. Is that the experience of creating any sort of asset, whether that is a slide deck, whether that is a Figma style design mode, mock -up, it's really easy to add things.
They have, I forget what they call it. It's called a comment tool. So I could just comment on this and I could just say, make number three, adding comments, great artifact editing on web.
And I can add this comment and you can see here, it automatically sends this to my duet. And so duet is now going to make the edits and it will show up here. And then I could come here and just say.
Hey, can you make these super easy to comment and then add an image on this slide right here, which is the thing three details page, except crop out the little edge here. And so the point I'm trying to get across for number three is that this like commenting and editing documents directly inside the web app is incredibly easy.
And it's as easy as their baked in design feature. And look at that. It added that image to the board.
And yeah, add two little bullets. Basically, super easy to create assets. Great for presentations and baked straight into the web app and iOS app.
And it should add these bullets here. But again, you can add these bullets directly. I don't have to do it through Duet.
The ability to even edit it directly is helpful. I don't like to keep chatting. Very practical.
Yeah.
Do you use voice mode for your agents? Not as much. I use a lot of voice mode when I do chat GPT codex coding, but not on my personal agents.
Wait, you code using voice mode? Yeah. You sit back and you just ask for changes.
Interesting. And what do you use? What harness do you use to edit or to code with?
Mostly chat ZPT for the voice mode. Oh, meaning just like their desktop app? Yeah, yeah, ZPT Live.
Wow, okay. So this also comes with the voice mode where you get to interact with your agent and chat with it. But I like how Riley Brown showed this way of talking.
through edits that you need. Watch this. I can very easily zoom in on the slide deck.
But instead of typing my change here, you actually can't use voice. What I'll do is I'll exit out and I'll just turn on voice right here with this button. Hey, I want you to make some changes to the presentation.
And yeah, I want you to make a good amount of changes. Now watch what he does here. On all the slides that has nine things you need to know about Duet, can you please make that like there should...
He didn't he didn't say it specifically, but now he moved back into the slide view and is continuing to be a line break after need and then make the text a little bit bigger for the voice mode above the little timeline we have at the bottom. Can you please make a cool little graphic that's only gray? That's just like very elegant and nice and kind of brings attention to the middle.
And then add one little bullet there based on how I've been using Duet so far. And then after slide 19, I want you to add a standalone slide that just has text in the middle. And have the text say, I'm not sure how I feel about this one.
And put this after slide number 19. Okay, so once I've done announcing all of my changes, I can hit the X. It's been recording my voice all the time.
And we can fire off that prompt. And now it's going to make all of those changes.
I like that. I would love to see real -time changes there, not just a block of text that's sent and you're waiting for the changes to happen. That would be a game changer.
Do you get that when you're chatting with ChatGPT on your code? Doesn't it take too long? It doesn't take too long for it to make the changes?
No, no, no. It's real time. You just need a few poses here and there, but it's working while you're talking.
This one is technically a transcript. Yeah. Okay.
I like your point of view. All right. I'm going to show two more.
Let's start with the frustration that he had, and then we'll close it out on a technique that he has to get more done. It's pretty easy to get lost between channels, tasks, and threads. I think having three types of chat threads is a little bit confusing.
You have your main chat, you have your channel chats, and then your task chats. And so it gets harder to explain the software. The more levels you go, it gets infinitely harder to explain because like, are these called chats?
Are these threads? Right here, you just have a chat. You're chatting with an agent.
And now you're chatting with an agent and channel. Now you're chatting with a task chat. inside a channel.
So you can get lost. I can feel the pain just from the UI. Yeah, it's a lot.
I really think what they need to do is stop with one agent, create multiple agents, give them computers. I mean, really. All right, let's look at a final problem that he has and a solution that he has for it.
I create a lot of these documents, sometimes to send to my team, sometimes for myself. And it's really useful to have a dedicated space where all I do is get, like, I'll get reminded to work on things that I've already created. And so every day at 9 a .m.
to 5 p .m., I get an artifacts recap. And so it'll be like, hey, you created this slide deck yesterday. Do you want to continue working on it?
And I'll be like, oh, yeah, I forgot I created that. I need to do that for my presentation or for my video. It's really useful to have that because oftentimes with AI, since they work for 15 to 40 minutes at a time, you want to be able to work on multiple things at once.
And it's easy to get lost. So I have this dedicated channel, which basically resurfaces all the things that I'm working on. I like it.
Scheduled check in. As we speak, I have 10 or 15 chats like that waiting for me to follow up. This is where I feel like the checklist would be helpful, that we're not just chatting, but every chat results in a task, or the important ones do, and then the unfinished tasks should become what nudges me back.
But barring that, this is a good solution to a common problem. Okay, overall, what's your take on this? On the last one?
No, on Claude Duet. I think it's a great start for Anthropic. The details that we discussed, maybe they are opinionated, but I think they really need more bots, more agents than just one.
But I think it's a great start, especially that it has Opus behind the scenes, which is one of the most important things about your agent. I'm very bullish for it. I'm glad they finally got in the game.
I'm disappointed that it's just so small. No computer access, no desktop software, no team of agents. It's just another way of interacting.
It almost feels like they saw everyone else getting in the game and they said, oh, shoot, we got to get into two. What can we do? How about if we take our Cloud Projects, which is essentially what this is, and we make it into its own name with its own little twist?
Okay, it's kind of a Cloud Projects product. We don't have to create something brand new. Boom, they launch it.
It's not ready. It's not ready. And partially it's not ready because, I mean, you can see because it's still in beta.
It's not available to everyone. And you can see some of the frustrations people have had. Having said that, if you want access to it, we will have a link to it below.
I'd love to know what your take on it is. Some of you people have been really harsh. Even when I love stuff, you're really harsh on it.
So I wonder what your take is on all this. And if you've heard Andrew and I talk about Grokbot, we've got a video for you where you can see what we're thinking about Grokbot right there. See you in that video.
The Hook
The bait, then the rug-pull.
The cold open borrows another creator's hype line, calling Duet the best personal agent out there, and the hosts spend the next half hour testing that claim against the best public demos. The verdict is split: the task system is the thing every agent should copy, and the missing computer is close to a deal breaker.
Named ideas worth stealing.
22:00list
Duet's four task states
- Active
- Up next
- Scheduled (recurring routines)
- Resolved
Every task is a sub-chat off the main thread and lives in one of four buckets, so background work is tracked like a project board instead of buried in chat.
Steal forany agent or ops dashboard that needs to show what is running, queued, recurring and done
03:30concept
Agent vs thread
- Own email address
- Own personality
- Own skills
- Own memories
A real separate agent carries its own identity, tools and memory; a separate thread with the same agent does not, so it cannot act in the world as a distinct role.
Steal fordeciding whether to split work into new agents or new conversations
27:23model
Duet's three chat layers
- Main chat
- Channel chats
- Task chats inside channels
The organizing model the hosts flag as confusing: every added layer makes the product harder to explain.
Steal forauditing your own product's navigation depth
19:40list
Zapier MCP bridge
- Create an MCP server in Zapier
- Pick the AI client (Claude Code)
- Paste the setup prompt into the client
- Connect the app (Drip)
- Ask the agent for stats and a dashboard
A five-step path to give Claude access to an app that has no native connector.
Steal forany tool in your stack missing a Claude connector
How they asked for the click.
VERBAL ASK
30:35next-video
“If you want access to it, we will have a link to it below. I'd love to know what your take on it is... we've got a video for you where you can see what we're thinking about Grokbot right there.”
Soft close: link in description, a comment prompt framed around harsh audience opinions, and a pointer to the companion GrokBot episode.
MENTIONED ON CAMERA
Visual structure at a glance.
