Claude Duet Is Here! Everything You NEED To Know

Peter Yang

AI-assisted summary of Claude Duet Is Here! Everything You NEED To Know — original source. Check the original for full context.

Source published: 2026-10-21

Key ideas

Anthropic introduced Claude Duet as a dedicated personal agent research preview for Claude Pro and Max users, designed to run proactive background checks across connected accounts without manual prompt triggers. (0:00)
Duet runs automated background checks every three hours between 9:00 AM and 10:00 PM (five times daily) to parse schedules, unread emails, and pending projects into actionable status updates. (4:46)
The agent can analyze past communication history—including 33 emails, 136 Slack messages, and 11 internal documents—to build a persistent writing persona for drafting emails and briefs. (6:23)
A major operational limitation is Duet's cloud browser isolation: it cannot bypass verification or bot checks (such as Uber's human check) because users currently have no way to view or control the browser session. (12:07)

Detailed notes

## Background Published on YouTube on October 21, 2026, by Peter Yang, this video provides an in-depth walkthrough of Anthropic's new personal agent, Claude Duet. The page description highlights Duet's proactive check-in system, hands-on artifact builds across daily productivity workflows, its existing limitations in browser execution, and how it stacks up against competitors like Grok Bot, Muse, and ChatGPT Dots.

## TL;DR Peter Yang tests early access to Claude Duet, examining its out-of-the-box architecture, autonomous background sweeps, and dedicated task threads. He demonstrates how Duet generates personalized daily artifacts—such as morning briefs, email triage boards, YouTube analytics engines, sponsor revenue dashboards, and interactive learning tools—by analyzing his writing style and schedule. While praising its sheer proactivity and design polish, Peter outlines major friction points, including its inability to solve browser bot checks or execute local machine skills. Ultimately, he compares Duet against Grok Bot, Muse, and ChatGPT Dots, concluding that Duet holds tremendous promise as an integrated chief of staff once cloud browser interaction is solved.

## Atmosphere & style The presentation feels calm, pragmatic, and refreshingly candid, carried by Peter Yang's structured, product-manager mindset. Rather than getting swept up in artificial intelligence hype, the delivery balances genuine enthusiasm for sleek artifact generation with dry, transparent critiques of real-world failures like stalled Uber bookings. Watching the video feels like an over-the-shoulder co-working session, moving seamlessly from productive automated routines to grounded evaluations of competitor trade-offs.

## Bulletin - [00:00](https://www.youtube.com/watch?v=D-pz8L1orVI&t=0s) Anthropic introduced Claude Duet as a dedicated personal agent research preview for Claude Pro and Max users, designed to run proactive background checks across connected accounts without manual prompt triggers. - [04:46](https://www.youtube.com/watch?v=D-pz8L1orVI&t=286s) Duet runs automated background checks every three hours between 9:00 AM and 10:00 PM (five times daily) to parse schedules, unread emails, and pending projects into actionable status updates. - [06:23](https://www.youtube.com/watch?v=D-pz8L1orVI&t=383s) The agent can analyze past communication history—including 33 emails, 136 Slack messages, and 11 internal documents—to build a persistent writing persona for drafting emails and briefs. - [12:07](https://www.youtube.com/watch?v=D-pz8L1orVI&t=727s) A major operational limitation is Duet's cloud browser isolation: it cannot bypass verification or bot checks (such as Uber's human check) because users currently have no way to view or control the browser session.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=D-pz8L1orVI&t=0s) Claude Duet differentiates itself from rival agents by being proactive by design rather than waiting passively for user prompts. - [00:48](https://www.youtube.com/watch?v=D-pz8L1orVI&t=48s) Duet automatically inherits existing Claude accounts, calendars, and email integrations without requiring separate re-authentication. - [00:48](https://www.youtube.com/watch?v=D-pz8L1orVI&t=48s) The interface organizes workflows into customizable channels and groups, splitting work across active tasks, scheduled routines, and resolved items. - [04:46](https://www.youtube.com/watch?v=D-pz8L1orVI&t=286s) The default background cadence of five check-ins per day functions like a chief of staff, though users may need to adjust the frequency to avoid notification fatigue. - [06:23](https://www.youtube.com/watch?v=D-pz8L1orVI&t=383s) Built-in skills such as '/morning' render clean visual HTML artifacts summarizing daily schedules and priorities. - [06:23](https://www.youtube.com/watch?v=D-pz8L1orVI&t=383s) Duet analyzes historical correspondence to adopt user-specific communication rules, such as bottom-line-up-front phrasing and numbered action items. - [08:12](https://www.youtube.com/watch?v=D-pz8L1orVI&t=492s) The email triage console stages drafted responses side by side with original threads, allowing line-by-line user edits before messages are sent. - [09:03](https://www.youtube.com/watch?v=D-pz8L1orVI&t=543s) With just a channel URL, Duet can synthesize an entire year of public YouTube data into an interactive analytics dashboard featuring timeline charts and behavioral insights. - [10:13](https://www.youtube.com/watch?v=D-pz8L1orVI&t=613s) Duet builds operational business tools like sponsor revenue tracking dashboards, surfacing financial summaries and contract renewal warnings from unstructured data. - [11:20](https://www.youtube.com/watch?v=D-pz8L1orVI&t=680s) Coupled with animation-capable Claude models, Duet can generate interactive educational software like the 'AI Explorer Club' for kids. - [12:07](https://www.youtube.com/watch?v=D-pz8L1orVI&t=727s) Real-world web actions like booking rides stall when third-party bot detection triggers, because Duet lacks a live screen view or manual takeover mechanism. - [12:07](https://www.youtube.com/watch?v=D-pz8L1orVI&t=727s) Duet currently cannot execute custom agent skills stored locally on a user's physical machine. - [13:54](https://www.youtube.com/watch?v=D-pz8L1orVI&t=834s) Grok Bot leads in real-time market research via native X integration, but suffers from spotty email handling. - [13:54](https://www.youtube.com/watch?v=D-pz8L1orVI&t=834s) Muse features outbound phone-calling capabilities to businesses, but relies on the Muse Wonder model rather than top-tier frontier reasoning models. - [13:54](https://www.youtube.com/watch?v=D-pz8L1orVI&t=834s) ChatGPT Dots benefits from shared ecosystem memory but is hindered by a slow, buggy rollout. - [13:54](https://www.youtube.com/watch?v=D-pz8L1orVI&t=834s) Claude Duet stands out as the most proactive agent out of the box, with its long-term potential tied to solving cloud browser handoffs and integrating directly into core Claude apps.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=D-pz8L1orVI&t=0s) Introducing Claude Duet

Anthropic has officially entered the personal agent race with Claude Duet. Having spent time with early access, Peter Yang highlights that it carries the familiar, warm Claude aesthetic while pushing further into autonomy than anything else on the market. Rather than waiting passively for a prompt, Duet acts as an omnipresent work partner. Peter plans to put it through its paces across several real-world workflows—generating morning briefings, orchestrating email triage, tracking YouTube analytics, managing sponsor dashboards, and even running a learning club for his kids—while also addressing its current limitations, such as failing to book an Uber, and comparing it against rivals like Grok Bot, Muse, and ChatGPT Dots.

As Peter Yang puts it right away: “Anthropic says that Duet is proactive by design. And that's actually pretty true. Of all the personal agents that I've tried, Duet tries the hardest to be proactive.”

[00:48](https://www.youtube.com/watch?v=D-pz8L1orVI&t=48s) Interface Architecture: Tasks, Channels, and Groups

Duet lives across a dedicated mobile app and a refreshed web workspace, framed by Anthropic as a 24/7 assistant built to help people get actual work done. Onboarding is practically friction-free because it links directly into an existing Claude account, bypassing the tedious step of reconnecting external services like email and calendars from scratch. Upon first launching the app, Duet immediately scanned Peter’s inbox and schedule—recognizing that he runs his business *Behind the Craft* like a product, juggles sponsor calls and podcast tapings, and handles customer support largely solo with AI—and instantly proposed three starting tasks: prepping for an upcoming sponsor call with Diana, compiling an interview guide for a podcast guest, and pruning calendar meetup entries.

```mermaid flowchart TD accTitle: Claude Duet Workspace Structure accDescr: Diagram illustrating how Claude Duet organizes channels, main threads, and sub-tasks into groups. subgraph Navigation ["Sidebar Organization"] G["Custom Groups"] C["Channels (Custom Icons & Colors)"] G --> C end subgraph Execution ["Task Management Engine"] M["Main Claude Thread"] T_Active["Active Tasks"] T_Sched["Scheduled Tasks (Routines)"] T_Res["Resolved Tasks"] M --> T_Active M --> T_Sched M --> T_Res end C --> M ```

Clicking into any recommendation spawns a dedicated task thread, creating a project-like hierarchy where the primary chat anchors high-level discussions while sub-chats manage individual initiatives. The interface neatly partitions active tasks, scheduled recurring jobs, and resolved history. On the left navigation bar, users can organize work into custom channels and drag-and-drop groups, personalizing icons and color accents. While Peter notes this is less playful than the dynamic animated avatars found in competitors like Muse, it aligns with Anthropic's productivity-first philosophy:

- **Unified Account Setup:** Inherits existing Claude integrations with email, calendar, and third-party tooling without re-authentication. - **Task Hierarchy:** Breaks projects into granular sub-threads featuring Active Tasks, Scheduled Routines, and Resolved archives. - **Customizable Channels:** Allows users to pick specific icons (such as home or pencil) and custom color themes organized into drag-and-drop groups. - **Standalone Launch Strategy:** Released as an isolated research preview, mirroring Anthropic’s earlier deployment of Co-work before eventual core integration.

Evaluating the design decisions, Peter Yang notes: “Now, I think it's an interesting choice that Anthropic decided to make Duet a completely separate app and experience from Claude, but I think this is pretty typical of Anthropic's playbook.”

[04:46](https://www.youtube.com/watch?v=D-pz8L1orVI&t=286s) Proactive Check-Ins and Daily Cadence

The driving force behind Duet’s proactive behavior is an automated, out-of-the-box check-in engine. Looking into the execution history, Peter discovers that the agent runs background audits every three hours between 9:00 AM and 10:00 PM. During each sweep, it parses incoming communications and calendar events to independently surface priorities, alert the user to upcoming milestones, and identify deliverables requiring attention.

In one morning check-in example, Duet proactively flagged an open sponsorship arrangement with Diana, reminded Peter to launch an upcoming Mercury payroll video, and queued preparation work for a Friday podcast taping with an Anthropic designer named Nate. While the experience closely mirrors having an attentive chief of staff, Peter finds the default frequency slightly overbearing for independent creators and suggests dialing back the cadence through a simple conversational prompt:

- **Check-In Frequency:** Automatically triggers every three hours between 9:00 AM and 10:00 PM, totaling five automated touchpoints per day. - **Context Synthesis:** Actively correlates unread emails, scheduling changes, and pending drafts into centralized status updates. - **Configurable Cadence:** Users can converse with Duet to restrict updates to preferred intervals, such as a morning brief and an afternoon recap.

Reflecting on the sheer intensity of the default background schedule, Peter Yang remarks: “So basically, you know, just like a chief of staff, it brings things to me, but honestly, I think five check-ins a day is a little bit too much.”

[05:51](https://www.youtube.com/watch?v=D-pz8L1orVI&t=351s) Sponsor: Wispr Flow

Peter takes a quick moment to spotlight Wispr Flow, a dedicated voice dictation tool he has integrated into his day-to-day workflow for over a year across Claude, ChatGPT, and other apps. Dictation has dramatically cut down his typing overhead, and his children regularly rely on it to build their own software and games without needing keyboard skills. Recent updates to its voice models have noticeably boosted transcription accuracy, making hands-free AI interaction significantly smoother.

Regarding how voice changes computer usage for younger creators, Peter Yang shares: “My kids also love using it to build games and apps with AI. Even though they don't know how to type, they just voice dictate.”

[06:23](https://www.youtube.com/watch?v=D-pz8L1orVI&t=383s) Slash Skills, Morning Briefs, and Writing Style Mimicry

Duet hides a substantial portion of its utility behind slash commands, which reveal an array of prebuilt system skills. Triggering the `/morning` skill prompts the agent to render a clean, styled HTML morning brief artifact. The generated document reflects Claude’s characteristic design restraint, summarizing Peter's calendar by noting two short morning calls before 9:00 AM, declaring the remainder of the day completely clear, and playfully embedding a sun graphic to nudge him outside.

To make Duet's communications genuinely useful rather than generic, Peter ran a skill called "Learn how I write." The agent ingested his recent sent messages and documentation to reverse-engineer his communication patterns, storing the output as an actionable persona profile that informs all future email and document drafting:

- **Dataset Analyzed:** 33 sent emails, 89 Slack channel messages, 47 direct messages, and 11 authored internal documents. - **Synthesized Voice Guide:** - Lead immediately with the core conclusion and cut unnecessary conversational preambles. - Automatically convert multi-part requests and action items into numbered lists. - Adopt casual, lowercase phrasing for internal team banter while maintaining structured professionalism with external partners.

Appreciating the thoughtful touches inside the visual morning artifact, Peter Yang observes: “And Duet even added the sun here to probably encourage me to touch grass after my morning meetings are over.”

[08:12](https://www.youtube.com/watch?v=D-pz8L1orVI&t=492s) Email Triage and Execution Artifacts

Leveraging the synthesized writing profile, Duet builds interactive operational artifacts, notably a daily to-do agenda and a dedicated email triage console. The triage artifact lists pending emails alongside recommended actions and auto-generated reply drafts. Peter can scan incoming requests, assign high-level instructions—such as declining an invitation from Freeman or confirming candidate recommendations from Roblox for Nail—and let Duet polish the responses in his authentic tone.

The system leaves full control with the user by presenting every reply in an editable staging view before anything leaves the outbox. Low-risk administrative notes can be cleared with automated sending, while high-stakes partner correspondence can be scrutinized line by line:

- **Triaged Processing:** Classifies correspondence by urgency, offering one-click confirmations for scheduling declines or basic administrative acknowledgments. - **Drafting Sandbox:** Presents drafts side-by-side with original threads, allowing direct in-line copy edits prior to final delivery. - **Safety Boundaries:** Distinguishes routine tasks suited for automated sending from critical sponsor communications requiring human sign-off.

Emphasizing the balance between automation and human oversight, Peter Yang points out: “Now, important emails I still want to review myself. So down here, it actually drafted all the replies and I can make edits to it before I hit send.”

[09:03](https://www.youtube.com/watch?v=D-pz8L1orVI&t=543s) YouTube Analytics Engine

To evaluate how Duet handles deeper analytical tasks, Peter instructed it to build a comprehensive analytics artifact for his YouTube channel simply by providing the channel link. Duet retrieved a full year of public upload data and converted it into an interactive dashboard, surfacing macro subscriber metrics alongside granular behavioral trends.

The generated report highlighted that deep dives into personal AI agents serve as his primary breakout topic, while solo videos consistently outpace interview episodes unless the guest delivers a practical live demo. The dashboard also incorporated interactive timeline charts mapping performance against upload dates, categorized top-performing search keywords like model comparisons, and broke down audience retention across different runtime lengths.

Reflecting on the low-friction setup required to generate the analytics suite, Peter Yang concludes: “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.”

[10:13](https://www.youtube.com/watch?v=D-pz8L1orVI&t=613s) Building a Sponsor Revenue Dashboard and an AI Explorer Club

Without relying on any complicated prompting, Claude Duet is able to pull scattered business data together and construct a clean, functional sponsor revenue dashboard. Peter Yang clarifies that while the numbers shown are mock data rather than his real financials, sponsorship serves as a major revenue driver for his creator business. The generated dashboard summarizes booked revenue for 2026, categorizes income by sponsor, flags outreach requiring follow-ups, highlights pending sponsor conversations waiting on his response, and tallies contracted deals. It even surfaces automated business insights—noting, for instance, that substantial contract revenue is scheduled to end in Q4, that very few sponsors have been locked in for the following year, and that three specific sponsors account for the vast majority of projected 2026 revenue.

Peter points out that whether someone runs an agency, a creator brand, or a subscription business, having an agent build custom dashboards is a game-changer: “You know you might have a different type of business. You might be making revenue from subscriptions or something else, but being able to build these dashboards is just extremely useful.” Moving forward, Duet can easily be instructed to schedule recurring weekly updates or dispatch snapshot emails directly to a phone. Shifting from business to family, Peter showcases an interactive artifact he built alongside his daughter: an “AI Explorer Club” gamified learning hub designed to teach kids AI fundamentals. Leveraging Anthropic models like Claude Opus and Claude Fable—which excel at animations and motion design—the artifact features a friendly robot companion and animated typography. Tapping “Start the Adventure” opens a visual path where the first lesson explains chatbots and provides a prompt kids can copy and paste to explore large language models. Although Peter notes the current iteration feels slightly too text-heavy, he explains that a simple follow-up prompt can quickly inject richer visuals and interactive flair.

[12:07](https://www.youtube.com/watch?v=D-pz8L1orVI&t=727s) Where Claude Duet Hits a Wall: Browser Hurdles and Skill Limits

Despite Duet's impressive artifact generation, Peter candidly addresses the product's friction points during its current research preview. The biggest operational limitation surfaces when testing out pre-built real-world actions, such as the default Uber ride and DoorDash skills. When Peter prompted Duet to summon an Uber ride to the Palace of Fine Arts, the agent asked the right logistical questions—confirming pickup location, destination, and immediacy—but ultimately stalled out. Uber's web portal triggered a “verify you're human” verification check, and Duet was unable to bypass or solve it autonomously.

```mermaid flowchart TD accTitle: Duet Cloud Browser Bottleneck accDescr: Diagram illustrating how Claude Duet gets blocked by verification checks without user intervention. subgraph DuetAgent["Duet Cloud Agent"] A["Trigger Skill: Book Uber Ride"] --> B["Prompt Clarifications: Time & Destination"] B --> C["Open Cloud Browser & Navigate Service"] end subgraph Bottleneck["Unresolved Bottleneck"] C --> D{"Bot Detection: CAPTCHA / Human Check"} D -->|"Agent Blocked"| E["Task Fails in Cloud"] D -.->|"Missing: User Takeover / Screen View"| F["User Unblocks Session"] end ```

Unlike competing personal agents that allow users to open up the underlying browser window to complete CAPTCHAs, sign in, or resolve challenges manually, Duet offers zero live browser visibility or interactive screen control. Peter emphasizes the severity of this gap: “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. But I think being able to interact with the browser is absolutely critical for personal agents that live on a cloud computer so that you can unblock the AI for things that still require a human to do like capture checks or signing.” Beyond the browser disconnect, Duet's current skill library also fails to index or access custom local skills stored on the user's physical computer, creating another boundary Anthropic will need to bridge when migrating the tool into their native desktop application.

- **Cloud Browser Isolation:** Users cannot view or take mouse/keyboard control of Duet's remote cloud browser to resolve security checks or two-factor logins. - **Local Skill Gap:** Duet cannot currently pull in or execute custom agent skills stored locally on a user's machine. - **Status:** Anthropic's product team is actively developing an interactive browser view ahead of the tool's public release.

[13:54](https://www.youtube.com/watch?v=D-pz8L1orVI&t=834s) Head-to-Head: Claude Duet vs. Grok Bot, Muse, and ChatGPT Dots

To wrap up the analysis, Peter stacks Claude Duet against the other major personal AI agents on the market: Grok Bot, Muse, and ChatGPT Dots. Grok Bot remains Peter’s day-to-day favorite, largely because its tight, native integration with the X platform makes it unmatched for live market research, though its internal email handling remains buggy and it requires a paid tier. Muse stands out for sheer product design and packs an unmatched real-world capability: placing live outbound phone calls to US businesses to handle tedious chores like negotiating cable bills or contacting tree removal services. However, Muse relies on the Muse Wonder model rather than top-tier frontier models like Opus or Astra, occasionally leading to lower reasoning quality. ChatGPT Dots benefits from living directly inside the ChatGPT ecosystem with shared context, threads, and memory, yet its buggy debut and sluggish performance make it feel less efficient than running a dedicated, manual Chief of Staff thread.

Claude Duet distinguishes itself as the most inherently proactive agent out of the box, with Peter noting, “And you know, last but not least we have Claude Duet which in my opinion is the most proactive out of the box.” Because Claude already serves as Peter's primary model and operating harness across his daily workflow, Duet's ability to capitalize on his existing context gives it immense long-term potential once Anthropic fixes the cloud browser handover and exits research preview. Currently, Duet is rolling out via a research preview waitlist for Claude Pro and Max subscribers, with Peter predicting that Anthropic will ultimately fold Duet's capabilities directly into the core Claude web, mobile, and desktop apps.

## Notable quotes > [12:07](https://www.youtube.com/watch?v=D-pz8L1orVI&t=727s) I think being able to interact with the browser is absolutely critical for personal agents that live on a cloud computer so that you can unblock the AI for things that still require a human to do like capture checks or signing. — Peter Yang

> [00:00](https://www.youtube.com/watch?v=D-pz8L1orVI&t=0s) Anthropic says that Duet is proactive by design. And that's actually pretty true. Of all the personal agents that I've tried, Duet tries the hardest to be proactive. — Peter Yang

> [00:48](https://www.youtube.com/watch?v=D-pz8L1orVI&t=48s) I think it's an interesting choice that Anthropic decided to make Duet a completely separate app and experience from Claude, but I think this is pretty typical of Anthropic's playbook. — Peter Yang

## Mentioned Claude Duet, Peter Yang, Anthropic, Wispr Flow, Grok Bot, Muse, ChatGPT Dots, Behind the Craft, Claude, ChatGPT, Uber, DoorDash, YouTube, Slack, Mercury, Claude Opus, Claude Fable, Muse Wonder, Astra, Roblox, X

## Video description 🎙️ Get one month of Wispr Flow free with my link: https://wisprflow.ai/?promo_code=PETERWISPRFLOW&months=1&utm_source=peteryang&utm_medium=sponsorship&utm_campaign=btc-q2-2026 Copy my best AI workflows to save time and automate busywork: https://www.behindthecraft.com 🤝 Subscribe to my practical AI newsletter read by 150K+ professionals: https://creatoreconomy.so

Claude Duet is Anthropic's new personal agent, and it's the most proactive one I've tried. In this video, I show you how Duet works, how its check-ins find work for you, and the artifacts I had it build: a morning brief, email triage, a YouTube report, a sponsor dashboard, and an AI Explorer Club for my kid. I also share where it still falls short and how it stacks up against Grok Bot, Muse, and ChatGPT Dots.

What I cover: (00:00) Why Claude Duet is the most proactive personal agent I've tried (00:48) First look at Duet: tasks, channels, and groups (04:46) How Duet's proactive check-ins work (06:23) Build a morning brief, to-do list, and email triage with Duet (09:03) Build a YouTube report and sponsor dashboard with Duet (11:20) Build an AI Explorer Club for kids with Duet (12:07) Where Duet hits a wall: Uber, the cloud browser, and skills (13:54) Claude Duet vs Grok Bot, Muse, and ChatGPT Dots

This video is sponsored by Wispr Flow, but opinions are my own.

Where to find me: X: https://x.com/petergyang LinkedIn: https://www.linkedin.com/in/petergyang/

Subscribe to this channel - more tutorials coming soon!

[Source](https://www.youtube.com/watch?v=D-pz8L1orVI)

中文简报

Claude Duet 正式登场！你想知道的一切都在这里 Peter Yang 体验了 Claude Duet 的抢先体验版，全面考察了其开箱即用的架构、自主后台扫描以及专属任务线程。他展示了 Duet 如何通过分析他的写作风格与日程安排，生成个性化的日常产出——例如晨报、邮件分流看板、YouTube 分析引擎、赞助商收入仪表板以及互动学习工具。在称赞其出色的主动性与精良设计的同时，Peter 也指出了若干主要难点，包括无法通过浏览器人机验证或执行本地设备操作技能。最终，他将 Duet 与 Grok Bot、Muse 和 ChatGPT Dots 进行了对比，并得出结论：一旦云端浏览器交互问题得以解决，Duet 作为一体化“幕僚长”将具有巨大的发展潜力。 Anthropic 面向 Claude Pro 和 Max 用户推出了 Claude Duet 专属个人智能体研究预览版，旨在无需人工提示词触发的情况下，跨连接账户主动运行后台检查。 (0:00) Duet 在上午 9:00 至晚上 10:00 之间每三小时运行一次自动化后台检查（每天五次），用于解析日程、未读邮件和待办项目并转化为可执行的状态更新。 (4:46) 该智能体能够分析过往的沟通记录——包括 33 封电子邮件、136 条 Slack 消息和 11 份内部文档——从而建立持久的写作风格画像，用于起草邮件和简报。 (6:23) 一个主要的运行局限在于 Duet 的云端浏览器隔离机制：由于用户目前无法查看或控制浏览器会话，它无法绕过验证或机器人检查（例如 Uber 的人机验证）。 (12:07)

Reading summary

Anthropic has entered the personal agent race with Claude Duet, an autonomous work partner designed to proactively manage daily tasks, communications, and business analytics. Based on early access testing, Peter Yang reports that while Duet leads the market in out-of-the-box proactivity and interactive artifact generation, its current inability to let users intervene in cloud browser sessions presents a significant operational hurdle. Workspace Structure and Proactive Cadence (0:48) Claude Duet operates across a dedicated mobile app and web workspace, linking directly to existing Claude accounts without requiring users to re-authenticate email or calendar tools. The interface organizes work through custom channels, drag-and-drop groups, and a task management engine that segments initiatives into active tasks, scheduled routines, and resolved archives. Upon initial setup, Duet automatically analyzed Yang's schedule and unread emails to propose starting tasks, such as prepping for sponsor calls and pruning calendar entries. The system's core mechanism is an automated background audit that runs every three hours between 9:00 AM and 10:00 PM, generating up to five check-ins daily. During each pass, Duet correlates unread messages, scheduling shifts, and pending deliverables into centralized updates. While acknowledging this acts like an attentive chief of staff, Yang noted that five daily pings can feel overbearing and recommended that users converse with Duet to reduce the frequency to a morning brief and an afternoon recap. Voice Modeling and Communications Triage (6:23) Duet relies heavily on slash commands, such as using `/morning` to generate visual HTML morning agendas. To avoid generic AI messaging, Duet features a 'Learn how I write' skill. In Yang's test, the system ingested 33 sent emails, 89 Slack channel messages, 47 direct messages, and 11 internal documents to compile a behavioral voice profile. The resulting guide directed Duet to lead with conclusions, eliminate preambles, format multi-part requests into numbered lists, and adopt lowercase phrasing for casual internal banter alongside formal external communications. Duet applies this profile through interactive triage artifacts, staging auto-generated email replies alongside incoming threads. Users can review drafts side by side with the original context, assign high-level handling instructions, or apply inline edits before sending. This sandbox creates a safety boundary, allowing routine administrative declines to clear with one-click approvals while reserving high-stakes correspondence for line-by-line human review. Operational Analytics and Custom Artifacts (9:03) Duet can synthesize unstructured business data into interactive, visual dashboards with minimal prompting. Provided only with a public channel link, Duet extracted a year of YouTube data into an analytics dashboard that mapped audience retention curves, categorized high-traffic search terms, and identified that solo personal-agent breakdowns consistently outpaced interview episodes unless guests ran live demonstrations. Similarly, Duet aggregated mock business data into a 2026 sponsor revenue dashboard. The artifact categorized income by partner, tallied contracted deals, and produced automated business insights—flagging that three sponsors accounted for most projected revenue and warning that major contracts were expiring in Q4. Beyond commercial operations, Yang tested Duet with Anthropic's Claude Opus and Fable models to build the 'AI Explorer Club,' an educational interface featuring animated typography and interactive prompts to teach children AI fundamentals. Cloud Browser Bottlenecks and Market Trade-offs (12:07) Duet's most prominent real-world failure occurs during browser-based actions, such as executing ride-hailing or food delivery skills. When prompted to book an Uber, the agent resolved the logistical details but completely stalled out when Uber's web portal triggered a human-verification CAPTCHA check. Because Duet operates inside an isolated cloud browser without interactive screen control or user takeover visibility, users cannot intervene to solve bot detections or two-factor prompts. Additionally, Duet currently cannot access custom skills stored locally on a user's machine, though Yang noted Anthropic product managers are actively developing an interactive browser view. Evaluating the broader agent ecosystem, Yang observed distinct trade-offs across providers. Grok Bot excels at live market research through native X platform data but requires a paid subscription and suffers from buggy email handling. Muse provides a free, polished interface capable of placing real-world outbound phone calls to businesses, yet it relies on the weaker Muse Wonder reasoning model. ChatGPT Dots integrates tightly with existing OpenAI memory but launched in a slow, buggy state. In contrast, Duet stands out for its out-of-the-box proactivity and workspace artifacts, even as browser bottlenecks remain unresolved. Currently rolling out as a research preview waitlist for Claude Pro and Max subscribers, Claude Duet demonstrates the strength of proactive, artifact-driven task management. However, until Anthropic bridges the cloud-to-human browser gap for CAPTCHAs and security checks, its capacity to independently execute real-world web transactions remains constrained.

精炼阅读摘要

根据 Peter Yang 的评测，Anthropic 推出的个人智能体 Claude Duet 凭借全天候主动巡检与精细的 Artifacts 界面架构，在自主协同办公领域展现出强劲的主动性，但由于缺乏远程浏览器交互控制以及无法调用本地技能，其执行现实任务时仍存在明显短板。界面架构与开箱即用的主动协同 (0:48) Claude Duet 采用独立的移动端应用与网页端工作区，定位于全天候数字助手。系统无需用户重复配置邮箱和日历授权，能直接继承原有的 Claude 账户生态。初次启动时，Duet 即通过扫描日程与收件箱识别出 Peter Yang 的业务运作模式，并自动提炼出赞助商会议筹备、播客嘉宾提纲整理等三项初始化任务。系统界面通过自定义频道、拖拽分组以及包含主会话与子任务的任务层级结构来归拢复杂事项，清晰区分进行中任务、定期例行事务与已完成归档，呈现出偏向严谨生产力工具的架构逻辑。后台巡检机制与个性化文风拟合 (4:46) Duet 的主动性核心源自自动化巡检引擎。该引擎在上午 9:00 至晚上 10:00 期间每三小时执行一次后台扫描，全天共触发五次，主动抓取未读信息与日程变更并生成汇总汇报。尽管该机制能够及时预警赞助合作进度或录制准备，但 Peter Yang 认为默认一天五次的提醒频次对独立创作者而言略显频繁，建议用户通过自然语言将其调整为早晚各一次等更舒缓的节奏。此外，通过执行内置的文风学习技能（Learn how I write），Duet 深度分析了 33 封已发邮件、89 条 Slack 频道发言、47 条私信及 11 份内部文档，提炼出直奔核心结论、多项待办转数字列表以及对外专业严谨、对内随性简洁的写作规范，以此驱动后续的文书与沟通拟稿。邮件分流与跨业务看板的交互式构建 (8:12) 在日常业务处理中，Duet 结合个性化文风生成可交互的邮件分流控制台，将待办邮件按紧急程度归类并附带推荐操作与预撰写草稿，同时为高风险沟通保留人工二次审核与直接编辑通道。面对数据分析诉求，用户仅需提供频道链接，Duet 即可调取 YouTube 一整年的公开视频数据，自动生成覆盖关键搜索词分布、观看留存率曲线与视频形态对比的分析看板。在商业运营上，它还能基于模拟数据搭建 2026 年度赞助营收仪表盘，直观预警第四季度合约到期与主要赞助商集中度风险；而在教育场景中，依托 Claude Opus 和 Claude Fable 模型，Duet 也能快速搭建出带有动态效果的儿童 AI 学习互动应用。执行瓶颈与主流个人智能体横向对比 (12:07) 在涉及现实外部操作时，Duet 暴露出核心瓶颈。在测试调用 Uber 叫车技能时，即便确认了出发地与目的地，系统在遇到 Uber 网页端的人机验证（CAPTCHA）时彻底卡死。由于 Duet 运行在云端且不提供实时屏幕回传或用户手动接管浏览器的窗口，导致外部阻塞无法被人工解除；同时，Duet 目前也无法调用存储在用户本地计算机上的私有技能。横向对比其他主流竞品：Grok Bot 依托 X 平台的原生整合在实时市场调研上表现优异，但邮件处理存在缺陷且需要付费；Muse 界面精美且支持拨打美国本地真实商户电话，但受限于底层 Muse Wonder 模型的推理能力；ChatGPT Dots 与主模型记忆体系深度绑定，但初期版本运行迟缓且漏洞偏多。Duet 虽在主动性与交互生成物上优势显著，但仍需等待 Anthropic 补齐浏览器接管等关键能力。 Peter Yang 认为，Claude Duet 是目前开箱即用状态下主动性最强的个人智能体，但云端浏览器不可见及本地技能隔离限制了其执行现实复杂任务的上限。据产品团队透露，交互式浏览器视图已在开发中；目前该产品正面向 Claude Pro 和 Max 订阅用户开启研究预览版候补，预计未来将被全面整合进 Claude 的核心跨端体验中。

Source quotes

[12:07] Peter Yang: I think being able to interact with the browser is absolutely critical for personal agents that live on a cloud computer so that you can unblock the AI for things that still require a human to do like capture checks or signing. [0:00] Peter Yang: Anthropic says that Duet is proactive by design. And that's actually pretty true. Of all the personal agents that I've tried, Duet tries the hardest to be proactive. [0:48] Peter Yang: I think it's an interesting choice that Anthropic decided to make Duet a completely separate app and experience from Claude, but I think this is pretty typical of Anthropic's playbook.

Original transcript

[0:00] Claw Duet, Anthropic's very own personal [0:02] agent, is finally here. I had early [0:05] access and is one of the most proactive [0:07] agents that I've tried with that [0:09] distinctive claw look that we all know [0:11] and love. In this video, I'm going to [0:13] show you exactly how Duet works and walk [0:16] through how I use it to create a morning [0:18] brief, a YouTube report, a sponsor [0:21] dashboard, and even an AI explorer club [0:25] for my kids to learn about AI. I'll also [0:28] be honest about where Duet falls short, [0:30] like what happened when I tried to get [0:32] it to book an Uber for me. And at the [0:35] end, I'll share my updated take on how [0:38] Claw Duet stacks up against the [0:40] competition, including Grockbot, Muse, [0:43] and Chat Dots. All right, let me show [0:46] you how Duet works. Okay, so Duet has [0:50] its own dedicated mobile app and its own [0:53] interface on cloud web. Anthropic is [0:56] positioning it as a 247 assistant that [0:59] helps you get work done. And the best [1:01] part about the onboarding is that it's [1:03] already connected to my cloud account. [1:05] So I don't have to connect my email, [1:07] calendar, and other apps all over again. [1:10] And Anthropic says that Duet is [1:12] proactive by design. And that's actually [1:14] pretty true. Of all the personal agents [1:16] that I've tried, Duet tries the hardest [1:19] to be proactive. And I'll tell you why [1:21] later in this video. But let me show you [1:23] my first conversation with Duet here. [1:26] Right off the bat, it read my email, [1:28] calendar, and connected apps to figure [1:30] out how I work. I run behind the craft [1:33] like a product. My calendar is full of [1:36] podcast recordings and sponsor calls and [1:39] I handle my newsletter, customer support [1:42] by myself with the help of AI agents. [1:45] Okay. Then duet suggested a few place [1:47] where he can start helping including [1:50] preparing for a sponsor call with Diana, [1:53] prepping a podcast interview guide with [1:55] my friend, and cleaning up some [1:58] meetup entries. [2:00] Now, when I click on the interview [2:01] guide, Duet will spin up a new task. [2:03] Let's take a look at the task folder [2:05] right now. So, a task is a thread inside [2:10] the main cloth thread that you see here. [2:13] And here's a bunch of tasks that I have [2:15] going on. Prepare Tuesday call with [2:17] Diana for sponsors. Plan tutorial about [2:20] Claude and find showtimes for you can [2:23] see everything the movie and so on and [2:25] so forth. And you can see here that here [2:28] are my active tasks. There's a bunch of [2:30] tasks coming up next. There's a bunch of [2:33] schedule tasks which are basically like [2:35] schedule routines and schedule jobs that [2:37] run every day or every week. And then [2:39] there's also a resolve task folder. So [2:43] in many ways this works a lot like cloud [2:44] projects. You have a main chat here with [2:47] sub chats for each task on the right. [2:50] Now let's take a look at the left nav. [2:53] So unlike Muse and Grockbot, you can't [2:56] customize the main thread and what the [2:59] avatar here is. It's basically just a [3:01] dot that animates when you get to do [3:03] stuff. I guess anthropic really want to [3:05] keep that distinctive claw look. But [3:07] what you can do is you can create [3:09] individual channels. So for example here [3:11] I can make a new channel and I just call [3:13] it test channel. Hit create and in the [3:18] channel I can actually customize the [3:21] icon of the channel. So in this screen I [3:23] can customize the icon from a pencil for [3:26] example to let's say home and also the [3:29] color of the icon. Now, in my opinion, [3:32] this isn't as interesting or engaging as [3:35] customizing your own bot or like Muse, [3:37] making your own animated avatars, but I [3:40] think Duet is very much designed for [3:42] work use cases. So, that's why I think [3:44] they went with these icon styles. And [3:47] after you create a channel, it's going [3:49] to suggest a few tasks that it can do [3:50] here. So, let's see what it comes up [3:51] with. Okay. And it read my email and [3:54] calendar and came up with three [3:55] different tasks. draft interview guide [3:57] for Nate who is a anthropic designer. [4:00] Write a post for the duet launch and [4:04] pull together a short prep guide for [4:06] inerson meetup with a friend of mine. So [4:09] you can see how it's always trying to do [4:10] work for you and is always trying to be [4:12] proactive. And over here you can also [4:16] create a new group. Let's call it test [4:19] group. And I can drag and drop my [4:21] channels into different groups here. So [4:23] pretty standard stuff. Now, I think it's [4:25] an interesting choice that Anthropic [4:27] decided to make Duet a completely [4:29] separate app and experience from Claude, [4:32] but I think this is pretty typical of [4:34] Anthropic's playbook. Launch a research [4:36] preview, iterate and prove that it [4:38] works, and then probably in the future [4:40] merge it back into the main cloud app, [4:43] just like what they did with co-work. [4:46] Okay, so I mentioned how Duet is the [4:49] most proactive assistant that I've tried [4:51] and the reason is actually pretty [4:53] simple. It has a schedule check-in task [4:55] that comes by default. And if I go here [4:58] and say view history, you can see that [5:00] this task basically runs every 3 hours [5:03] from 9:00 a.m. to 10 p.m. And I'm sure [5:06] it checks my email calendar and other [5:08] connected apps to find work that it can [5:11] do for me. So, here's an example. It [5:13] says, "Morning to you, Peter. Here's a [5:15] sponsorship deal with Diana that you're [5:17] working on. And there's also a Mercury [5:20] payroll video that you should launch. [5:22] and you got to prepare the Friday [5:23] podcast recording with Nate. So [5:25] basically, you know, just like a chief [5:27] of staff, it brings things to me, but [5:29] honestly, I think five check-ins a day [5:31] is a little bit too much. I prefer maybe [5:33] once in the morning and once in the [5:35] afternoon, but you can also just ask [5:37] Claude and do it to change the schedule. [5:40] All right, so that's a quick tour of the [5:41] main screens and the proactive checking [5:44] that do it has. But I think what really [5:46] makes this product stand out is the [5:48] artifacts that it can create. Okay, but [5:51] first I want to give a quick shout out [5:53] to the sponsor of this video, Whisper [5:55] Flow. You know, I've been using Whisper [5:57] Flow for over a year to dictate in [5:59] Claude, Chatbt, and many other apps, and [6:02] it's been a massive timesaver for me. My [6:05] kids also love using it to build games [6:07] and apps with AI. Even though they don't [6:09] know how to type, they just voice [6:10] dictate. And with the recent update of [6:13] Whisper Flow's voice dictation model, [6:14] the dictation is just way more accurate. [6:17] So, go try out Whisper Flow. You can [6:19] check out the link in my description to [6:21] get 1 month's free. Awesome. Now, let's [6:24] come back to Duet and talk about the [6:27] artifacts that I had in make. So, first [6:29] of all, a hidden feature in Duet is that [6:31] if you type slash, you can see all kinds [6:34] of skills that you can run inside this [6:36] app. And down here, there is a morning [6:39] skill which is basically renders the [6:41] user's morning brief as a styled HTML [6:44] artifact. So, that's basically what I [6:46] did. I said, "Morning, let's try out the [6:48] skill." And then it rendered a morning [6:49] brief. Let's take a look. And here we [6:52] have it. As you can see, it also has [6:54] that clawed aesthetic. I love how it [6:57] says that I have two short calls before [6:59] 9:00 and then the day is wide open. You [7:01] know, now that I'm independent, I try to [7:03] avoid meetings as much as possible. And [7:05] Duet even added the sun here to probably [7:08] encourage me to touch grass after my [7:10] morning meetings are over. If you have a [7:13] job, you have a busy schedule, you [7:14] probably have a lot more meetings than [7:16] me. And down below it has all the stuff [7:18] that I should address this morning. [7:20] Starting for example with the sponsor [7:23] contract with Mercury, one of my [7:25] partners. Okay. So another skill that I [7:28] ran right after morning brief is learn [7:31] how I write. You know, if you're going [7:33] to have Duet reply to emails for you, [7:35] draft documents for you, it's important [7:37] for it to learn your style. So, what it [7:39] did was it read 33 emails that I sent, [7:42] 89 Slack messages that I posted, and 47 [7:45] DMs, 11 docs that I wrote, and it came [7:47] up with a writing guide for my style. [7:50] And the writing guide has you open with [7:52] the point and skip the warm-up. You turn [7:54] any set of asks into a number list, you [7:57] like to use lowerase of your team, and [7:59] polish writing with your partners. So, [8:01] basically, it went off and it saved my [8:03] writing style. And now we can use this [8:05] writing style skill to help us draft [8:09] emails and more. Okay, so next I had [8:12] duet build two more artifacts for me [8:14] which also come by default. So there's a [8:16] daily to-do list which is basically [8:18] calls I have with folks during the day [8:20] and also stuff that I need to get done [8:22] in a day. And there's also email triage [8:25] artifact that basically prioritizes the [8:27] email that I should respond to. And as [8:30] you can see on the left with this [8:32] artifact, I can just say, "Hey, for [8:34] Freeman, let's decline. Can you send a [8:36] reply? For Nail, I recommended two [8:39] colleagues from Roblox. Can you reply [8:41] directly?" And a bunch of other [8:43] comments. So basically, with the writing [8:45] skill and this artifact, I can rely on [8:48] Claude to reply to emails for me. Now, [8:51] important emails I still want to review [8:52] myself. So down here, it actually [8:55] drafted all the replies and I can make [8:57] edits to it before I hit send. But some [9:00] emails maybe just get cloud to send for [9:02] you. Okay, I want to show you guys a [9:05] more sophisticated artifact that I had [9:06] to make next which is an Alex report on [9:10] my YouTube channel. It basically looked [9:12] at a year of my published videos and [9:14] their public data to come up with this [9:16] report. So we have subscribers and [9:19] number of videos published up here. And [9:21] what's more interesting is down here [9:24] we're sharing insights about my videos. [9:26] So for example, it found that personal [9:28] AI agent videos are my breakout topic [9:31] and actually my solo videos are doing [9:33] better than the guest unless the guest [9:35] has a practical demo or something to [9:37] show. If I scroll down even more, there [9:39] is some pretty amazing charts. So here [9:40] is a chart that shows every video by [9:43] publish date and how they performed. And [9:45] you can see here that my solo videos [9:47] started doing better this year compared [9:49] to my podcast videos. And if I scroll [9:52] down more, there is also keywords that [9:55] perform, personal AI agents, model [9:56] showdowns, claude videos perform, and [9:59] also how long each video is by length. [10:03] And by the way, to make this awesome [10:05] dashboard, all I said was like, hey, can [10:07] you make a dashboard and insights on my [10:09] YouTube performance data? And I linked [10:11] my YouTube channel. So I didn't give any [10:13] kind of fancy prompt. That was basically [10:15] it and it found all this data for me. [10:18] Now, real quick, let me show you another [10:19] very useful dashboard that it made for [10:22] me. And that is my sponsor revenue [10:25] dashboard. So, like any other creator, [10:27] sponsorship is a big revenue driver for [10:29] me. And by the way, this is not real [10:30] data. This is block data on my [10:32] sponsorships. But you know it has [10:34] information on how much revenue I booked [10:35] for 2026 revenue by sponsor [10:39] conversations that I need to follow up [10:40] on you know sponsorship conversations [10:42] that are waiting on me and stuff that's [10:44] contracted out and it has a bunch of [10:48] insights here as well. So for example [10:50] there's a lot of revenue ending in Q4 [10:52] but I haven't really booked a ton of [10:53] sponsors for next year and you know [10:56] these three sponsors are the majority of [10:58] revenue for 2026. [11:01] You know you might have a different type [11:02] of business. You might be making revenue [11:03] from subscriptions or something else, [11:05] but being able to build these dashboards [11:07] is just extremely useful. And of course, [11:09] I can also ask Duet to set up scheduled [11:12] tasks to keep these dashboards updated [11:15] every week or maybe email me these [11:17] dashboards so that I can take a look at [11:18] them on my phone. Awesome. So, another [11:22] artifact that I had duet built with my [11:24] kid is an AI explorer club for her to, [11:28] you know, pick up AI skills. And if you [11:30] know Opus, if you know Fable, you know [11:32] that these models from Anthropic are [11:34] extremely good at motion design, [11:36] animations, and such. So for this one, I [11:38] decided to make it fun. We have a little [11:39] robot buddy here and animated text. [11:42] Let's go ahead and click start the [11:43] adventure. And there is a adventure [11:46] path. And let's click one. And this is [11:48] the first lesson about what a chatbot [11:51] is. And there is a prompt here that I [11:53] can copy and paste to have my kid learn [11:56] all about LMS. Now, I think this is a [11:59] little bit too textheavy for my taste. I [12:01] wish it was more interesting, but you [12:03] know, we can easily prompt claw to make [12:04] it more visually engaging. All right, so [12:08] artifacts are awesome into it, but I [12:10] also want to be honest about the [12:12] shortcomings of this app. Now, note that [12:14] this app is still research preview, so [12:16] maybe some of the stuff will be failed. [12:18] But one shortcoming is, for example, I [12:20] asked it to book an Uber ride because [12:22] there's actually a default skill called [12:24] Uber ride in Duet. So if it's an Uber [12:28] ride, there's also skills to book Door [12:29] Dash and so on. So when I type in the [12:32] skill, it asked me, you know, where [12:34] should a car pick you up and is it right [12:36] now and where is it going? And I said [12:39] palace of fine arts and unfortunately it [12:41] was not able to do it because US browser [12:44] required verify your human check from [12:46] Uber and it was able to click through [12:49] this. So that required me to basically [12:52] like the other personal agents actually [12:54] open Duet's browser and navigate this [12:57] human check for it. But unfortunately at [12:59] least right now duet doesn't actually [13:02] let me see it browser or let me interact [13:04] with this browser. I couldn't find a [13:06] single place in this app where I can [13:08] actually engage. Now talking to the duet [13:10] PM I know that he's hard at work on this [13:12] feature. So I think it'll be coming [13:13] eminently. But I think being able to [13:15] interact with the browser is absolutely [13:17] critical for personal agents that live [13:19] on a cloud computer so that you can [13:21] unblock the AI for things that still [13:24] require a human to do like capture [13:26] checks or signing. Another minor [13:29] annoyance is that duet in this skill [13:31] library can't actually see all the [13:34] skills that I have on my local computer. [13:36] So I'm sure they're going to launch duet [13:38] in the cloud desktop app at some point [13:39] too and this problem will be fixed as [13:42] well. But basically right now the two [13:44] shortcomings in my opinion is lack of [13:47] ability to engage with Duet's cloud [13:49] browser and also I can't seem to add my [13:52] own personal skills on my local [13:53] computer. Okay, so let's wrap up by [13:57] comparing Duet to all the other personal [14:00] agents that I tested. You know, I made [14:02] videos on Grockbot Muse Chat Dots and [14:05] all these personal agents now. So I want [14:07] to give you guys a brief recap on which [14:09] one I think is the best. So, spoiler [14:12] alert, I think Grockbot is still my [14:14] favorite agent right now. Basically, I [14:16] treat each bot as an employee and the [14:19] free X integration that Grockbot has is [14:22] really awesome for market research. The [14:25] downside right now is that Grockbot's [14:27] email doesn't actually work that well. [14:28] If I tag it on its own email, it doesn't [14:31] actually respond and Grockbot obviously [14:34] is not free. Now, if you want to give a [14:36] a try, I have some free coupons on [14:38] behindthecraft.com [14:40] for pay subscribers. [14:42] The next one is Muse, which is free. And [14:46] Muse is also incredible. I think it's [14:48] super welldesigned, probably the most [14:50] well-designed out of all the personal [14:52] agents. Maybe Duet comes close. And it [14:55] has a killer feature in that it can call [14:58] US businesses for you. So, for example, [15:00] here I'm calling Davy Tree Expert [15:02] Company to try to get them to remove a [15:05] tree from my backyard and you know, Muse [15:07] is making a call. I've also made calls [15:09] before to negotiate cable bills and more [15:12] and it's just a massive timesaver. The [15:14] primary drawback of Muse is that it's [15:17] not really using a Frontier model like [15:18] Opus and Astra. So, sometimes it can [15:21] make mistakes. I think Muse Wonder is a [15:23] good model, but in my opinion, Opus, [15:26] Astra, Fable is a better model. Okay, [15:29] next we have chat GBT dots which lives [15:31] inside chatbt which is probably its [15:34] biggest advantage because it's part of [15:36] catchb it has access to chat memory [15:39] chatb threads and more. Now the problem [15:41] with chat dots is that honestly it [15:45] launched in a really buggy state and [15:47] they made a lot of improvements since [15:48] then but personally I still find it [15:50] slower and less useful than just using [15:54] my own chief of staff thread inside [15:56] chatbt. [15:58] And you know last but not least we have [16:00] claw duet which in my opinion is the [16:03] most proactive out of the box. I know [16:05] Anthropic has been working on this for a [16:06] while and the product has also changed a [16:08] lot and it's actually a pretty polished [16:11] experience. I love the claw look. It's [16:13] just missing some key features right now [16:15] that I'm sure they're going to add once [16:17] they exit research preview like be able [16:19] to see what the browser is doing and be [16:21] able to add my own skills easily. All [16:24] right, so overall I would say I still [16:26] prefer Grockbot as number one. I still [16:29] use it every day along with chatbt and [16:31] claude for individual work streams. If [16:34] Duet becomes more robust after it exits [16:36] research preview, I can see myself using [16:39] it a lot because right now Claude is my [16:42] main harness and model and this has a [16:44] ton of memory about how I like to work. [16:46] So overall, I think this is generally a [16:48] great start to a personal agent from [16:50] Anthropic and I'm really excited to see [16:52] them improve it. Claw Duet is available [16:55] today as a research preview with a [16:57] weight list for pro and max subscribers [17:00] and I'm sure Anthropic is trying to [17:01] launch it to as many people as possible [17:03] very very soon. And again, my prediction [17:06] is that Anthropic will improve to it and [17:08] eventually make it a feature inside the [17:12] main cloud app both on desktop, web, and [17:14] mobile. As always, if you enjoyed this [17:17] breakdown, please like and subscribe for [17:20] more practical AI tutorials, and I'll [17:23] see you next time.