Source: raw/I_tested_Claude_Duet._Is_it_worth_switching_from_Grok_Bot.md (Claire Vo), raw/Anthropic_Just_Dropped_Claude_Duet_Full_Breakdown.md and raw/Claude_Duet_is_the_Grok_Bot_Killer_LIVE_TEST.md (Brock Mesarich), ai-research/every-vibe-check-claude-duet-shipper-2026-10-21.md (Dan Shipper, public portion only), ai-research/peter-yang-x-personal-agents-ranking-claude-duet-2026-10-21.md (Peter Yang, partial X capture), ai-research/eesel-claude-duet-review-2026-10-22.md (eesel, no hands-on access; Hacker News quotes).
What people with early access said about Claude Duet in its first week. Four people had hands-on time in that week (Whittemore, Jones and a Vo benchmark were added 2026-11-01); their verdicts split on one question: does Duet’s quiet, careful design make it useful, or easy to forget? Every reviewer below had a few hours to two weeks with a research preview, so treat each as one data point.
The Claire Vo and Brock Mesarich sections were moved here unchanged from the original claude-ai/claude-duet article on 2026-10-28.
At a glance
| Reviewer | Access | Verdict | Liked | Disliked |
|---|---|---|---|---|
| Dan Shipper (Every) | Two weeks, early access from Anthropic | ”First personal agent I’d call a pro tool”; moved from mostly ChatGPT to “roughly 50/50” | Channels, running task list, editing docs, slides and dashboards in-app, “suggestions I actually want” | Can see but not direct his Claude sessions; email drafts worse than Dot’s; scheduled runs can’t reach custom connectors; voice is transcription only |
| Peter Yang | Early access | ”Most proactive agent”; Artifacts “could be its secret sauce”; “arguably the best model behind it” | Simple to use, proactive by default | Missing preview features, “like being able to drive its browser” |
| Claire Vo (How I AI) | A few days, TestFlight builds | Worth trying, “not sure it’s going to become part of mine” | Onboarding, polish, task lists, artifacts, channels | Too quiet, slow (10+ minute threads), phone only, local skills don’t carry over, single-player |
| Brock Mesarich | A few hours, phone only | Unsure he would switch from Grok Bot | Make section, Tasks view, quick voice | Prefers Grok Bot’s separate agents to channels; no way to add connectors in-app |
| Nathaniel Whittemore (AI Daily Brief) | A few days, web and app | ”Much more optimistic about this approach” for proactive work; not his main way of using AI | Drafted replies to emails he would have starred, sent from chat; web interface; channels; artifacts | Over-proactive on a non-problem; a brand design task was “much worse” than in Claude’s own design tool |
Shipper’s disclosure: “Anthropic provided Every early access to Claude Duet. Anthropic had no input on this review.”
Dan Shipper, Every (Vibe Check, 2026-10-21)
Only the free portion of the review is saved; the rest is paywalled.
- Why he tested it. His OpenAI Dot had “become less proactive” and started losing tasks: “I sent it off to do my taxes, and it didn’t finish and didn’t alert me.”
- Result. “I’ve landed closer to 50/50 between ChatGPT and Claude. It’s the most time I’ve spent with Anthropic in close to a year.”
- What makes it a pro tool: it “splits recurring topics into their own channels, keeps a running list of what it’s working on and what it needs from me, and can edit docs, slides, designs, and dashboards and keep them current without leaving the app.”
- Gaps: “it can see the Claude sessions on my computers, but it can’t direct them; its email drafts are worse than Dot’s; and its scheduled runs can’t reach my custom connectors. There’s no real voice mode yet, just transcription.”
- Advice: “if you’re a Claude subscriber who does knowledge work, get on the waitlist.” On X: “I’m opening Duet every day.”
Peter Yang (X, 2026-10-21)
Captured from search snippets; his full video review was not watched.
- Launch post: “the most proactive agent”, and “Artifacts could be its secret sauce.”
- His four-agent ranking: Grok Bot (“a team of bots is the easiest to understand”, but “email feature doesn’t really work”), Muse (“free, best consumer UX”), Claude Duet (“simple to use. Arguably the best model behind it”, but missing preview features “like being able to drive its browser”), ChatGPT Dots (“can drive all your cloud and local threads” but “still buggy”).
- He caps himself at four agents connected to his email at once.
Hacker News launch thread (via eesel)
eesel’s reviewer had no invite and scored Duet 3.5/5 from the documentation and the launch thread (544 points, 510 comments when pulled). Reactions split three ways:
- Model fans were excited about Opus 5.5 behind a personal agent.
- Early users found the UI busy and the pace slow. joshstrange: “I hated the UI/UX within seconds. It started ‘working’ and dumping ‘Cards’ in the chat of the tickets it was looking up.” He also said that when Duet lacked data it needed, it pushed ahead instead of asking. robgough agreed with a video preview’s verdict: “so slow that it’s not actually particularly pleasant to use, even if it is smart.”
- Skeptics: loveparade, “I still don’t understand how this is different from just a Claude session, with a browser.” eesel’s answer: for a non-technical owner, the packaging (brief, approvals, payments, phone app) “is the whole product.”
eesel’s scorecard: judgment 4.5, safety design 4.5, value for money 4, apps it can reach 3, transparency 2.5, availability 1.5.
First hands-on review (Claire Vo, October 2026)
Added 2026-10-27. Creator: Claire Vo · URL: https://www.youtube.com/watch?v=HYb5w4PfPZM · Platform: YouTube (How I AI) · Source: raw/I_tested_Claude_Duet._Is_it_worth_switching_from_Grok_Bot.md. One user with early access, after a few days of use.
What she liked:
- It inherits your Claude setup. Duet “inherits all your plugins from Claude”, so connectors such as Gmail, Google and Figma are already there.
- Task lists that track both your goals and what the agent is doing.
- Artifacts are first-class. Duet already shows the artifacts from your existing Claude account, and creates new ones as you chat.
- Channels. You can split work off the main chat into Slack-style channels, for example a “sales” channel.
- Polish. “Approximately 100 times more polished than Dots was when it came out.”
What she did not like:
- Not proactive in practice. “Once I set up Duet… it never pushed or bothered me to do anything ever again.” That is consistent with Anthropic’s design of messaging “only when something needs attention”, but for her it meant she forgot it existed. She notes she may not have turned notifications on.
- Slow. Multi-part requests split into separate threads that took “10 plus minutes” to finish or come back. A ticket-shopping errand (Warriors tickets) “didn’t close the loop.” This matches Anthropic’s own caveat that personal errands are weaker than work tasks.
- No desktop app. It is phone-only during her workday, which she calls “a really disconnected experience.”
- Local skills don’t carry over. “Duet does not ingest all your local skills.” To use them you would have to publish them to Claude as plugins.
- Single-player only. There is no way to share agents with a small team, which she says Claude Tag and Claude in Slack have not solved either.
Task results:
- Email and Slack triage “did okay”, but “the priority didn’t seem right.”
- Sales pipeline: it pulled data from Slack, email and the CRM quickly, because those were already connected.
- A Google deck started from her phone: “wasn’t bad”, but not on-brand, because it did not use her Figma design.
- Launch video: not possible without her local skills.
Her verdict: worth trying from the waitlist, but “I’m just not sure it’s going to become part of mine.”
What this means for setup: connect and test the plugins you need in Claude first, since Duet inherits them. Publish any local skill you depend on as a plugin. Turn notifications on, and check open threads yourself rather than waiting for Duet to report back.
Brock Mesarich field notes (2026-10)
Added 2026-10-27. Creator: Brock Mesarich · Platform: YouTube · two videos: “Anthropic Just Dropped Claude Duet (Full Breakdown)”, https://youtu.be/nv1aaP3dHzs, 7:39, uploaded 2026-10-21 (raw/Anthropic_Just_Dropped_Claude_Duet_Full_Breakdown.md), and “Claude Duet is the Grok Bot Killer? (LIVE TEST)”, https://youtu.be/PkrjD2IOCm4, 11:55, uploaded 2026-10-22 (raw/Claude_Duet_is_the_Grok_Bot_Killer_LIVE_TEST.md). Brock uses Grok Bot as his daily agent, and the live test covers a few hours on his phone only.
Launch details he reads from Anthropic’s post. Moved 2026-10-27: the launch details Brock read from the post (built on Opus 5.5; memory, skills and apps carry over; one ongoing thread; red teams and an outside firm; the early-user examples) are confirmed by the full post, saved as ai-research/claude-blog-introducing-claude-duet-full-2026-10-27.md, and now sit in the body of Claude Duet with that file as source. His reading matched, with two small differences: he calls Boris “the guy who created Claude Code”, where the post says Cherny “leads our Claude Code team”; and he says “pretty much no setup”, where the post says “minimal work”. One item is his alone:
- Launch-video example: a coffee shop owner is told her new location opens in 26 days and the health permit is not filed. Duet drafts the permit and says a client renewal can wait until Monday. This is from Anthropic’s launch video as Brock describes it; it is not in the saved post.
What the app showed in his live test (2026-10-22, off the waitlist):
- Access panel: files and documents, mail, calendar, team chat and payments. It could not make payments yet, but you can add payment methods.
- Voice mode answered “what’s on my calendar over the next couple of days” with his flight times. He found the latency “very quick”.
- Channels: from the three-dot menu (“Get focused with channels”), give a channel a name, a purpose, an icon and a colour. He made “YouTube videos”. He calls them “very similar to Slack”. A new channel offers to suggest tasks it can do for you.
- Tasks: a spinner on a running job opens its step list. The Tasks tab shows Active, Up next and items ready for review. It includes things he did not ask for: an important email that arrives gets added to tomorrow’s list.
- Make: every artifact on the account in one place, including PDFs, CSVs, Claude docs, presentations and work made earlier in Claude Code. Use ”+” to start a new one by type, such as a dashboard.
- Data test: asked to check his YouTube analytics, it pulled from the Supabase database he had set up through Claude Code and reported 1.35 million views over the last 28 days. A second request produced a shorts dashboard with views, likes and engagement for his last three videos.
When an app has no Claude connector. He wanted email stats from Drip, which has no Claude connector, and could not find a way to add connectors inside Duet. Duet uses the connectors on your Claude account, so set them up on your computer. His workaround:
- In Zapier MCP, go to MCP servers → Create server, pick the client (he picked Claude Code; the list also offers Claude, ChatGPT, Cursor and Grok Bot), and paste the setup prompt it gives you into that client.
- Under Apps, add Drip (seven tools, including pulling stats, tagging subscribers and adding someone to a campaign) and sign in.
- In a Duet channel, ask for the last campaign’s results as a dashboard. About a minute later it returned open rate, click-through rate, unsubscribe rate, delivery rate and where readers dropped off.
The transcript does not show how a server set up for Claude Code reached Duet. The likely route is a Zapier server already linked to his Claude account, since apps added in Zapier show up for every client connected to that server. ^[inferred] He gives a Zapier link in his video description. More on the connector workaround in Claude Connectors.
His verdict against Grok Bot.
- He likes Duet’s Make section, because Claude “has been really good at building artifacts” and he hasn’t seen that in Grok Bot, and the Tasks view.
- He prefers Grok Bot’s separate agents to Duet’s channels, “a cleaner experience”, but has used channels for only a couple of hours.
- After the launch-day video he was unsure he would switch from Grok Bot. The waitlist is US-only.
Usage limits. Anthropic says background check-ins count against your Claude limits. Brock’s own impression is that Opus 5.5 is “really efficient at spending my tokens” compared with his earlier Claude Code use. He measured nothing.
Nathaniel Whittemore, The AI Daily Brief (added 2026-11-01)
Source: raw/How_I_m_Actually_Using_Claude_Duet.md · Creator: Nathaniel Whittemore · URL: https://www.youtube.com/watch?v=HyEfRBW10Kk · Platform: YouTube. A few days of use, by someone who had dropped earlier proactive-agent setups (an OpenClaw chief-of-staff and research agents) because “the juice wasn’t worth the squeeze.”
- Onboarding pushes proactivity. It asked him to connect Slack, Gmail and Drive first, then offered “a stack of options for proactive projects” before he had done anything else.
- Over-proactive on a false alarm. From his inbox it decided he had promised the same podcast ad slot to two sponsors. He has several slots per day, so it was not a problem, but it reminded him “like four times” until he told it that work stream did not matter.
- Where it paid off. For emails “of exactly the type that I would star and go back to a couple days later”, it proposed replies he could edit or authorise from chat. He let it send a couple; on the most substantive one it got him “about 90% of the way there.”
- Web over phone. He “quite like[s] its web interface” despite the iOS push. Tasks are grouped active, up next, scheduled and resolved, and can be edited from the main chat or added directly.
- Not for design. A UI and brand design task was “much worse than doing it in Claude’s native design experience.”
- Verdict: less a new main way to use AI than “an easy to set up little baby version of myself that does a bunch of stuff that I don’t want to do.” He compares Duet favourably with his Dot, which he found “not to be particularly proactive.”
Two more data points (added 2026-11-01)
- Claire Vo’s nine-task benchmark put Duet last of four agents at 79%. It asked for permission four times where the others asked once, was slowest overall, failed to download a video, and was blocked by “almost every shopping site”. It won the launch-video task (on-brand, ElevenLabs MCP voiceover) and found an extra bug in the coding task. Full results: Claire Vo’s Personal Agent Benchmark.
- Nate B Jones (
raw/The_Future_Of_The_Internet_Isn_t_Open_For_Your_Agent._Muse_Dots_And_Claude_Duet_Will_Find_Out..md, https://www.youtube.com/watch?v=c3HHFL_lsLM): “within the first day or two it was proactive with me. It reminded me of something that I didn’t realize was coming up.” He puts Duet’s and Dots’ proactive alerts at useful “50 to 75% of the time.”
Where the reviews agree
- Judgment and polish are strong. Every hands-on reviewer praised onboarding, design or output quality.
- Speed is the main complaint. Vo’s 10-minute threads and the Hacker News comments describe the same thing. Neither Anthropic nor any reviewer gave timings beyond Vo’s.
- Quiet cuts both ways. Shipper and Yang call it proactive; Vo forgot it existed. Whittemore and Jones (2026-11-01) found it proactive within days, and Whittemore found it too insistent on one false alarm. The difference may be notification settings or use pattern; nobody tested that directly.
- Reach is the limit, not intelligence. Custom connectors in scheduled runs, Microsoft 365, desktop, team sharing and local skills all came up as gaps.
Key Takeaways
- The two longest tests (Shipper, two weeks; Vo, a few days) reached opposite conclusions about daily use, so run your own one-week trial before relying on it.
- Turn notifications on and check the Tasks tab yourself in week one; the most common failure was work finishing silently.
- Expect multi-step jobs to take minutes, not seconds.
- If your stack depends on custom connectors, test one in a scheduled run early (Shipper’s gap).
- Email drafting was rated below OpenAI’s Dot by one reviewer; review drafts before approving.
Open Questions
- Shipper’s full review and Peter Yang’s video were not read; their task-level results are not captured here.
- No reviewer measured how much of a Pro plan’s limits a week of background checks uses.
Related
- Claude Duet — the product overview.
- Getting Started with Claude Duet — a first-week plan that tests the reviewers’ complaints.
- Limits and Limitations — the documented gaps behind several complaints.
- Personal Agents Compared (October 2026) — the wider category.
- Grok Bot — the agent Vo and Mesarich compared against.
- Claude Connectors — including the Zapier MCP route for apps with no connector.
- Claude Tag — the team agent Vo wanted Duet to be.
- Claire Vo’s Personal Agent Benchmark — Duet against Grok Bot, Muse and Dot on nine tasks.