Claude Duet review (2026): honest verdict on Anthropic's proactive AI agent on Opus 5.5 — proactivity, connectors, approvals, and the waitlist preview.
Claude Duet is one of the more convincing proactive AI agents of 2026: it watches your connected apps, hands you a morning brief, and works on tasks in its own cloud computer, pausing for approval before anything sensitive. On Opus 5.5 the reasoning is strong and the safety posture is sane. Two honest caveats hold the score back — it launched as a US-only waitlist research preview that shares (and spends) your plan's usage limits, and Anthropic itself admits it can still misjudge what's urgent. It's a genuinely useful assistant for managing your work; just don't expect it to generate or publish your content.
Claude Duet is Anthropic's answer to a simple question: what if the assistant reached out first? Launched October 21, 2026, Duet is a proactive agent that connects to your apps, checks them through the day, and surfaces what needs you — a short morning brief with drafted next steps, then alerts when something comes up. It runs on Opus 5.5 in a standalone app for iOS and Android and on the web at claude.ai, and it works in its own cloud computer so it can carry out tasks without local access to your machine.
This review scores Duet as what it is: a personal proactive agent. It is not a content-creation suite, and I don't grade it as one. Where it competes — against other proactive consumer agents and against Claude's own Cowork — it holds up well on reasoning, connector breadth, and the discipline of asking before it acts. The scores below reflect that, while being honest that this is an early preview with real access limits.
Two things anchor the verdict. First, the proactivity is the product, and it mostly works: a morning brief that drafts next steps, periodic app checks, and tasks executed in a sandboxed cloud browser is a legitimately new shape for an assistant. Second, the guardrails are the right ones — it pauses for approval before sending a message, making a payment, signing a document, or deleting a file, it doesn't see your passwords, and it shows its work so you can steer it. The limits are about maturity and reach, not philosophy.
Everything below reflects Duet's state as of 2026-10-22 and is based on Anthropic's own product materials plus launch-day reporting. Specs, connector counts, and availability can change fast on a research preview, so confirm current details on the official page before you rely on them.
Claude Duet is a standalone proactive AI agent from Anthropic, built on Opus 5.5 and released as an early research preview on October 21, 2026. Rather than waiting for a prompt, it connects to your apps — Anthropic cites more than 6,000 connectors and plugins, including Gmail, Slack, Stripe, and QuickBooks — checks them periodically, and reaches out: a morning brief with drafted next steps, then alerts through the day when something needs attention. It carries over the memory, skills, and connected apps from your existing Claude account, so setup is light, and it keeps everything in one ongoing thread. Duet runs tasks in the cloud on its own virtual machine with its own browser, so it can act without local access to your computer — handling light admin like scheduling a meeting or heavier work like building a dashboard or small app, then reporting back. The safety model is explicit: it asks for approval before sending messages, making payments, signing documents, or deleting files; it doesn't see passwords or logins; and it runs a dedicated classifier and safety prompt against sensitive-information leaks. At launch it is available by waitlist to Pro and Max subscribers in the United States only — Team and Enterprise are not included yet — with no separate charge, sharing your plan's usage limits (its background checks count toward them). What Duet does not do is generate on-brand multi-format content or publish to social platforms; it manages your work, it doesn't manufacture your content.
Duet fits professionals and small-business owners who spend their day across email, Slack, calendars, and a stack of SaaS tools and want an agent to triage all of it — flag what's urgent, draft the reply, set up the meeting, and take on the occasional build task. It is strongest for people already on a Claude Pro or Max plan in the US who want their assistant to initiate rather than wait, and who are comfortable approving sensitive actions as they come up. It fits poorly for anyone expecting a content tool: Duet makes no short-form video, carousels, quote graphics, or blogs and publishes to no platform, so a creator whose bottleneck is producing and shipping a week of posts will find it manages the surrounding work but not the content itself. It's also an early preview with US-only, waitlist access and shared usage limits, so treat it as promising rather than production-ready.
| Dimension | Score | Why |
|---|---|---|
| Proactivity & morning brief | 4.3 / 5 | The core idea works: a morning brief with drafted next steps plus periodic app checks is a genuinely new, useful assistant shape — though Anthropic admits it can misjudge urgency. |
| Task execution & autonomy | 3.8 / 5 | Runs real tasks in its own cloud browser, from scheduling to building a dashboard, on Opus 5.5 — capable, but the approval gate means it is semi-autonomous by design, not hands-off. |
| Connectors & integrations | 4.4 / 5 | Anthropic cites 6,000-plus connectors and plugins including Gmail, Slack, Stripe, and QuickBooks, and it inherits your existing Claude apps and skills. |
| Safety & approvals | 4.3 / 5 | Asks before sending, paying, signing, or deleting; doesn't see passwords; runs a dedicated classifier and safety prompt — a sensible default for an agent that acts on your behalf. |
| Transparency & control | 4.1 / 5 | Shows what it has done and is doing and lets you steer it mid-task, which matters a lot for trust in a proactive agent. |
| Availability & maturity | 2.8 / 5 | US-only, waitlist, Pro/Max only, no Team or Enterprise, and an explicit research preview Anthropic says it is still tuning — reach and readiness are the weak spots. |
| Value & plan inclusion | 4.0 / 5 | No separate charge during the preview is a plus, but it shares your plan's usage limits and its background checks spend against them, so heavy use eats your Claude quota. |
| Content-workflow scope | 1.5 / 5 | Not a content tool: no multi-format generation (video, carousels, blogs) and no social publishing. Not what Duet is for, and scored only to set expectations. |
Duet's pricing is, for now, an inclusion story rather than a line item. During the research preview there is no separate Duet charge: it is bundled into Claude Pro and Max, available by waitlist to US subscribers, and it draws on your plan's existing usage limits. That makes it effectively free to try if you already pay for Claude — Pro runs about $20/month, and the Max tiers about $100/month (5x usage) and $200/month (20x usage) — but "free" here has an asterisk, because Duet's background checks count against the same limits as the rest of your Claude usage. A proactive agent that polls your apps every few hours is, by design, spending quota in the background.
That changes the value math depending on your plan. On Pro, a busy Duet could consume a meaningful slice of your monthly usage just staying aware, which may push heavier users toward Max purely for headroom. On Max, the shared-limit model is far more comfortable. There is no standalone Duet subscription to evaluate yet, and no Team or Enterprise access, so businesses can't buy it for a team at launch.
The honest read: as a preview bundled into plans you may already hold, Duet is fairly priced — you're not paying extra to try a genuinely novel agent. Watch the usage-limit interaction, and watch for how Anthropic prices it if and when it leaves preview. What the price never includes is the content work: Duet does not write your posts, make your visuals or video, or publish anything, so if that is the job you're trying to fund, you're pricing the wrong tool.
| Use case | Fit | Why |
|---|---|---|
| Triaging a busy inbox, Slack, and calendar | Strong | This is the core job: Duet watches your connected apps, flags what needs you, and drafts the next step in a morning brief. |
| Running light admin and the occasional build task | Strong | Scheduling meetings or building a dashboard in its own cloud browser is squarely what it is designed to do, with approval gates on anything sensitive. |
| A proactive assistant you already pay for via Claude | Strong | If you are a US Pro or Max subscriber, it inherits your Claude memory, skills, and apps and costs nothing extra during the preview. |
| Hands-off, recurring automation of sensitive actions | OK | The approval gate is the right default but means it pauses for sending, paying, signing, or deleting — deliberately not fully autonomous. |
| Team or enterprise rollout | Weak | Launch is Pro and Max only, US-only, by waitlist; Team and Enterprise plans are not included yet. |
| Generating social posts, scripts, or blogs | Weak | Duet is a general agent, not a content generator — it writes no multi-format content and has no brand-voice layer for a content week. |
| Producing short-form video, carousels, or quote graphics | Weak | It creates no visual or video content assets; that is outside its scope entirely. |
| Scheduling and publishing across social platforms | Weak | No social connections for posting and no scheduler — Duet publishes nothing to a feed. |
To be clear where I stand: I run Kompozy, and Kompozy is not a Claude Duet competitor in the way a comparison page implies. Duet is a proactive personal agent — it manages your attention, your inbox, your calendar, and the odd build task. Kompozy is a content operation — it manufactures and publishes the posts. They share one trait, a proactive "do it before I ask" posture, pointed at two different halves of the day.
That shared trait is exactly why a creator might land here confused. If you adopted Duet hoping a general agent would run your content, the honest answer is that it won't: Duet writes no short-form video, no carousels, no quote graphics, no blog, and it posts to no platform. Kompozy does that part — copy under a Persona Brief so it stays on-voice, Persona Shorts and avatar video with built-in HeyGen TTS, brand-exact carousels, quote graphics, photo posts, a blog, and a newsletter, all scheduled and published across nine platforms with its own Autopilot and a per-post review queue. The clean pairing: Duet to surface which ideas are worth posting and keep your day clear, Kompozy to turn those ideas into a week of on-brand content and ship it. Two proactive agents, two jobs.
For managing your work, yes — if you are a US Claude Pro or Max subscriber, Duet is a genuinely novel proactive agent that triages your apps, drafts next steps, and runs tasks with your approval, at no extra charge during the preview. It is not worth it as a content tool, because it generates no video, carousels, or blogs and publishes nothing to social; for that you need a content engine like Kompozy.
During the research preview there is no separate Duet charge — it is bundled into Claude Pro (about $20/month) and Max (about $100/month for 5x usage and $200/month for 20x), available by waitlist to US subscribers. Note that Duet shares your plan's usage limits and its background checks count toward them.
Claude Duet runs on Opus 5.5, Anthropic's flagship model, and executes tasks in its own cloud virtual machine with its own browser so it does not need local access to your computer.
Anthropic built in guardrails: Duet asks for approval before sensitive actions like sending messages, making payments, signing documents, or deleting files; it does not see passwords or logins; and it runs a dedicated classifier and safety prompt to guard against leaking sensitive information. It also shows what it is doing so you can steer or stop it.
No. Duet is a general proactive assistant; it can draft content and connect to apps, but it does not generate finished social formats (short-form video, carousels, quote graphics) or schedule and publish to social platforms. Creators pair it with a content engine such as Kompozy, which generates on-brand content and publishes across nine platforms.
Both are Claude agents, but Duet is proactive — it polls your connected apps and reaches out with a morning brief and alerts before you ask — whereas Cowork is a collaborative agent you direct. Duet runs always-on in its own cloud computer; Cowork is more of a session you drive.
At launch Duet is an early research preview available by waitlist to Claude Pro and Max subscribers in the United States. Team and Enterprise plans are not included yet, and availability may expand as it leaves preview.
Yes. Duet is a standalone app for iOS and Android, and it also runs on the web at claude.ai. It carries over the memory, skills, and connected apps from your existing Claude account across those surfaces.