Early Claude Duet review finds scheduled tasks cannot use custom connectors

Claude Duet can keep work moving between conversations, but some connections stop at scheduled tasks. In an October 21 early-access review, Every’s Dan Shipper says Anthropic’s new assistant cannot use custom connectors during scheduled runs. He also finds its email drafts weaker than OpenAI’s Dot and says it cannot direct existing Claude sessions on his computers.

A professional tool, with uneven strengths

Shipper’s overall verdict is enthusiastic. After two weeks of early access, he calls Duet the first personal agent he would describe as a professional tool. He says it separates recurring subjects into channels, tracks what it is working on and what it needs from him, and sends suggestions he actually wants.

The appeal is not just reminders. Shipper says Duet can edit documents, slides, designs and dashboards inside the app and keep them current. He credits the experience with bringing his usage closer to an even split between ChatGPT and Claude, after previously spending most of his time with OpenAI’s products.

That does not make it his winner at every job. His preference for Dot’s email drafts sits alongside a separate Duet restriction: it can see Claude sessions on his computers but cannot direct them. He also reports that voice input is transcription, rather than a full voice mode. These are Shipper’s observations, not independently verified performance results.

Every disclosed that Anthropic supplied early access and had no input on the review.

Background work has a usage cost

Anthropic’s launch announcement describes an assistant that inherits memory, skills and connected apps from a Claude account. It runs on Opus 5.5 and has a virtual computer with its own browser, allowing tasks to continue while a user’s phone or laptop is off. The company says it sends notifications when work finishes or needs review.

“Always-on” does not mean constant checking. Anthropic says Duet checks connected accounts every few hours because nonstop checks would exhaust plan limits faster. Those background checks share the same usage allowance as ordinary Claude work. Larger assignments, such as financial models or slide decks, require additional model turns and background processing.

The research preview opened October 21, initially through a waitlist for US Pro and Max subscribers. It is included in those plans and available through standalone iOS and Android apps and claude.ai, separate from the main Claude app. Anthropic is targeting professionals and small-business owners rather than presenting personal errands as its strongest use.

Preparation is automatic; consequential actions need approval

Anthropic draws a line between preparing work and acting on someone’s behalf. It says Duet can read, draft and prepare independently, but must show the proposed action and wait before taking these steps:

• Sending a message or making a payment.
• Signing a document or deleting anything.
• Trading securities and moving money between accounts remain prohibited, even with approval.

The company also says a dedicated safety system checks messages for sensitive information. It is designed to reject requests to share credit card numbers or government identification, and to permit health or personal-finance information only on explicit instruction. Those are Anthropic’s stated protections, distinct from Shipper’s assessment of everyday usefulness.

Anthropic acknowledges that Duet may misread conversations or misjudge urgency. It also cannot yet show users its browser, and personal errands such as booking tickets remain weaker. The preview therefore pairs background execution with approval requirements, while leaving users without a view of that browser.