Claude Duet Is Anthropic's Most Polished Agent Launch — and Still Forgettable

Anthropic's new personal agent, Claude Duet, is the most polished agent launch to date — and its reviewer still forgot to use it. That tension is the core finding of a hands-on review by product veteran Claire Vo, who tested the phone-based agent across real work tasks and found execution gaps that polish alone cannot close.

Claude Duet lives on the user's phone, inherits all existing Claude plugins and connectors, and targets small business owners and busy professionals. Vo, creator of ChatPRD, put it through concrete workloads: inbox and Slack triage, a sales pipeline, a Google Docs sales deck, a launch video, and finding Warriors tickets for her kids.

What does Duet get right?

The launch stands out on design. Vo cited thoughtful onboarding, task lists, first-class artifacts, and Slack-style channels — three primitives that structure how an agent organizes and hands back work. Task lists let users queue work; artifacts give outputs a durable home; channels group conversations the way teams already communicate.

Even the app icon drew a dedicated segment of the review, a sign of how far the surface-level craft goes. The takeaway: the interaction layer is ahead of nearly every competing personal agent launch.

Where does it fail?

The failure modes are structural, not cosmetic:

• Proactivity. An agent that never nudges its user is a growth problem. Vo found Duet insufficiently proactive — it waits to be summoned in a market where every personal agent competes for the same limited attention.
• Latency. Model response time, the delay between request and usable output, erodes willingness to work with an agent synchronously. Vo treats this as a deciding factor for daily adoption, not a nice-to-have.
• Task quality. The agent stalled across the board on real work: inbox triage, sales decks in Google Docs, and launch videos all exposed gaps, including a shortfall in local skills.
• Missing platforms. No desktop app and no multiplayer support keep Duet out of Vo's working stack. She draws a line between single-player and multiplayer agents, and places Duet firmly on the wrong side of it for team workflows.

The verdict: polish is not enough when the competition is every other agent vying for the same habit slot.

What can product managers take from this?

The review names the conditions under which a personal agent succeeds or fails. It works when it is proactive enough to earn a reflexive habit, fast enough for synchronous work, and present on every surface where work happens — desktop included. It fails when users must remember it exists, wait on it, or switch devices to finish a task.

Vo also positions Duet against a crowded field, referencing Claude Cowork, OpenAI's Codex, xAI's Grok Bot, and Meta Muse as context for where agent product design now stands.

What comes next?

Anthropic has shipped the best-dressed entrant in the personal agent category, but retention will be decided by proactivity, latency, and desktop presence — the unglamorous constraints that turn a beautiful agent into a default one.