TL;DR
The short version
Claire Vo found Anthropic's Claude Duet the most polished personal agent she has tried, and still stopped using it within days. A personal agent now has to pull you back, finish fast, sit where you work and include your team. Design alone doesn't do that.
On How I AI, Vo walks through what Duet gets right (onboarding, inherited plugins, artifacts, channels) and why it fell to the bottom of her list: it was quiet, slow, phone-only and single-player.
Built on Claire Vo's How I AI review (21 October 2026). Product facts checked against Anthropic's launch post and the Claude Help Center. Latency and polish comparisons are Vo's impressions.
The design is the best part
Anthropic opened a research preview of Claude Duet on 21 October 2026: a phone-first agent for professionals and small business owners that works in the background and asks before it acts. Claire Vo tested early builds and opened with her real problem. She has personal-agent fatigue. To win her attention, a new agent has to be "not good, not great. You have exceptional."
Her praise is specific. Onboarding is clear about what Duet can do and what it can reach. It inherits the plugins and connectors already set up in Claude, so Gmail, Google and Figma are there on day one. It keeps a task list of what you and the agent are each working on.
Artifacts (the HTML pages, decks and designs Claude generates) are first-class objects, and Duet could already see the ones in her Claude account. Side conversations become Slack-style channels off one main thread. She even likes the icon. Her summary: roughly "100 times more polished" than OpenAI's Dots was at launch.
Calm is a retention problem
Then she stopped opening it. After setup and a few first tasks, "it never pushed or bothered me to do anything ever again." The morning briefing didn't help: "There are only so many three-point morning briefings that can pull me back into a product."
That calm is deliberate. Anthropic's help pages say Duet checks your accounts at set intervals and messages you when something needs you. Vo reads the same restraint as a growth problem. With several agents competing for her day, the quiet one gets forgotten. "I need you to be a little bit more annoying."
Slow agents lose to fast ones
Her second complaint is speed. Duet runs on Claude Opus 5.5, and you can't choose another model. She likes the model's work but felt the wait. Tasks she fired off from one chat spun into separate threads that took 10 minutes or more to finish or come back for permissions. Because Duet wasn't noisy, she forgot about them.
Her ticket-shopping test (find Warriors seats she could take her kids to) took so long that she had moved on before it closed the loop. Anthropic's launch post admits personal errands "such as booking tickets" aren't as strong yet.
If that colleague is slow, if that colleague goes to a meeting to get work done, like, you kind of forget and don't want to go to that colleague over your more responsive ones.
Claire Vo, How I AI
She would take a faster agent that does about 10% worse work, she says. An agent is supposed to be a coworker who answers right away. A slow one breaks that.
The gaps are about where you work
The rest of her list is about reach, not quality.
- No desktop app. She doesn't want to be on her phone during meetings and work. The agents she keeps using sit on her desktop where she sees them.
- No local skills. The launch-video skills she built in Claude didn't come across. Duet can't see files on your computer, so she'd need to repackage them as plugins.
- Single-player. She runs a small business with about five people she works with regularly. If she can't share an agent with them, she doubts it will "worm its way into my working stack." Team and Enterprise plans aren't eligible yet.
Some work went fine. Duet pulled her sales pipeline from Slack, email and the CRM quickly and built a passable deck in Google Docs from her phone. For a busy owner with a lower design bar than hers, she thinks that alone would be useful.
What the review is and isn't
This is one heavy user's taste after a few days on a research preview, and Vo says so. She uses many agents and has a very high design bar. Her latency and "100×" numbers are impressions, not measurements. Anthropic calls Duet an early version that "might misread a thread or miscalibrate what's urgent."
Her test still holds for any agent product. Being well made is no longer enough. Agents now compete for the user's return visits.
