Key Takeaways

  • Anthropic structured Claude Duet around three concrete interface primitives: a dual-track task list, cross-platform artifacts, and Slack-style channels.
  • While OpenAI's Dots stumbled on polish, Vo points out that Duet eliminates sharp edges by explaining exact tool permissions during onboarding.
  • Duet splits work into channels for specific jobs like sales workflows, preventing the context drift common in single-thread chats.
  • Output artifacts generated in Claude Code or Co-work persist directly inside Duet as first-class objects rather than disappearing in chat history.
  • The product still trails competitors like Grok Bot and Muse on core mechanics, suffering from slow latency, no desktop app, and zero multiplayer support.

Task Lists Separate Human Work From Agent Work

Most agent interfaces treat autonomous software like a standard chatbot. You send a prompt, watch a spinner, and hope the model understood your intent. Claire Vo points out that Claude Duet changes this pattern with a dedicated task list view.

Instead of burying status updates in a running text feed, Duet displays a split view. “I like the idea of keeping track of what goals and tasks I'm doing and what goals and tasks the agents doing,” Vo explained. “So, I think the task list view is something that's really nice here and something that we're going to come back to a lot.”

This dual-track setup solves a basic coordination problem. When delegating to an agent, you must verify what it is handling versus what you still own. If you are reviewing contracts while the agent drafts outreach briefs, seeing both lists side by side prevents duplicate work. It also removes the anxiety of wondering whether the agent stalled or went off track.

Channels Prevent Single-Chat Chaos

When users run an AI agent across complex company workflows, standard single-thread chats break down. A user reviews sales metrics, switches to outbound copy, and then asks for code edits. The model scrambles context and loses earlier instructions.

Anthropic addressed this by borrowing from team communication tools. “Anthropic borrowed a little bit from Slack, which I feel like is everybody's favorite collaboration service, which positions these subhats, these subthreads, these like subplaces that you're going to work in as channels,” Vo noted.

Spinning off dedicated channels for specialized tasks, such as sales outreach, isolates context. The agent stays on task, and the user avoids scrolling through endless messages.

Anthropic also made generated files permanent. “If you're used to artifacts in Claude Code or in Co-work, if you're generating HTML artifacts, designs, slides, Anthropic has made those first class citizens in Duet,” Vo said. These outputs persist as standalone items rather than disappearing up the feed.

Polish Wins Onboarding but Latency Stalls Daily Use

Duet's visual craft stands in sharp contrast to OpenAI's Dots. Vo argued that what “Anthropic did so so so well that OpenAI's Dots did so so so poorly is the polish of the agent design. This app is beautiful. It is welldesigned. There are no sharp edges.”

That care starts at signup. “From the second you onboard, they are really thoughtful about walking you through what Duet does, what it has access to, and how to get onboarded,” Vo noted. Instead of dumping users into a blank text box, Anthropic clearly defines tool access and integrations upfront.

Yet design polish cannot mask infrastructure limits. Vo warned that Duet suffers from slow model latency that disrupts flow. It also lacks proactive notifications to alert users when tasks finish, has no native desktop app, and offers zero multiplayer collaboration. Competitors like Grok Bot and Muse push ahead on raw execution. Good onboarding wins initial interest, but slow response times risk user churn.

What to Do With This

Audit your product's agent workflows this week. Strip your primary interface out of an endless chat window and structure it around three explicit zones: a split task list showing user versus agent responsibilities, dedicated subchannels for isolated jobs, and a persistent side panel for generated artifacts.