> ## Content Index
> Fetch the complete content index at: https://www.livain.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Claude Duet gets your things done. The harder question is who gets our things done.
- URL: https://www.livain.com/blog/claude-duet-personal-assistant-vs-orchestrator/
- Published: 2026-10-26T08:23:49.000Z
- Updated: 2026-10-27T05:42:26.000Z
- Description: Anthropic's Claude Duet is the proactive assistant many of us were waiting for. For a solo operator it could be the back office. For a team of five, it raises a harder question: when does a personal assistant become an orchestrator?
- Author: Remco Livain
- Tags: AI, Claude, Agentic Workflows, Team Design

Anthropic has opened a research preview of [Claude Duet](https://claude.com/product/claude-duet?ref=livain.com), and I'm more excited about this one than about most launches this year. Not because of the model behind it, but because of the shape of it. For the first time, Claude doesn't wait for me to open a chat. It comes to me.

Duet is a separate app where Claude works as a proactive agent. It briefs you in the morning, flags what needs you in your inbox and calendar, drafts next steps, and then goes quiet until something actually needs your attention. It runs on its own cloud computer, so it keeps working when you close the app. And it asks before anything sensitive, like sending a message or making a payment.

That's the version of AI most of us were hoping for when this all started. Something that sees what could be done, should be done, or already has been done, and tells you.

### Why this one lands for me

I've been working with Claude since the beginning of this year. The last six months have changed what I do, what I'm able to do, and how fast a rough thought turns into finished work. I've built a setup around it: memory, skills, connectors, scheduled routines. Most days it just works.

Duet carries all of that over. It uses the same memory, skills and connected apps as your Claude account, so there's very little to set up. That matters more than it sounds. The months you've put into teaching Claude how you work don't reset. They compound. It's the same logic as when [Claude merged Chat and Cowork](https://www.livain.com/blog/claude-merged-chat-and-cowork/): fewer places to decide where work lives, more time spent on the work.

The trust model is right too. Duet shows you what it has done and what it's doing, and waits for approval before it acts on your behalf. I've written before about how [an agentic routine owes you more than "the draft is ready"](https://www.livain.com/blog/draft-is-ready-not-a-status-agentic-routine-verify-own-claims/). A proactive agent that shows its work is the only kind I'd let near a client inbox.

### It needs to be about business, not just life admin

Meta's Muse and ChatGPT Dots got there first, and both are framed around your personal life. I tested Muse [in its first week](https://www.livain.com/blog/meta-muse-week-one-integrations-beat-model-quality/), and the integrations were the interesting part, not the model. But a dinner reservation is not where the real pain sits.

The real pain is the business inbox. The client reply that slipped. The proposal that should have gone out on Tuesday. The invoice nobody chased. For a solo operator or a fractional executive, there's no assistant, no EA, no office manager. A proactive agent that understands your clients and deadlines isn't a convenience. It's the back office you never had.

Anthropic itself calls Duet a personal assistant, and for now it's limited to Pro and Max subscribers. Team and Enterprise aren't part of the preview yet. That's exactly why I think the business case deserves more attention than the personal one.

### When does an assistant become an orchestrator?

Here's where it gets tricky. Picture a project with five people, each with their own proactive assistant. Five agents, each optimizing for their own person's priorities. You'd get duplicate follow-ups, conflicting nudges, and context stuck in five separate silos. Every assistant is proactive toward a slightly different version of the same project.

An assistant works for one person's goals. An orchestrator works toward a shared goal and decides who does what. The line is crossed when the AI can see everyone's work, holds one shared picture of what the team is trying to achieve, and has the authority to assign tasks. That last part is the real shift. Today, you delegate to the AI. In an orchestrated team, the AI delegates to you, based on who is best placed for the job.

> A personal assistant gets your things done. An orchestrator gets our things done, and that only works if we all agree on what getting things done means.

That's not a technology problem. It's an alignment and governance problem. If a team agrees on the goal, proactive AI makes it faster. If it doesn't, AI makes the disagreement faster. And context is the steering wheel: whoever feeds the agent the most context steers where it goes. I learned a small version of this when [a scheduled routine of mine failed three weeks running](https://www.livain.com/blog/scheduled-claude-code-routine-failed-agentic-workflows-context/) because it didn't have the context a written lesson couldn't give it. Multiply that by five people with five different ideas of the priority.

### Start small, then scale

Nobody knows yet how this plays out in teams, and that's exactly why it needs testing in team setups. But the sensible order is clear. Solo entrepreneurs first: one person, one goal, no alignment problem, so the cleanest possible test. Then small teams with an explicitly shared goal and shared context. Organizations last, once there are clear rules about who sets priorities, what the AI may assign, and who is accountable when it gets something wrong.

Anthropic is honest that Duet can still misread a thread or misjudge what's urgent. That's fine for a preview. It's also a reminder that the early-adopter operators are the ones who'll figure out where the edges are.

### And yes, Europe waits again

The preview is a waitlist for US subscribers only. I'm in Switzerland, so I'm waiting like everyone else in Europe. Muse and Dots didn't reach us either. GDPR is a hard bar for any company to clear, and hardest for big tech at scale. That's a trade-off we chose, and I understand it. I'd still like it to mean a shorter wait, because some of us are already deep in this ecosystem and ready to test it on real business work.

I've joined the waitlist. If it opens up, I'll report back on what a proactive Claude does to a one-person business. My bet: it gets my things done. The interesting part will be what happens when it has to get ours done.

### Sources & further reading

**External**

- [Claude Duet product page, Anthropic](https://claude.com/product/claude-duet?ref=livain.com)
- [Claude Duet announcement, Claude on X](https://x.com/claudeai/status/2112984669943968099?ref=livain.com)
- [Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet, 9to5Mac](https://9to5mac.com/2026/10/21/anthropic-takes-on-chatgpt-dots-and-meta-muse-with-claude-duet/?ref=livain.com)
- [Claude Duet: Anthropic's proactive agent app, AI/TLDR](https://ai-tldr.dev/releases/anthropic-claude-duet/?ref=livain.com)

**Related posts**

- [Claude just merged Chat and Cowork](https://www.livain.com/blog/claude-merged-chat-and-cowork/)
- ["The draft is ready" is not a status](https://www.livain.com/blog/draft-is-ready-not-a-status-agentic-routine-verify-own-claims/)
- [Meta's Muse, one week in](https://www.livain.com/blog/meta-muse-week-one-integrations-beat-model-quality/)
- [My scheduled Claude Code routine failed three weeks running](https://www.livain.com/blog/scheduled-claude-code-routine-failed-agentic-workflows-context/)