Anthropic just launched the most ambitious version of Claude yet: a proactive, background-running AI agent that monitors your work, surfaces what needs attention, and keeps completing tasks while you sleep. Claude Duet went into research preview for US Pro and Max subscribers on October 22, 2026 — and it changes the basic contract between humans and AI assistants.
For three years, the dominant metaphor of AI assistance has been the prompt box. You open a chat, you type, the AI responds. The exchange is transactional and synchronous: you must be there to start it. Every product from ChatGPT to Gemini to earlier versions of Claude followed this model. The user initiates; the AI reacts.
Claude Duet breaks the model. It doesn’t wait.
What Claude Duet Actually Does
Released as a research preview on October 22, Claude Duet is a new application — available on iOS, Android, Mac, and Windows — that keeps Claude working between and during conversations. The key behavioral difference is directionality: Claude Duet can come to you rather than waiting for you to come to it.
According to Anthropic’s announcement, Claude Duet does four distinct things that prior Claude releases could not:
It surfaces what needs your attention. Rather than requiring you to ask “what should I be working on,” Claude Duet monitors the tasks and context you’ve shared with it and proactively alerts you to things that need a decision, a reply, or an action. Think of it as a chief of staff that reads your situation and tells you what’s urgent — before you ask.
It completes tasks you assign it. This isn’t new for AI, but the scope is larger. You can assign a task and disconnect. Claude Duet will work through it — drafting, researching, analyzing, scheduling — and have results waiting when you return.
It keeps working when you step away. Unlike a standard chat session that ends when you close the window, Duet maintains continuity. Work that starts in the morning can resume in the afternoon even if you never reopened the app. Tasks persist, context is maintained, and progress doesn’t evaporate when you’re offline.
It checks with you before acting on your behalf. This is the safety valve. Claude Duet will not fire off emails, schedule meetings, or commit to external actions without a human confirmation step. It flags the action it wants to take, explains its reasoning, and waits for a thumbs-up. Anthropic is describing this as “human-in-the-loop by default.”
Underneath, Claude Duet runs on Opus 5.5 — Anthropic’s current highest-capability model, the same one powering the most complex enterprise deployments.
The Race to Build the Ambient AI Layer
Claude Duet arrives in a field that got very crowded very fast.
OpenAI launched ChatGPT Dots earlier this year — a conversational agent capable of maintaining persistent tasks across sessions, watching your calendar, and checking in proactively. Meta shipped Muse in September, positioning it as an always-available creative and productivity collaborator that learns your preferences over time. Google’s Gemini has been building toward ambient assistance through its deep Android integration, surfacing suggestions before they’re requested.
The battleground isn’t model quality anymore. GPT-6, Gemini, and Claude are all capable enough that users can’t tell the difference on most tasks. The competition has moved to presence and persistence: which AI assistant is actually there when you need something — and smart enough to know what you need before you articulate it.
Anthropic’s positioning with Duet is deliberate. The “checks with you before it acts” framing is a direct response to the anxiety that accompanied early proactive AI deployments — the concern that agents moving autonomously through your calendar, inbox, and documents are a privacy and control disaster waiting to happen. Duet is trying to be proactive without being presumptuous.
Whether that balance works in practice is what the research preview is designed to test.
Who Gets It and What It Costs
Right now, Claude Duet is available exclusively to US Claude Pro and Max subscribers. Pro subscribers pay $20/month; Max subscribers pay $100/month. The research preview designation means Anthropic is being explicit that the product is incomplete — it expects to surface rough edges and will use feedback to shape the general availability release.
The decision to restrict to paid subscribers is pragmatic. Proactive agents that run continuously in the background are far more computationally expensive than reactive chat sessions. A model that’s maintaining context, checking for relevant triggers, and preparing to ping users isn’t sitting idle between prompts — it’s actively consuming inference compute. Anthropic needs the economics of usage to make sense before opening this to free users.
International availability is listed as “coming soon,” though no specific timeline has been given.
What This Means for the AI Landscape
Claude Duet is the clearest signal yet that Anthropic views the future of Claude not as a chat interface but as a long-running AI colleague. The terminology in Anthropic’s messaging is telling: “proactive agent” rather than “assistant,” “comes to you” rather than “responds to you.”
This matters for enterprises in particular. The organizations most likely to get business value from AI aren’t the ones with employees who remember to open a chat window — they’re the ones where AI is embedded in workflows deeply enough that it’s monitoring, completing, and escalating without requiring a human to initiate every interaction. Claude Duet is Anthropic’s first real step toward that version of AI deployment.
For individual users, the more immediate question is trust. Giving an AI agent license to work in the background, to monitor context continuously, and to proactively interrupt you requires a level of confidence in the agent’s judgment that most users haven’t been asked to extend before. Anthropic’s “checks before acting” design is clearly engineered to earn that trust incrementally — but it’s a new psychological contract with AI, and not everyone will be comfortable signing it immediately.
The research preview will test exactly where that line is.
The Bigger Picture
Claude Duet launches the same week Anthropic’s IPO S-1 is circulating on Wall Street and the company’s AI agents are officially outworking its human employees. The company is, in a very literal sense, eating its own cooking — and now asking users to bring AI into the background layer of their own working lives.
The bet is that ambient, proactive AI is more valuable than reactive AI. History suggests that’s probably right. The most useful technologies become invisible infrastructure that works whether or not you’re paying attention to them. Email servers don’t wait for you to open a client before they receive your messages. Your calendar app doesn’t require you to ask if you have conflicts.
Claude Duet is Anthropic’s first attempt to make Claude that kind of ambient presence. The research preview starts today. The general availability version — and what it does to how people think about working with AI — will be the story that follows.
Claude Duet is available as a research preview to US Claude Pro and Max subscribers via the Claude app on iOS, Android, Mac, and Windows. Anthropic expects to expand availability internationally in the coming months.
