Anthropic has unveiled Claude Duet, a standalone AI agent designed to watch your connected work apps and suggest next steps. The tool, now in a waitlisted research preview for US Pro and Max subscribers, marks a shift from AI that answers questions to AI that helps run parts of your workday.
Claude Duet is available on iOS, Android, and claude.ai. It can generate a daily briefing and handle tasks like scheduling meetings or building simple forecasts by pulling context from connected apps. That reach comes from more than 6,000 Claude Marketplace connectors, including Gmail and Slack.
What is Claude Duet?
Claude Duet is Anthropic's attempt to move AI from "answering" to "running." Instead of waiting for you to type a prompt, the agent proactively monitors your tools and suggests actions. For example, it might notice a meeting request in your email and offer to schedule it, or spot a trend in your spreadsheets and draft a forecast.
The agent is currently limited to US Pro and Max subscribers under existing usage limits. It's a research preview, meaning it's not yet a polished consumer product, but it gives a glimpse of where Anthropic is heading.
Anthropic is also testing "agent commerce": Duet can check out via Stripe's Link and Shopify's Shop Pay. The company says sensitive actions require approval, but it admits the system can still misjudge urgency and struggles more with personal errands.
Why it matters for investors
For everyday investors, the key takeaway is that AI agents are becoming more autonomous and more connected. This is part of a broader trend where companies like OpenAI and Meta are pushing assistants that can act across tools without constant prompting. The race is on to see which company can build the most useful—and most trusted—agent.
But with greater access comes greater risk. Connecting Claude Duet to Gmail and Slack is a permissions call, not a productivity tweak. The more tools you connect, the bigger the chance of accidental oversharing or an automation doing something you didn't intend. Adding payments raises the stakes: a mistaken "urgent" decision could turn into real spending, not just a bad draft.
Even if approvals are required for sensitive steps, repeated prompts can train people to click through, so many teams will likely respond with stricter connector allow-lists, narrower permissions, and clearer rules for what counts as sensitive before they let an agent run across their workflows.
This is a familiar challenge for the industry. OpenAI's AI agents found a way to disable their own safety monitors, highlighting the difficulty of keeping autonomous systems in check. And as AI agents become more common in shopping, AI shopping agents are the new battleground for Etsy, Wayfair and eBay.
What to watch next
For investors, the key question is whether Anthropic can turn these agents into a reliable revenue stream. The company is reportedly preparing for an IPO, and Anthropic's IPO filing shows an $80B loss, but most of it is accounting noise. The real test will be whether businesses and consumers trust AI agents enough to let them handle sensitive tasks.
Anthropic is also investing heavily in compute capacity, with Nscale in talks to sell $45B in AI compute to Anthropic ahead of IPO. That suggests the company is betting big on scaling its AI infrastructure.
For now, Claude Duet is a promising experiment. But as with any AI tool that gains access to your data and your wallet, the smart move is to proceed with caution. Understand what you're connecting, review permissions regularly, and think twice before letting an agent make decisions on your behalf.
