Cover Claude, a generative AI assistant developed by Anthropic (Photo: Getty)
Claude, a generative AI assistant developed by Anthropic (Photo: Getty)
1 / 1
Anthropic’s standalone agent, Claude Duet—billed as a highly “proactive” agent that can anticipate tasks—is the newest competitor to advanced generative AI tools, including OpenAI’s Dots, Meta’s Muse and Google Gemini’s Spark
Generative AI tools in the workplace often serve as hardworking interns who can execute tasks when prompted—and, as is to be expected of an intern, the output often includes mistakes. The arrival of San Francisco-based tech giant Anthropic’s Claude Duet, an autonomous software agent powered by the Claude Opus 5.5 model, is marketed as a higher-level hire. Billed as a “digital chief of staff”, the tool rolled out by the makers of the popular chatbot Claude is designed to anticipate tasks and workloads, operating from its own dedicated virtual cloud—running background tasks, monitoring schedules, and even reaching out to the user to flag pending items or issues without being prompted.
How well it works, though—particularly in Hong Kong, where it’s only available through VPN networks and other workarounds—depends a lot on the user, and it faces fierce competition in the market for AI agents. Before Claude Duet's official debut, Tatler took a closer look at Claude Duet to answer burning questions about the new, higher-level era of enterprise computing.
What exactly is Claude Duet?
Claude Duet is an AI agent—referring to a generative AI tool that can execute tasks autonomously, without being constantly guided or prompted—that runs on an independent cloud, allowing it to work continuously in the background without requiring huge amounts of space on a personal computer. Instead of cramming every conversation into one endless messaging thread, the workspace divides tasks into organised project folders, much like office communication apps. A user can set up one dedicated channel for incoming client queries, another for flight bookings and a third for project research. Because the assistant connects directly to main account memory, it remembers ongoing goals and documents without requiring users to repeat background explanations.
The software is currently offered as a preview for users paying upwards of US$20 monthly for Anthropic’s Pro and Max subscription plans.
How does Claude Duet’s “proactive” check-in feature work?
For an agent to be “proactive”, as Anthropic describes, it essentially performs scheduled background monitoring. Rather than remaining inert until a user enters a query, Duet is conducting background sweeps every three hours by default across email systems, calendars and other software linked to its platform. According to Anthropic, Claude Duet can connect to 6,000 of these apps and tools, such as Gmail, Slack, QuickBooks, Gusto, Stripe and LinkedIn.
What this looks like in practice is that Duet can send morning briefings before the working day begins, flag upcoming deadlines, help initiate and draft replies to routine email correspondence, and also identify delays to progress on tasks. When events unfold—such as an urgent client message or an impending calendar clash—the system crafts a message and texts, or “pings”, the user directly. On the backend, its connections made through a Model Context Protocol (MCP) tool allow the software to gather information and assemble data points into predictions—or in human terms, “learning” about its user’s schedule so that it can anticipate upcoming events, needs or behaviours.
Is Claude Duet safe? Features for “human approval”
Automating computer tasks carries natural concerns about mistakes, which is why Anthropic has fitted the software with strict safety catches. Duet is programmed to pause and ask for explicit human permission before completing any major action, such as sending an external email, signing an agreement, deleting a file, or making a card payment through services like Stripe or Shopify. When a financial transaction or important message is ready, the system stops and sends an approval alert to the user’s phone. The software’s internal rules also bar it from buying or selling shares on financial markets—so, no risk of going bankrupt thanks to Duet’s stock-picking ideas—and it cannot share sensitive personal details such as passport numbers, according to Anthropic.
Read more: Meet Randy Lai: the McDonald’s Hong Kong CEO on leveraging AI for fast food experiences
How does Claude Duet compare with rival agents?
Major technology firms are each pursuing distinct approaches to automated working tools which are “always on”—meaning, they can monitor and access your information across different channels without requiring oversight to be switched on or off. While they offer similar support in multi-task operations and automation, they differ in the chatbot provider and services—as well as price point.
OpenAI offers Dot, an autonomous assistant embedded within ChatGPT that costs US$100 monthly and uses the GPT-6 Astra model to run background research and schedule tasks; Google provides Gemini Spark for US$19.99 monthly, integrating closely with its existing email, document and calendar applications. Meta focuses on mobile phones and hardware with its Muse assistant—just launched in Singapore—which ranges from free basic tiers to US$100 monthly, while developer platform Cursor and Elon Musk’s xAI offer Grok Bot at prices between US$20 and US$200 per month. Duet differs by combining an independent cloud workspace with organised project channels at a familiar US$20 monthly entry point.
What is driving autonomous agent adoption? Looking beyond conversational AI
Widespread interest in independent software tools may reflect a level of fatigue with everyday digital clutter and AI agents offering quick, high-volume, but inaccurate output. Professionals spend large parts of their week jumping between messaging channels and overflowing inboxes, creating administrative fatigue; and while first-generation AI tools offered clever writing advice, they still demanded constant supervision, essentially adding yet another open tab to an already crowded screen.
Independent agents alter that balance by handling coordination in the background, assembling data while their human counterparts focus elsewhere, and only asking for input where needed. Their more advanced models work to make fewer mistakes, or flag areas where errors may occur—such as Claude Duet’s “pings” for human review in sensitive transactions—to avoid issues. With open-source developer alternatives like OpenClaw alongside proprietary offerings from Anthropic and OpenAI, technology companies are increasingly betting that proactive digital assistants will become the standard, though the future norm lies behind a paywall for now.
Cat Wang is an editor at Tatler Asia based in Hong Kong, where she covers business, wealth, innovation, and impact. Previously, she was a Forbes Asia reporter and assistant editor of the Forbes Asia 100 to Watch list, spotlighting startups and small companies on the rise across the Asia-Pacific. She began her career as a trainee reporter at the South China Morning Post, Hong Kong’s newspaper of record. Born in New York City, she graduated from the University of California, Los Angeles, with a B.A. in political science.
Curated for you
The stories, profiles, and cultural moments featured on our Instagram feed
