Anthropic’s Claude Duet Enters the Agent Race With a Work-First Pitch
Anthropic’s Claude Duet Enters the Agent Race With a Work-First Pitch
Anthropic launched Claude Duet on Wednesday, entering the increasingly crowded race to build AI agents that can keep working after a user steps away. The company’s wager is that professionals and small-business owners want an assistant that monitors their work, flags what matters and tackles tasks across connected apps—not simply a chatbot waiting for prompts.
Duet is powered by Anthropic’s Opus 5.5 model and is being released as a research preview through a waitlist. In a post announcing the app, Addy Osmani said it works across “6,000 connectors/plugins,” alerts users to what needs attention and “keeps going when you step away.”
The product’s pitch is deliberately proactive. Each morning, Duet produces a brief from connected applications; during the day, it watches for issues that require intervention. Users can hand it work ranging from meeting scheduling to dashboard-building, forecasting, prototypes and web apps.
That puts Anthropic alongside OpenAI’s Dots, Meta’s Muse, Google’s Gemini Spark and agents from startup Instinct. But Anthropic is drawing a line around its target customer: Duet is designed for work and is less capable at personal chores such as booking reservations, a contrast with the consumer-oriented framing of some rivals.
The company is also foregrounding control. Duet runs with a virtual machine and browser, but asks users for approval before consequential steps such as sending payments or deleting material. Anthropic acknowledges the early version can still “misread a thread or miscalibrate what’s urgent,” and says a dedicated classifier and safety prompt are meant to block requests involving highly sensitive data, including credit-card numbers.
For now, the launch is less a finished consumer breakthrough than a public test of a harder promise: an AI assistant that is useful enough to act, yet restrained enough to be trusted.
Write a comment