Anthropic unveiled Claude Duet on 21 October 2026, a standalone AI agent app that acts as a proactive assistant for professionals and small-business owners, according to Reuters. Instead of waiting for a prompt, Duet delivers a daily briefing, monitors connected apps and surfaces items that need attention, in a direct shift from reactive chatbots towards software that watches work continuously and speaks first.
What happened
On 21 October 2026, Anthropic unveiled Claude Duet as a standalone app, according to Reuters' report of the launch. The company described Duet as a proactive assistant aimed at professionals and small-business owners. Its daily briefing is the centre of the product: rather than opening a blank chat window, a user starts with a summary of what has changed across connected services and what appears to need a decision.
According to Reuters, the tasks Duet is designed to handle include scheduling meetings, building financial forecasts and building web apps. Duet is built on the Opus 5.5 model, the company said, and it is available on iOS, Android and claude.ai through a waitlisted research preview. Access is initially for US subscribers on the Pro and Max plans, and Duet shares the existing usage limits on those plans rather than carrying a separate allowance, according to the reporting.
The integrations are a large part of the pitch. Duet integrates with more than 6,000 connectors in the Claude Marketplace, including Gmail and Slack, according to Reuters. It can also make purchases via Stripe Link and Shop Pay. That combination matters: an assistant that can read email and messages, watch a calendar, prepare a forecast and then complete a purchase is moving from advice into action, which is why the safeguards around it are central to the launch.
Anthropic said Duet requires user approval for sensitive actions, according to Reuters. The company also acknowledged limits plainly. Duet can misjudge urgency, the company acknowledged, and it is weaker at personal tasks such as booking tickets. That admission is useful context for a research preview: a proactive system that interrupts too often, or ranks the wrong item first, quickly teaches users to ignore it, which would defeat the purpose of the briefing.
Further detail on the launch and its positioning was also covered by TechSavvyEd's report on Claude Duet, which framed Duet as Anthropic's answer to the always-on agent products now arriving from rivals.
Why it matters
The important change is timing. A conventional assistant is useful only when a user remembers to ask it something. A proactive agent tries to close the gap between something happening in an inbox, a calendar or a project tool and the user noticing it. For a small-business owner, that gap is where missed replies, forgotten follow-ups and late decisions accumulate. A reliable daily briefing that ranks a small number of items well could be more valuable than a more powerful chatbot that still depends on the user to drive every interaction.
Context reported by Reuters shows this is now a competitive category, not a single company's experiment. According to Reuters, OpenAI launched always-on “dots” agents last month, and Meta's Muse AI agent drew millions of downloads. Three major developers pursuing proactive agents at once suggests the next contest in consumer and professional AI is not only about answer quality, but about which system earns permission to monitor work, interrupt at the right moment and act with approval.
That permission is the hard part. Connecting Gmail, Slack and thousands of other services, plus purchasing through Stripe Link and Shop Pay, asks users to trust one system with a wide view of their working life. Anthropic's approval requirement for sensitive actions is therefore not a footnote. It is the mechanism that decides whether Duet feels like a helpful chief of staff or an automated risk. The company's own caution about misjudged urgency points to the other trust test: accuracy of prioritisation, not just safety of individual actions.
What to watch
First, watch the waitlist. Duet is in a research preview for US Pro and Max subscribers, according to Reuters, so the early questions are how quickly access widens, whether it reaches other countries, and whether the shared usage limits feel generous enough for an always-on product. A proactive agent that consumes the same allowance as ordinary chat could push heavy users towards higher tiers or towards rationing the monitoring itself.
Second, watch the approval experience in real use. If sensitive actions arrive with clear context and a simple choice, professionals may accept delegation. If approvals are frequent and vague, the briefing becomes another inbox to clear. Third, watch whether Duet improves at the personal tasks Anthropic acknowledged are weaker, including booking tickets, because professionals rarely separate work and personal logistics as neatly as product categories do.
Finally, watch the rivals named in Reuters' context: OpenAI's dots agents and Meta's Muse agent. Download counts and launch coverage will matter less than retention — how many users still let an agent brief them each morning after the novelty fades. That is the test every proactive assistant now faces, and Duet has entered it with unusual candour about what it still gets wrong.
Sources: Reuters, 21 October 2026 and TechSavvyEd. Related reading: more coverage in our AI News section.
Be the first to share a useful question or thought.