English
Español
Anthropic has quietly crossed a line the rest of the industry has only talked about. On October 21 it released a preview of Claude Duet, an AI agent that does not sit and wait for you to type a request. Built on Claude Opus 5.5, Duet is designed to look at the work in front of you, decide what matters, and start handling it on its own. For the first time, a mainstream assistant is acting more like a colleague who takes initiative than a chatbot that answers questions.
What Claude Duet actually does
With your permission, Claude Duet connects to the tools you already use, starting with your email. It reviews the messages you authorize it to see, works out which ones need action, and prioritizes them the way an assistant would. From there it can carry out tasks end to end, including multi-step ones. Anthropic’s own example is striking: Duet can book a flight and complete the check-in, not just draft an email about the trip. The key word is proactive. Instead of responding to a single prompt, it chains together the steps a real task requires and keeps going until the job is done.
Anthropic is not alone in chasing this idea. The company points to similar efforts from rivals, including OpenAI’s Dots and Meta’s Muse, as proof that the whole industry is racing toward agents that act rather than answer. What sets Duet apart in this preview is that it is built on Opus 5.5, Anthropic’s most capable model, and aimed squarely at everyday knowledge work rather than a narrow demo.
Why a proactive agent is a bigger deal than it sounds
For two years, “AI assistant” has really meant “very good autocomplete.” You asked, it answered. Duet flips that relationship. When software can decide what to do next and then do it, the bottleneck stops being how well you can prompt and starts being how much you are willing to delegate.
That shift brings an obvious tension: trust. Handing an agent your inbox and the authority to book travel is not a small thing. People will reasonably ask what happens when the agent misreads an email, acts on the wrong one, or takes a step you would not have approved. Anthropic’s “with your permission” framing is doing a lot of work here, and how much control users actually feel they have will decide whether Duet becomes a daily habit or a party trick. The promise is enormous time savings. The question everyone is circling is whether they are comfortable letting the software move first.
What it means for your business
If you run a team, Duet is a preview of where the next year of work is heading. The repetitive, multi-step chores that eat your week, triaging email, scheduling, booking, following up, are exactly what a proactive agent is built to absorb. The companies that win will be the ones that figure out which tasks they are willing to let an agent own, and which still need a human in the loop.
This is also a nudge to get your systems ready. An agent is only as useful as the tools it can reach, so clean data, connected apps, and clear rules about what it may and may not do are about to matter more than ever. The teams already building AI agents into their workflows will be the ones able to take advantage of Duet-style automation the moment it is reliable, instead of scrambling to catch up.
Claude Duet is still a preview, and Anthropic is right to roll it out carefully. But the direction is clear. AI is moving from a tool you operate to a teammate that operates alongside you, and the gap between the two is closing faster than most businesses planned for. The smart move now is not to wait for the finished product, but to start deciding what you would hand an agent today.
