Traditional use of AI in the past few years has largely meant just telling it what to do and firmly being in the driving seat. But Anthropic’s newly announced Claude Duet wants to change that. Described as a proactive “always-on” AI agent, it can identify what needs your attention, complete tasks, and continue working even when you’ve stepped away.
With OpenAI’s Dots and Meta’s Muse exploring similar territory, it’s become clear that some of the biggest names in tech see proactive AI as the next frontier. But how much responsibility are we really prepared to hand over? And what happens when these tools get things wrong?
What Is Claude Duet?
The easiest way to really understand Claude Duet is to compare it to a traditional chatbot. Rather than waiting for you to ask a question, Duet checks your connected applications every few hours and identifies things that might need your attention. It can provide a morning briefing, flag outstanding tasks, and carry out work ranging from scheduling meetings to building dashboards.
Powered by Claude Opus 5.5, Duet uses the memory, skills, and connected applications from your existing Claude account. It also operates through its own virtual computer, meaning it can continue working even when your laptop is switched off.
Vitally, sensitive actions such as sending messages, making payments, or deleting information still require your approval. The tool is currently available through a research preview waitlist for US-based Pro and Max subscribers.
What’s most interesting is how accessible this could make AI agents to non-technical users. You don’t really need to understand complex workflows or spend time refining prompts. For a small business owner juggling customer emails, invoices, and meetings, the idea of an assistant that recognizes what needs doing could be hugely appealing.
However, there are certainly some concerns around working with always-on agents. While Anthropic’s strong reputation among AI enthusiasts could help attract early users, there’s a significant difference between trusting AI to answer a question and trusting it to make decisions on your behalf.
Could Always-On AI Create More Work Than It Saves?
An AI assistant that never switches off sounds like a productivity dream on paper, but introducing an always-on agent into a business is very different from asking ChatGPT to draft you an email. Could tools designed to reduce workloads end up adding to them?
One concern is false positives. Imagine an agent repeatedly flagging routine emails as urgent, identifying problems that don’t exist, or requesting approval for unnecessary tasks. If employees spend their time checking, correcting, and dismissing AI-generated suggestions, you have to question how much time is really being saved.
Accountability is also another issue. Anthropic has built approval mechanisms and safeguards into Duet, but mistakes remain possible. If an agent misreads a sensitive customer exchange or mishandles information, the consequences don’t fall on the AI – employees and businesses will be responsible for how they use it. In an enterprise environment, one mistake is one too many, and a single serious error could significantly damage trust.
For Salesforce pros, there’s a familiar theme here. Salesforce has spent two years promoting autonomous agents through Agentforce, as well as the recent introduction of Coworker. But Duet takes a more personal approach, working across an individual’s applications rather than a single business process.
Imagine a Salesforce Admin using a proactive assistant to flag unresolved support tickets, summarize Slack conversations before a release, or spot potential project delays. These are hypothetical examples, but the appeal is obvious for professionals already stretched across multiple responsibilities.
However, another AI system also means considering permissions and oversight. Someone needs to know what it can access and when to intervene. It seems the next challenge for Salesforce professionals is learning how to supervise agents rather than just knowing how to use them.
Join the Waitlist
As Claude Duet is in early research preview, if you want to get your hands on the technology, it’s recommended that you get on the waitlist, which you can find here.
Interestingly, one of the questions on the form is “What do you do, and what would you hand off to Claude Duet first?”, prompting you to not only evaluate your day-to-day workflows and processes, but also indicating that Anthropic is eager to gather more data on Claude’s use cases.
Final Thoughts
There will undoubtedly be people who are excited about Claude Duet, as well as the competitor tools emerging. There are also others who find the idea of an always-on AI assistant slightly terrifying. Both reactions are understandable, especially when we’re talking about technology that can carry out work without being explicitly asked.
I do find it interesting how quickly the conversation around AI has shifted. The once-dominant fears of AI eventually destroying humanity seem to have taken a back seat to a more immediate question – how much of our work will AI end up doing for us?
Personally, I think proactive agents are a fascinating development, but an enormous amount is riding on their reliability. Anthropic may have earned the trust of many AI users, but that trust will be tested when its technology starts taking action rather than simply offering advice.
