$ today_in_10_seconds
Anthropic just put a proactive agent in your pocket. Claude Duet runs in the background on iOS, Android, and the web, and it reports back without waiting to be asked. It's waitlist-only for US Pro and Max users for now, but the bigger news is the interaction model.
$ lead_story
Anthropic unveils Claude Duet, a proactive background agent
Anthropic unveiled Claude Duet, an agent app that works in the background on iOS, Android, and the web. It launches as a waitlisted research preview, starting with US Pro and Max subscribers. The change is in who starts the conversation: Duet doesn't sit in a chat box waiting for a prompt. It runs on its own and brings you updates when it has something. That moves the hard problems away from model quality and toward trust, permissions, and deciding when to interrupt. For builders, the bar for agent UX just moved from 'answers well' to 'knows when to speak up and when to stay quiet.'
// why_it_matters: Proactive agents are going mainstream, and the competition is now about judgment as much as capability.
$ also_worth_attention
How Duet is being pitched: reports updates, works autonomously, takes tasks off your plate
TechFlow's coverage sums up Duet in three verbs: it proactively reports, it works autonomously, and it handles tasks for you. That's close to a spec for what users will now expect any background agent to do. The headlines say little about what Duet can access or how users approve its actions, and those details are what to watch as the preview opens up.
link →
$ bigger_picture
The trend behind the headlines
Until now, agents have mostly lived where developers live: terminals, IDEs, and browser extensions. Duet moves the agent loop onto the phone. There, background work shows up as notifications, and every notification costs the user some attention. A waitlisted, US-only research preview is a careful rollout. That suggests the open questions are about how the agent behaves in real use, not whether the model is capable enough. If proactive agents catch on, apps will compete on permission models and on how often they interrupt. Every agent builder will need a good answer to 'what does this do when I'm not looking?'
Model Context Protocol (MCP)
A background agent is only as useful as the systems it can reach, and MCP is an open standard for connecting agents to tools and data. Whether or not Duet uses it, an MCP server is the cheapest way to make your product something agents can act on.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime