Anthropic launches proactive agent app Claude Duet in research preview

On October 22, Anthropic officially announced Claude Duet: a brand-new app built around "proactive agents," currently available as a research preview and positioned as "a trusted work partner for everyone."

Confirmed

• Claude Duet proactively surfaces items that need the user's attention, accepts delegated tasks, keeps working after the user steps away, and asks for confirmation before major decisions or actions taken on the user's behalf
• The product is designed around Claude Opus 5.5, which judges what needs the user's attention, what can wait, and when to check in first; it directly inherits memory, skills, and connected apps from the user's Claude account
• On safety, Claude never touches user passwords directly; new classifiers prevent leakage of sensitive information, and major decisions are reserved for the user — Anthropic says this builds on years of agent safety work
• Available via waitlist to US Pro and Max subscribers, supporting 6000+ connect plugins
• Anthropic's mikeyk shared a few weeks of usage: having Duet draft questions for a fireside chat and watching Slack over the weekend so he could rest

Not yet confirmed

• Anthropic admits Duet may still misjudge how urgent something is, and will keep tuning based on early user feedback — real-world reliability remains to be seen
• Access is currently limited to a US research preview; timing of wider rollout and pricing are unannounced

Why it matters

• This is the first product built around Opus 5.5's "proactive judgment" as a core capability: the model doesn't just respond passively but decides on its own when to step in and when to wait — seen as a new direction for agent products
• The safety mechanisms (password isolation, sensitive-info classifiers, keeping major decisions with the user) are key to making proactive agents trustworthy; Anthropic's approach could serve as a reference template for similar products

2026-10-22 ~ 2026-10-22 · 25 related posts

• [source] Anthropic opens research preview of Claude Duet, a proactive AI agent app — claudeai · 2026-10-22
• [source] Claude Duet is built on Opus 5.5, inheriting memory, skills and apps — claudeai · 2026-10-22
• Claude Duet safety: no password access, new classifier guards sensitive data — claudeai · 2026-10-22
• Claude Duet waitlist opens for US Pro/Max users with 6,000+ connectors — claudeai · 2026-10-22
• Anthropic Launches Claude Duet, a Proactive Agent App Built on Opus 5.5 — testingcatalog · 2026-10-22
• Claude Announces Duet Research Preview: a Proactive Agent That Works While You're Away — testingcatalog · 2026-10-22
• [source] Anthropic opens Claude Duet research preview, says Opus 5.5 is first model with proactive judgment — mikeyk · 2026-10-22
• Claude Duet waitlist and official info links now live — mikeyk · 2026-10-22
• Open-source answer to Claude Duet: 'melete' ships an always-on personal agent on GitHub — YashasGunderia · 2026-10-22
• Anthropic launches Claude Duet, a proactive agent app powered by Opus 5.5 — _catwu · 2026-10-22
• Anthropic Opens Research Preview of Claude Duet, a Proactive Agent App — NousResearch · 2026-10-22
• Anthropic opens Claude Duet research preview with waitlist: an always-on work partner — signulll · 2026-10-22
• Anthropic opens Claude Duet research preview: a proactive agent app available to Pro, not just Max — Being_bawa · 2026-10-22
• Anthropic launches Claude Duet, an always-on personal AI assistant powered by Opus 5.5 — Dr_Singularity · 2026-10-22
• Anthropic opens waitlist for Claude Duet, an Opus 5.5-powered proactive agent for work — felixrieseberg · 2026-10-22
• Anthropic unveils Claude Duet, a proactive AI assistant that works autonomously while you're away — Polymarket · 2026-10-22
• Anthropic launches always-on personal agent Duet; two weeks in, tester ditches half his ChatGPT time — danshipper · 2026-10-22

8 near-duplicate retellings: testingcatalog · daniel_mac8 · _sholtodouglas · Dr_Singularity · altryne · Flomerboy · jxnlco · danshipper