Claude Duet is the first personal-assistant-type harness that feels useful
Anthropic’s new Claude Duet personal AI assistant is drawing praise from some early users for feeling more genuinely useful than earlier “agentic” tools, thanks to features like project-oriented channels and strong performance likely backed by the latest Claude models. Others question whether its advantages are mostly interface polish that could be recreated with custom setups, or express reluctance to grant any cloud assistant broad access to personal data, money, or credentials given unresolved security and privacy risks. Comparisons with rivals such as Grok Bot, Muse, Dots, and DIY Hermes-style rigs highlight a split between users who value turnkey, hosted assistants and those who prefer self-managed, privacy-preserving alternatives.
Product clarity and documentation
- Several commenters were initially confused about what “Duet” referred to; they wanted an explicit first-sentence definition like “Anthropic’s new personal AI assistant.”
- Others argued niche blog posts can assume shared context and that missing definitions are acceptable for an insider audience.
Build-your-own vs. proprietary harnesses
- Many technically inclined users prefer rolling their own agents (e.g., Raspberry Pi setups, custom Swift apps, Hermes/Nanoclaw-style rigs) to avoid vendor lock-in and gain flexibility.
- Counterpoint: building and maintaining harnesses is itself work; some would rather spend attention on tasks than on infrastructure.
Comparisons with other assistants
- Duet is discussed alongside Grok Bot, Muse, Dots/Dot, OpenClaw/NanoClaw, Hermes, Instinct, and others.
- Some claim all the described Duet features already exist in Grok Bot or other tools; others say Duet feels like the first one that “clicks.”
- A recurring view: the underlying model quality (e.g., Opus vs Grok) is the main differentiator, not UI ideas, which competitors can copy quickly.
Perceived strengths of Duet
- Channels and subthreads seen as genuinely useful for organizing work.
- Many think it’s likely powered by a strong model (e.g., Opus 5.5), with concise, focused answers and transparent task views.
- Framed more as a project-management/work assistant than a life-automation bot, which some find more comfortable.
Limitations and missing features
- Complaints: lack of full voice mode/phone calls, no shell/desktop access like some rivals, difficulty with certain integrations (e.g., Dropbox, desktop workers), less “make it yours” config.
- Some users report it isn’t significantly faster than competitors.
- Availability constraints (U.S.-only, issues on GrapheneOS, no Teams support) frustrate potential adopters.
Privacy, security, and control
- Strong skepticism about giving any assistant broad access to identity, payments, or 1Password vaults; many design “low blast radius” setups (separate email, limited tokens, small balances).
- Discussion of the “lethal trifecta” (internet access, untrusted inputs, private data) and prompt injection risks; claims that newer models may mitigate this are met with doubt.
- Ideas surface for special “AI credit cards” or reservation flows requiring explicit user confirmation.
Use cases and broader attitudes
- Popular uses: coding, repo/project management, research and monitoring (e.g., bundles, events), news/library access, shopping/stock checks, travel planning, job search pipelines.
- Some only want simple STT/TTS, dislike anthropomorphic voice modes, and even prefer robotic voices.
- A portion of commenters feel agents are overhyped, privacy-invasive, or a minor convenience not worth the risk; others report “can’t live without it” experiences, especially with Grok Bot or custom setups.