Claude Duet

Original Article ↗ Hacker News Discussion ↗

Persistent AI “assistant” agents like Anthropic’s new Claude Duet are emerging as competitors to OpenAI’s Dots, Meta’s Muse, Grok Bot and various self‑hosted tools, promising to proactively handle email, scheduling, shopping, travel, finances and other digital chores. Commenters are split between enthusiasm for offloading tedious logistics — with some describing complex real‑world workflows already run by agents — and deep concern about security, privacy, platform lock‑in and overreliance on opaque, always‑on services wired into everything from email to password managers. Many also see signs of commoditization and hype fatigue: the core tech feels genuinely powerful, but the products look increasingly similar, trust is a major bottleneck, and some users prefer local or open-source setups that keep long‑term control of data and automation logic.

Product positioning & comparisons

  • Duet is seen as Anthropic’s answer to OpenAI Dots, Meta Muse, Grok Bot, Instinct, etc.
  • Runs on Opus 5.5; several argue model quality alone could differentiate it from Muse/Dots where outputs feel weaker.
  • Others say it “doesn’t really differentiate” at the product level and simply repeats the “persistent agent” pattern.
  • Some question why it’s a separate app/UX instead of just a built‑in Claude capability, attributing this to “shipping the org chart” and Conway’s Law.

Use cases & reported value

  • Common assistant tasks: email triage, scheduling, calendar management, travel logistics, reservations, grocery ordering, subscription/bill monitoring, recurring chores.
  • Several users describe powerful real‑world flows:
    • Buying a car and even a house abroad (language barrier, complex logistics) largely handled by an agent.
    • Continuous monitoring of accounts, catching unexpected bills.
    • Meal planning, grocery carts, apartment searches, contract workflows, and 311 reports.
  • Others run long‑lived agents (Hermes/OpenClaw/muse-like setups) that: watch IMAP mail, fill databases, auto‑file train compensation, restock groceries, monitor travel, and manage home automation.

Skepticism, limitations, and UX issues

  • Many still “don’t see the point” over a strong one‑off Claude/ChatGPT session plus some scripts or routines.
  • Complaints about: slow or brittle autonomy, poor coding performance, website blocking/CAPTCHAs, over‑eager notifications, and agents that still need constant prompting.
  • UX critiques: clunky waitlists, poor in‑app search of old chats, weak cross‑product integration, and generally half‑baked flows compared to what the underlying models could enable.

Privacy, security, and control

  • Strong concern about giving an always‑on, hosted agent access to everything (email, passwords, finances, calendars).
  • 1Password integration is polarizing: some like scoped vaults and wrappers; others see it as “the mother of bad ideas.”
  • Some argue risks are comparable to long‑standing Gmail/Dropbox exposure; others see persistent agents as a qualitatively worse privacy and security nightmare.

Local/self‑hosted alternatives

  • Multiple commenters prefer self‑hosted agents (Hermes/OpenClaw + local or cheap cloud models) on home servers or tailnets.
  • Benefits cited: owning memory/state, easier model switching, better transparency, and less data exposure.
  • Downsides: weaker models, more setup effort, and missing some of the polish of big‑lab products.

Market dynamics & AI fatigue

  • Many expect numerous venture‑backed agent startups to die or be acquihired as labs bundle similar functionality “near free” atop their own models and compute.
  • Debate over moats: compute, models, enterprise integration, and ecosystems vs thin B2C agents.
  • Several express fatigue with constant AI/LLM launches on HN, though others see this as the natural focus given AI’s perceived career impact and technological significance.