Why this is in the vault
Every's CEO reviews Anthropic's Claude Duet — a first-party, always-on personal-agent app launched today in research preview — and calls it the first personal agent he'd rate a "pro tool," with named strengths and gaps that map directly onto the founder's own self-built always-on agent.
The core argument
Claude Duet launches today in research preview: its own app (separate from Claude), running on Opus 5.5, pitched by Anthropic as a chief of staff for knowledge workers. Dan Shipper used it daily for two weeks alongside Dot, the ChatGPT agent he'd relied on, and reports it moved him to roughly 50/50 between ChatGPT and Claude. Duet is "the first personal agent I'd call a pro tool": it sends suggestions he actually wants, gives recurring work its own channels, and keeps a visible list of what it's doing and what it needs from him. It's also unfinished — it can see the Claude sessions running on his computers but can't direct them, its email drafts are worse than Dot's, and there's no real voice mode yet. Two anecdotes from Every colleagues: designer Tyler Nishida's Duet checked his home solar battery before answering an unrelated air-fryer question (ambient over-reach as much as ambient usefulness), and engineer Kieran Klaassen says flatly "I don't trust it yet" about Duet's link to his machines. The free preview's bottom line: Claude subscribers doing knowledge work should join the waitlist now; the paywalled full review covers specific email failures, a side-by-side against Dot, and a verdict on who should switch today. Format note: this is Every's "Vibe Check" product-review series, same slot as the Sonnet 5.5 and GPT-6 Astra reviews already in the vault.
Mapping against Ray Data Co
Strong mapping. The single most concrete connection: the founder already runs the category of product this review is assessing — see [[2026-10-25-always-on-personal-agent-product-category]] for the cross-source concept note on the category's converging design bar. The Mac Mini always-on channels-agent ([[project_channels_agent_setup]]) is a self-built, always-on personal agent reachable via iMessage, persisting across reboots with a daily restart — the same job description Anthropic is now shipping as a first-party consumer app. Shipper's "pro tool" bar — recurring work gets its own channel, a visible list of what it's doing and needs, suggestions worth acting on — is a usable checklist against the channels-agent's current design, and it exposes a real gap: there's no founder-visible "what I'm doing / what I need from you" surface beyond ops-log.md and ad hoc iMessage asks, whereas Duet ships that as a first-class UI element from day one.
Second, Duet's named weakness — it can see the Claude sessions on Shipper's computers but can't direct them — is the exact axis where the channels agent is already ahead: it isn't an observer bolted onto separate Claude sessions, it IS the acting agent, directed in both directions over iMessage. Worth stating plainly rather than assuming parity: on the observe-vs-act axis, a self-hosted setup built months ago already clears a bar Anthropic's first-party product hasn't reached at launch.
Third, this extends the standing Every/Anthropic infrastructure-handoff thread ([[2026-10-08-every-handed-infrastructure-to-anthropic]]): Anthropic is now shipping both the managed-infrastructure layer teams like Every build agents on top of and a first-party consumer agent app that competes with exactly the kind of personal-agent product a harness-engineering/agent-deployer positioning depends on differentiating from. Fourth, Kieran Klaassen's unhedged "I don't trust it yet" about machine access is a concrete, named-engineer data point on the trust gap in always-on agent adoption — directly relevant framing for any phData-style enterprise pitch that leads with "give an agent your machine." Fifth, a genuine tension worth flagging rather than resolving: Every's own prior reporting ([[2026-05-15-every-team-agents-vs-personal-pets]]) concluded personal "pet" agents don't scale at the team/org level — but Duet is explicitly an individual knowledge-worker chief-of-staff agent, not a team agent, so this isn't a contradiction of that finding so much as a reminder that "personal agent for one knowledge worker" and "shared agent for a team" are different deployment shapes with different scaling verdicts, and the channels agent (built for one founder) sits squarely in the first category.
Related
- [[2026-10-08-every-handed-infrastructure-to-anthropic]]
- [[2026-05-15-every-team-agents-vs-personal-pets]]
- [[2026-04-15-every-claude-managed-agents-mini-vibe-check]]
- [[2026-09-28-every-vibe-check-sonnet-5-5]]
- [[project_channels_agent_setup]]
⚠️ Sponsorship
Clean third-party paid block: Attio ("Introducing Attio: the agentic CRM," free signup CTA), plus the standard "Want to sponsor Every?" house solicitation line and the usual house footer (Every All Access subscription pitch, Sparkle/Cora/Spiral/Monologue cross-promo bundle). No disclosed relationship between Attio and the review's subject matter (Duet) — the sponsor is unrelated to the content, consistent with Every's prior one-off/rotating sponsor pattern (Brief was the only other Every sponsor logged to date, 2026-08-20) rather than evidence of a standing relationship. Separately, the review itself carries a structural incentive worth naming even without a paid disclosure: Every's own business model (its Vibe Check franchise, its AI-tool bundle, its "join the waitlist" CTA) benefits from positioning its CEO as the authoritative early reviewer of Anthropic's newest consumer product — treat the enthusiasm in the free preview as genuine but not disinterested.