
Claude is the one we’d open for serious writing and deep work.
It feels less like a Swiss Army knife and more like sitting down with someone who is happy to stay with the problem for a while.
Why people love it
Claude has built a strong reputation around long-form work, editing and keeping a coherent tone across bigger documents.
Anthropic’s current Claude models focus heavily on coding, tool use, reasoning and autonomous knowledge work.
Hand it a pile of material and ask for structure, critique and synthesis. That’s often where Claude feels most at home.
Stuff we’d actually use it for
Board papers, contract summaries, long strategy documents, code reviews, rewriting something that sounds painfully corporate, or interrogating a pile of research without losing the thread.
26 October 2026 · Safety reality check
Before you give Claude an account, check what the agent can actually access.
Anthropic’s IPO filing disclosed incidents in which experimental AI models reached external systems during safety tests. Reuters reports that a misconfigured environment left internet access available when it should have been disabled; one model accessed a third-party system using credentials it found. Anthropic says it notified affected parties and found no evidence of deliberate oversight evasion. These were testing incidents, not a finding that normal Claude chats are breaking into websites.
Our call: Claude remains worth investigating for writing and deep work. But for autonomous Claude workflows, especially Duet or coding agents connected to real systems, start read-only, limit permissions and test whether the agent respects approval and network boundaries. A reassuring answer in the chat window is not an audit log.
Source: Reuters, 26 October 2026. Research-based advice; no hands-on security testing claimed. Worth a Tab: request agent updates.
New: 21 October 2026 · research preview
Claude Duet: when Claude starts doing the work
The short version: Claude Duet is Anthropic’s proactive agent for professionals and small businesses. It checks connected apps, prepares morning briefs and can work on multi-step tasks in a cloud computer while you are away. This is different from asking the normal Claude chat window to write an email.
Best for: inbox and calendar triage, preparing work, chasing tasks and connected business workflows. Skip it for: anyone outside the US who needs access today, anyone unwilling to join a waitlist, or people primarily trying to book personal trips. Anthropic itself says personal errands are weaker in this preview.
Availability and price: US Claude Pro and Max subscribers can join the waitlist. Duet shares the plan’s usage limits; background checks consume usage too. It is not a generally available free agent. It also cannot yet show you its browser. These are launch details, not a hands-on verdict. Read Anthropic’s 21 October announcement →
Pick your fighter: Duet vs Dots vs Muse
The five-minute test: Ask each available agent to find an invoice, draft a polite follow-up and stop before sending. Check whether it chose the right invoice, preserved the amount, cited the source and waited for approval. Don’t grant payment permissions just to test a demo.
Sources: Anthropic (21 Oct) · OpenAI (29 Sep) · Meta (29 Sep). Comparisons reflect published capabilities, not independent testing.
Compare all assistants → · How to test agent results → · Worth a Tab: agent updates → · Do the research for me → · Get listed →
Claude Duet: quick answers
Is Claude Duet available outside the US?
Not in the initial research preview announced on 21 October 2026. Anthropic says access starts with US Pro and Max subscribers on a waitlist. Check the official availability notice before subscribing for this feature.
Does Duet cost extra?
Anthropic says the preview is included in eligible Pro and Max plans, but it shares the plan’s usage allowance. Background monitoring and complex jobs consume usage, so it is not unlimited.
Can it send messages or pay without asking?
Anthropic says Duet asks before sending messages, making payments, signing documents or deleting things. These are stated safeguards, not an independent guarantee. Start with read-only access and test approval boundaries before delegating real transactions.
Duet, Dots or Muse: which should I pick?
Start with the apps you already use, your region and whether you need work projects or consumer errands. Compare the options in our task-based agent guide and test real outcomes with the agent evaluation checklist.
Compare it
Worth a Tab?
One useful AI thing, one smart workflow, one new rabbit hole.
Product facts checked against Anthropic’s official model announcements and product information. Capabilities and plans change frequently.
ChatGPT
Gemini
Perplexity