In partnership with
What's Actually Happening
The rumor mill spent all day promising Fable 5.5. What Anthropic actually shipped is arguably a bigger change.
Claude Duet is a new app where Claude stops waiting for you. It connects to your inbox, calendar and work apps, sends you a short brief every morning, checks back every few hours, and messages you only when something needs your attention. You can also hand it whole tasks, and it works on them on its own cloud computer while you do something else.
Meta has Muse and OpenAI has Dots. Now Anthropic has its own always-on agent, and that means all three big labs have made the same bet: the next AI product isn't a better chatbot, it's one that comes to you first.
ARTIFICIAL INTELLIGENCE
🎻 What Claude Duet Actually Does
The daily loop:
☀️ Morning brief. Every morning, Duet summarizes what needs your attention across your connected apps and drafts next steps. Then it stays quiet.
🔔 Check-ins. It checks your apps every few hours and messages you only when something needs a human.
🧵 One thread. Everything happens in a single ongoing conversation instead of a pile of separate chats.
The bigger jobs:
🖥️ Its own cloud computer. Duet works in a virtual machine with its own browser, so it keeps going after you close the app. It never touches your own computer.
📋 Real work. Anthropic's examples include flagging unpaid invoices and drafting reminders, preparing payroll for review, getting contracts ready to sign, scheduling meetings, building dashboards and prototyping web apps.
🔌 6,000+ connectors. Gmail, Google Calendar, Slack, Notion, HubSpot, Shopify, Stripe, QuickBooks, Docusign, Gusto, Ramp, Microsoft 365 and more.
🧠 No setup from scratch. It inherits the memory, skills and connected apps already in your Claude account from the first message.
It runs on Claude Opus 5.5.
🔐 The Safety Design (This Is the Part to Read)
An agent that can read your email and touch Stripe is only useful if you trust it. Here's what Anthropic says it built:
✋ It asks first. Duet needs your approval before sending messages, making payments, signing documents or deleting files.
🙈 No passwords. Anthropic says Duet never directly sees your passwords or sign-in tokens.
🛡️ A dedicated classifier. A new classifier and safety prompt check every message and block sensitive data like card numbers and government IDs. Health and personal financial details only go out when you explicitly tell it to send them.
🔎 Tested from outside. Internal red teams and an outside security firm tested it, and it's covered by Anthropic's bug bounty.
And what Anthropic says it gets wrong: it can misread a thread, misjudge urgency, or miss something that arrives between check-ins. It's also weaker at personal errands like booking tickets.
That timing matters. Yesterday's lead story was attackers who left their agent dashboards open on the internet, leaking hundreds of passwords and API keys. Approvals and never seeing your logins are exactly the safeguards that story showed are needed.
285 Free Prompts to Save Time
Your inbox isn't the problem. Neither are your meetings. It's the hours you spend on work AI could do in minutes.
285 copy-paste prompts: built for email, meetings, admin, and delegation. Each one has a clear role, a specific task, and a defined output.
No setup. No prompt engineering. Just copy, paste, and get your time back. Free when you subscribe to Mindstream.
📱 Who Can Get It
🇺🇸 US only, for Pro and Max subscribers.
⏳ Waitlist. It's a research preview, and Anthropic will email you when you're in.
📲 iOS, Android and web. It's a standalone mobile app, plus claude.ai on a computer.
🏢 Team and Enterprise plans aren't included yet.
💳 No extra price. It uses your existing Claude plan limits, and its background check-ins count toward them.
🥊 Muse vs Dots vs Duet
All three labs now have an agent that works before you ask. They're aimed at different people:
🟣 Meta Muse | 🔵 OpenAI Dots | 🟠 Claude Duet | |
|---|---|---|---|
Built for | Consumers | Big companies | Solo pros and small businesses |
Lives in | Instagram, WhatsApp, Facebook | ChatGPT, Slack, Teams | Its own app plus claude.ai |
Status | Live, 14M+ installs (Apptopia estimate) | Beta for Enterprise, Edu, Healthcare | US waitlist, Pro and Max |
Duet sits in the gap between the other two, which is a smart place to be. Small business owners have the most scattered tools and the fewest people to watch them.
One more twist: some big sites are fighting back against this whole category. Amazon, Yelp, Delta and several banks are reportedly blocking third-party AI agents from their sites, so how far Duet can actually reach will depend on more than Anthropic.
🔓 Why It Matters
For you: if you run a small business or freelance, this is the first big-lab agent built for you specifically, rather than for an enterprise IT department or for teenagers on Instagram.
For your usage limits: a proactive agent uses compute even when you aren't using it. Every check-in is a model run, and that's why they count against your plan. Remember Character.ai: engagement in AI costs money, and an agent that's always watching is always spending.
For the industry: the question has moved from "how smart is it?" to "will you let it act for you?" Every lab can build a capable agent. The one that wins will be the one people trust with their inbox and their bank account.
Top 5 In AI Research 🔬
The stories moving fast beyond today's headlines:
📈 Anthropic's draft IPO prospectus reportedly shows $16.3B in first-half revenue: plus about $9B spent on compute and more than $600B in future infrastructure commitments.
⚖️ Anthropic pitches a federal AI safety bill: the government could seek to halt a model's release if it posed an imminent risk of more than 50 deaths or $1B in damage.
🦂 Claude ran a 64-hour protein-design campaign against scorpion toxin: 17 of 211 tested designs met the binding criteria. Anthropic is clear that neutralization hasn't been shown yet.
🧬 OpenAI is acquiring Orchestra Bio: OpenAI plans to pair the biopharma coordination platform with its Rosalind life-sciences tools to cover more of drug development.
🧠 "Neural Networks Are Already Clean": Nick Cammarata argues that neurons long thought to be messy may be far more interpretable than we assumed. He calls it a working hypothesis.
🛠️ Tools That Are Hot Right Now!
💾 OpenAI 12-hour prompt caching: shared API context can now stay cached for up to 12 hours, which cuts repeated-input costs for agents that run all day.
🎙️ Mistral Voxtral 3: a speech recognition and text-to-speech family with a voice-agent SDK, 50+ languages, and either API or self-hosted deployment.
🔗 Composio: 1,000+ integrations that handle authentication and permissions when your own agents need to act in outside apps.
What's The Recap?
For three years, using AI meant opening a box and typing. Duet, Dots and Muse turn that around: the AI opens the conversation, and your job becomes approving its work instead of asking for it. That's a bigger shift than it sounds. A chatbot that's wrong wastes a minute of your time. An agent that's wrong can send the email, pay the invoice or miss the deadline it was supposed to catch. So the features that matter most in Duet aren't the clever ones. They're the boring ones: it asks before it acts, it never sees your passwords, and Anthropic admits it can misjudge urgency. If you get in off the waitlist, connect one or two apps first, watch what it flags for a week, and widen its access only once it has earned it. Give an AI agent trust the same way you'd give it to a new hire.
What Do You Rate Today's Newsletter?
Stay building. 🤖