Claude Duet: Anthropic's Agent That Works While You're Away

Host: Hello and welcome to AI Shadow Daily. I'm your host, and today we have two stories. First, Anthropic's new agent, Claude Duet, which keeps working while you're away.

Co-host: And I'm the American co-host. Then we'll look at OpenAI bringing its models and Codex to Google Cloud, which is a pretty surprising pairing.

Host: We'll take our time on both, so let's start with Duet.

Host: According to Engadget, Anthropic announced Claude Duet on October 21st as a research preview. It's a new app where Claude works as a proactive agent.

Co-host: Okay, I want to slow down on proactive. Most chatbots just sit there until you type something. What's different here?

Host: Exactly that. An agent is software that takes actions for you, not just answers. Proactive means it starts the conversation. Anthropic says Duet comes to you with what needs your attention, and checks with you before it acts.

Co-host: So it's like an assistant who taps you on the shoulder, rather than waiting in the corner.

Host: A nice image. And it's built for professionals and small business owners, according to Anthropic, people with more on their plates than they can track.

Co-host: What does a normal day with it look like?

Host: Each morning it sends a short brief on the most important items across your connected apps. You can also tell it what you want to hear about and how often.

Co-host: Does it watch those apps constantly?

Host: No. It checks those accounts every few hours, because checking nonstop would use up your plan's limits much faster.

Co-host: That's a nice honest detail. Constant watching sounds great until you see the bill, or the usage meter.

Host: Quite. Anthropic also says it designed Duet around Claude Opus 5.5.

Co-host: Let's talk about doing things. If I'm not at my laptop, where does the work actually happen?

Host: Duet has a virtual machine with its own browser, so it doesn't need your laptop or phone to be on. It's a bit like a small computer in the cloud that belongs to your assistant.

Co-host: And how do I give it work?

Host: You describe a task in a short message and it works out which apps to use. It then notifies you when the task is complete or needs review.

Co-host: What kinds of tasks are we talking about?

Host: Anthropic lists scheduling meetings, building dashboards, financial or sales forecasts, and web apps.

Co-host: That's a wide range. How does it connect to all my tools?

Host: It can use over 6,000 connectors and plugins in the Claude Marketplace. And it carries over your memory, skills and connected apps from your Claude account.

Co-host: Give me a business example.

Host: It can check QuickBooks or Stripe for unpaid invoices and draft reminders. It can even prepare and run payroll in Gusto.

Co-host: Payroll. That's the point where my pulse goes up a bit. What about buying things?

Host: For purchases it can use Link, Stripe's digital wallet, which pays with a virtual card so your real card number isn't shared. It can also use Shop Pay from Shopify.

Co-host: So the obvious worry is an agent with access to money and messages. What stops it going wrong?

Host: The first layer is approval. Before it sends a message, makes a payment, signs a document or deletes anything, it waits for your approval.

Co-host: Is anything simply forbidden?

Host: Yes. Trading securities and moving money between accounts are off-limits even with approval. And Duet never directly sees your passwords or sign-in tokens.

Co-host: What about private data leaking out in a message?

Host: Anthropic released a new dedicated classifier and safety system prompt that checks each message. They're designed to reject requests to share the strictest tier of information, including credit card numbers and government identification.

Co-host: And a classifier is basically a small checker model that flags risky content, right?

Host: Right, a second pair of eyes. Health or personal finance details go through only when you explicitly instruct it.

Co-host: Has anyone tried to break it?

Host: Anthropic says red teams and an external firm penetration tested Duet. It also says it will share more on the safety design in the coming days.

Co-host: Which means we should read that follow-up carefully when it lands.

Co-host: Okay, can I use it today?

Host: Access is by waitlist, for Pro and Max subscribers in the US, according to MacRumors.

Co-host: And Team and Enterprise?

Host: Those plans aren't included yet, says The Decoder. Anthropic says that during the preview it is included in those plans and uses the same usage limits as the rest of Claude.

Co-host: What about the price later?

Host: Fortune reports that Anthropic hasn't disclosed what it will charge after the preview. Background checks count toward the limits, and larger tasks like building a financial model or a deck need more model turns.

Co-host: So a heavy user could burn through their allowance quickly. Now, this isn't the first agent of its kind, is it?

Host: No. Fortune notes Meta launched Muse on September 8, and OpenAI followed with Dots on September 29. Engadget calls Duet Anthropic's answer to both, and said Anthropic is late to the party.

Co-host: And Muse has done well.

Host: CNBC says it rocketed to the top of Apple's App Store. Zuckerberg told analysts that more than 10 million people in the US and Canada have set up Muse agents over the past month and a half.

Co-host: Fair to say Anthropic is playing catch-up on attention, at least.

Host: Though to its credit it's candid. It admits this is an early version that might misread a thread or miscalibrate what's urgent. Personal errands, such as booking tickets, aren't quite as strong, and it can't yet show you its browser.

Co-host: And it says it will keep tuning Duet based on what early users tell it.

Host: On to our second story, and it's a surprise. On October 21st, OpenAI and Google Cloud announced a partnership.

Co-host: Google hosting OpenAI. What exactly is coming?

Host: OpenAI's frontier models and Codex agentic workflows are coming to Google Cloud's Model Garden. That's essentially a catalogue where cloud customers pick models to build with.

Co-host: Which models?

Host: Three, rolling out in private preview: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, according to Securities.io.

Co-host: How will companies pay for them?

Host: Through Google Cloud Marketplace, on a single invoice that draws down existing cloud commitments.

Co-host: That matters more than it sounds. A company that already promised Google a certain spend can use it on these models, rather than opening a separate contract.

Host: Sensible reading. And Microsoft remains OpenAI's primary cloud partner.

Co-host: When can people get it?

Host: General availability is planned in the coming weeks, according to AlphaSignal, though no launch date has been given. So watch that space.

Host: Time to wrap up. I'll be watching Anthropic's promised safety write-up on Duet. An agent that handles payments lives or dies on trust.

Co-host: I'll be watching the Google Cloud rollout, and whether a launch date appears. Pricing for Duet is my second one.

Host: That's the show. The show notes list every source, and this episode was made with AI voices. Follow AI Shadow for the daily videos, and we'll see you tomorrow.