SAnthropic just opened a waitlist for an AI agent that runs tasks on its own cloud computer.
SIt's called Claude Duet. You hand it work, it goes off to a virtual machine, and it emails you a daily brief.
ESo the AI does the chores while I check the highlights? I am listening.
SI knew you'd like this. Fady Eid posted on X about it — Anthropic's own page confirms the details.
EIt runs on Opus 5.5, their top model, and it's open to Pro and Max subscribers in the US.
SRight, but there's no general release date. This is a research preview.
EThat's the difference between a chatbot and an agent. A chatbot talks. This one acts.
SIt can click, open files, navigate apps — all without you watching.
EWhy does that scare you? You already trust it with your inbox, your calendar, your search.
SBecause my inbox can't buy anything or send a message I didn't write. This can.
EIt sends a daily brief, Sloane. You approve the work. You can turn it off.
SCan I? Anthropic says no release date, and it's still a research preview.
ECome on, that's unfair. They built it with controls. It's not a runaway agent.
SWe've seen models do unpredictable things once they can click around the web. One bad click is all it takes.
ESo you want an AI that just tells you what it would do, and you do it manually?
SI want it to ask before anything irreversible. That's not too much.
EThat's exactly the problem! If it has to ask for every click, it's not saving you any time.
SIt's saving you time until it's not. There's a reason this is waitlisted.
EFine. You can keep your paper calendar. I'll take the robot intern.
SNot a paper calendar. Just a healthy pause before the machine sends the email.
EOkay, that's one way to live. But think about the busywork — scheduling, monitoring, research.
EThis is about cutting the low-value stuff so humans do judgment. The summary is the point.
SOr it just raises the bar. Now you're expected to run ten agents and read ten summaries.
ENo, you're expected to make better decisions because you're not digging through spreadsheets.
SSpoken like someone who hasn't tried to cancel a subscription through a chatbot.
EWe're not talking about customer service bots. This is your own machine, your own files.
SFair point. If it stays in a sandbox and logs every action, maybe I'd relax a little.
ESee? That's all I'm asking. Sandbox, logs, daily brief. That's built in.
SMaybe the real test is what happens on day thirty, when you stop reading the briefs. Then it's just... running.
ESo the failure mode is human laziness, not the AI. That's a much older problem.
SAnd laziness plus autonomy is exactly the combination I worry about.
EThen build a daily brief that flags anything unusual in red.
SNow you're designing the product.
ESomebody has to. Besides, you're cautious and I'm an early adopter. It keeps things honest.
SIf my caution saves us from one accidental email, I'm saying I told you so.
SAnd on the complete opposite end of technology, a viral video claims a therapy horse plays piano to wake patients from anesthesia.
EOf course it does. Next it'll be doing their taxes.
SDigital Brain on Bluesky called this one out: real therapy horses help with balance and rehab.
ENot with Beethoven.
SThe piano part is likely staged or trained, and there's no credible evidence it wakes anyone from anesthesia.
EI choose to believe the horse has a side gig.
SMedical teams use sedatives and monitoring for that, not hooves on a keyboard.
ESomebody trained a horse to hit keys and called it medicine. That's a new one.
SThe internet is undefeated at turning a cute animal into a health claim.