Every AI tool you’ve used so far works the same basic way. You type, it answers, and then it sits there until you type again. Proactive agents flip that. They keep an eye on the apps you connect them to, notice things, and come to you.
All three of the big consumer players shipped one within about seven weeks. Meta’s Muse came first in early September, OpenAI launched Dots on September 29, and Anthropic followed with Claude Duet on October 21.
What it means
A proactive agent is an AI assistant that starts the conversation. You connect it to things like your email, calendar and files, and it checks them on its own every few hours. When something looks like it needs you, it messages you, often with a fix already drafted.
Most of them run on their own cloud computer, so they keep working after you close the app. And they’re built to stop and ask before doing anything you can’t take back.
Why it matters
The examples are pretty mundane, which is kind of the appeal. Claude Duet sends a morning brief from your email and calendar, follows up on threads that went quiet, and can flag unpaid invoices through Stripe. One early Dots tester said his agent noticed he’d forgotten to bill a publication, wrote up the invoice and waited for his OK.
The approval step is basically the whole safety plan. Duet asks before it sends a message, makes a payment, signs a document or deletes a file. Dots pauses for a check before anything that touches your accounts or shares information. That works great right up until you get tired of tapping “approve.”
They’re not free, exactly. Duet and Dots don’t cost extra, but they’re limited to the pricier plans (Claude Pro and Max, ChatGPT Pro and Business Premium), and Duet’s background checks count against your usage limits. An agent that reads your inbox every few hours is spending your plan while you sleep.
Simple example
A good waiter doesn’t wait for you to flag them down. They see your water glass is empty and fill it. They notice you checking your watch and bring the check over. Nobody asked. That’s the proactive part.
But they don’t order you dessert. They ask, because dessert costs money and you might have opinions. Proactive agents are supposed to work the same way. Small stuff, they just handle. Anything with a price tag, they come back to the table.