AI
Claude's built-in agents rival OpenAI's Dots—but Anthropic is playing a different game
Anthropic quietly folded persistent agent capabilities into Claude while competitors chase downloads; the real battle is winning developer trust and finding lasting work.
The New Stack
OpenAI made waves this week with Dots—autonomous agents that live in the cloud, monitor Slack, and trigger actions without prompting. Meta's Muse is driving massive download numbers. Yet Anthropic's answer has been quietly embedded in Claude for two weeks, without the hype, the mascot, or the fanfare.
The competition heating up around always-on agents represents one of the fastest-moving races in AI. Within months, competitors have shipped overlapping features while borrowing liberally from one another's approaches. But the strategies diverge sharply: Meta needs blockbuster consumer adoption, OpenAI is betting on premium paid tiers, while Anthropic is counting on developers and teams who already use Claude to discover recurring work worth delegating.
A crowded field emerged almost overnight#
The current wave of persistent agents barely predates mid-2024. Peter Steinberger's Clawdbot—a weekend project that lived on spare hardware and took commands over messaging apps—kicked things off last November. It faced a trademark dispute, became Moltbot, then evolved into OpenClaw. Within ten months, the space had spawned three names, brought an OpenAI hire (Steinberger himself), and spawned an enterprise edition.
The pace of feature diffusion across competing projects has been striking. Meta's Nat Friedman acknowledged that Muse drew inspiration from OpenClaw, and users examining Muse's code found personality configurations nearly identical to OpenClaw's. Anthropic's September merge of Cowork into Claude (which shipped scheduled task capabilities on July 7, two days before OpenAI's Work mode) shows how quickly ideas propagate across the sector. When the category is this nascent, shipping a comparable feature is how labs attract their core users.
Three different bets, three different audiences#
Meta's high-volume play: Muse exploded out of the gate. In its first dozen days, iOS users reached an estimated 359,000 daily—outpacing ChatGPT's comparable period. By late September, downloads exceeded 3.4 million. Meta needs this win; its Reality Labs division has burned through tens of billions on metaverse bets with little consumer traction, and last year's Llama 4 landing fell flat. Muse is the first breakthrough from Zuckerberg's rebuilt AI division, and it got there by leveraging Meta's existing 2-billion-person social footprint. More than 95% of Muse users also use Facebook. Pricing starts at free, climbs to $20 and $100 monthly, and includes integrations for Shopify and Stripe—meaning agents can transact.
OpenAI and xAI start with wealthy customers: Dots requires a ChatGPT Pro subscription ($100–$500 monthly) or a business account—Altman framed it as a premium product because each agent demands substantial compute. Grok Bot reached 418,000 weekly users by mid-September and xAI has launched Team Bots. Both companies avoid chasing consumer volume; they're testing retention among paying users to see whether agent utility outlasts the initial novelty.

