OpenAI unveiled Dots at its DevDay conference in San Francisco, entering a fast-forming race for always-on AI agents after rivals including Meta and Instinct drew attention. The GPT-6 Astra-powered product presents itself as a friendly, customizable blob with eyes — but the strategic target is clear: Meta’s Muse.¹

At launch, OpenAI argued that Dots can do more than answer requests. Demonstrations showed the agent handling tasks and even building virtual worlds, while its interface lets users watch it work through a virtual machine. Yet the defining difference is access. Muse is free and comes with a dedicated virtual machine, while Dots is initially gated behind paid OpenAI tiers — a limitation that could put it beyond most of ChatGPT’s enormous user base.²

That divide also shapes the product’s personality. One early hands-on account found that, despite the cute avatars and ability to help buy things, Dots feels less like a personal shopper than a colleague: “business software that can use other software.” The agent can work with apps including Blender and GIMP, connect through the ChatGPT desktop app, and be called for voice help. The tester used a $100-a-month Pro account, underlining the workplace-first message behind the playful design.³

Sam Altman, meanwhile, has framed the appeal in intimate terms. He called Dot his “favorite openai product so far,” saying it improves daily as it learns his workflow and takes on the chores that otherwise become a “gravity well of dread.”Sam Altman (sama)⁴

The broader bet is that agents could create a deeper kind of loyalty than chatbots. Users may freely spread questions across several AI services, but handing an agent access to personal data, files and routines raises the cost of switching. That makes Dots’ paywall both its immediate weakness against Meta and, for OpenAI, a potential foundation for a more lucrative lock-in.¹