At OpenAI’s DevDay in San Francisco, Sam Altman unveiled Dots, an “always on” agent powered by GPT-6 Astra. The launch arrived after rival agents from Meta and Instinct had seized attention, turning personal AI helpers into the industry’s newest contest for customer loyalty.¹

OpenAI presented Dots as more than a chatbot with a face. Like Meta’s Muse, it uses customizable, blob-like avatars and can work inside a virtual machine. But its emphasis is squarely on getting things done: it can access software including Blender and GIMP, connect to a user’s computer through the ChatGPT desktop app, and even be called for voice help. One tester’s verdict was blunt: Dots feels like “workplace software that happens to be able to order you a burrito.”²

Altman’s own early review framed the product as a tool for the chores that accumulate into a “gravity well of dread,” saying it improves as it learns his workflow and style.Sam Altman (sama)³ That learning loop is central to OpenAI’s wider bet. It is easy to spread occasional questions across several chatbots; it is much harder to grant multiple agents access to personal data and daily routines. The company hopes that friction will produce something closer to durable, one-provider attachment.¹

But the rollout exposes the weakness in that strategy. Dots is initially reserved for Enterprise and Business Premium customers and ChatGPT Pro subscribers, with access beginning at $20 a month and higher tiers. Muse, by contrast, is available free to anyone with a Meta account, including a dedicated virtual machine for the agent and its associated data.⁴

That leaves OpenAI trying to sell an agent as a high-value coworker while Meta can pitch its rival as a habit anyone can form at no cost. The cute avatars may look alike; the commercial models—and the audiences they can reach—do not.