OpenAI launches Dots, its answer to Meta's Muse

OpenAI used its DevDay keynote on Tuesday to announce Dots, a fleet of always-on AI assistants designed to "do nearly anything" across connected apps while the user is doing something else — a direct response to Meta's buzzy Muse AI agent.

Like Muse, Dots are represented by cute, customizable avatars. And like Muse, they are built around a promise that has become the defining bet of the consumer AI market in 2026: that software agents will act on a user's behalf in the background, learning preferences over time, rather than waiting to be prompted for one answer at a time.

The announcement puts the two largest consumer AI companies in open competition over the same product category. Meta launched Muse to significant attention, and its hands-on coverage emphasized the shopping-oriented, message-threaded experience of interacting with an agent. OpenAI is now countering with its own take on the same idea, delivered through the ChatGPT ecosystem that the company has spent years building into a mainstream platform.

What OpenAI announced

According to OpenAI's DevDay presentation, Dots are always-on assistants. They work in the background across connected apps, and the company says they can "do nearly anything" — the exact phrase OpenAI used in its keynote. The agents also learn a user's preferences over time, which positions them not as one-off tools but as persistent software companions that accumulate context about the person they serve.

That persistence is the core of the pitch. A Dot is not a chat window that goes dormant between queries. It is, in OpenAI's framing, a continuously available helper that operates across the applications a user has connected to it.

GPT-6 Astra under the hood

Dots are powered by OpenAI's GPT-6 Astra model, the capable flagship the company released earlier and has positioned as its most advanced system. Each Dot also gets access to its own cloud computer, which it uses to operate a web browser and to reach more than 4,000 supported apps.

Those numbers matter for two reasons. First, the model: agentic products live or die on the reliability of the underlying model, because an agent that browses, clicks and transacts autonomously makes errors far more consequential than a chatbot that merely suggests text. OpenAI tying Dots to GPT-6 Astra signals that it views agent-grade competence — sustained, multi-step task execution — as a flagship-model capability rather than a lightweight feature.

Second, the app count: more than 4,000 supported apps gives Dots a wide surface area for action on launch, an attempt to answer the most common criticism of agent products, which is that they work well in demos and poorly across the messy reality of a user's actual digital life.

How you talk to a Dot

Interaction happens through two channels. The primary interface is a text-message-like thread — an experience similar to the one Meta offers with Muse, where talking to the agent feels closer to texting a person than to operating software.

Users can also hop on a voice call with a Dot from ChatGPT on the web, on desktop, or on mobile. That voice-calling capability, spread across all of OpenAI's client surfaces, extends Dots beyond text and into the conversational voice territory that OpenAI has been pushing across its product line.

The choice of a messaging-style interface, now shared by both Meta and OpenAI, is itself a signal about where consumer AI agents are heading. Both companies have concluded that the chat thread — not a dashboard, not an app store of skills — is the metaphor users will accept for delegating tasks to software.

Why the Muse comparison is the story

OpenAI did not launch Dots in a vacuum. The Verge's coverage frames the announcement explicitly as a response to Meta's Muse, the AI agent Meta rolled out to considerable buzz and which its hands-on reviews described as a shopping-capable assistant with a friendly, message-based interface.

The competitive dynamics here are straightforward. Meta distributes AI through its social platforms and a hardware ambitions of its own; OpenAI distributes through ChatGPT, one of the most widely used AI products in the world. Both companies now offer an always-on, avatar-bearing, preference-learning agent that text-messages you and does things on your behalf. The overlap is nearly total, and it comes at a moment when the consumer AI market is consolidating around a handful of dominant platforms.

For users, the immediate stakes are which ecosystem captures their agent relationships. An assistant that learns your preferences over time accumulates a switching cost: the longer you use one, the more it knows about you, and the harder it becomes to move. That is precisely why both Meta and OpenAI are racing to establish their agents now, before habits form around a rival's product.

For the broader industry, the stakes are the shape of the app economy itself. If Dots and Muse-style agents become the primary way people interact with services — browsing, shopping, scheduling through an intermediary — then the apps become backends and the agent becomes the front door. More than 4,000 apps supporting Dots at launch suggests OpenAI has been negotiating that inversion with developers for some time.

The agentic bet, restated

Dots also mark a continuation of the strategy OpenAI has pursued throughout its recent product history: turning ChatGPT from a conversational text model into a general-purpose operating layer for digital life. GPT-6 Astra gave the company a model it describes as capable enough for demanding tasks. Dots give that model a persistent, personalized consumer form.

The "always-on" framing carries real technical and commercial weight. An agent that works in the background, holds its own cloud computer, and maintains an ongoing relationship with a user is a fundamentally different product from a query-and-response chatbot — in infrastructure cost, in privacy exposure, and in the trust it must earn before people let it act inside their email, their calendars, and their shopping accounts.

OpenAI's claim that Dots can "do nearly anything" is the kind of broad promise the agent category invites and that hands-on testing tends to qualify. What is not in dispute is the direction: both of the industry's largest consumer AI players have now shipped always-on, avatar-based, preference-learning assistants within the same competitive window, each backed by a flagship model and thousands of app integrations.

What comes next

OpenAI announced Dots during its DevDay keynote, its developer-focused event, which means the company is pitching the platform to builders as much as to consumers. The more than 4,000 apps supported at launch represent the first cohort of that ecosystem; the DevDay framing suggests OpenAI expects the number to grow as developers build for the agent layer directly.

The immediate question for the market is how Dots and Muse diverge in practice — in reliability, in the depth of their app integrations, and in how aggressively each learns and uses personal preferences. Both products now exist, both are backed by flagship models, and both are asking users for the same thing: permission to act. Which permission users grant, and to whom, will shape the next phase of the consumer AI market.