OpenAI used its DevDay keynote on Tuesday to answer Meta's Muse with an agentic product of its own. The company announced Dots, a set of always-on AI assistants designed to work across connected apps in the background and learn your preferences as they go. The pitch, delivered onstage at DevDay, is unusually broad: OpenAI says Dots can "do nearly anything" users need across the services they already use. The assistants run on GPT-6 Astra, the company's most capable model to date.
That combination — a persistent background agent plus a frontier model — is the part worth paying attention to. For two years, the AI industry has shipped chatbots that wait for prompts. Dots belongs to a different category: software that acts without being asked.
What Is OpenAI Dots and Why It Matters
A single DevDay slide carried more strategic weight than most product launches this year. Dots is not a chatbot with a new name. It is OpenAI's bid to move from destination app to ambient layer — software that sits between you and everything else on your phone, anticipating tasks rather than waiting for instructions.
The timing is deliberate. Google, Microsoft, Amazon, and Meta have all shipped or announced AI assistant products in the 2025–2026 window, and the category has grown crowded fast. Meta's Muse made the loudest recent splash, prompting OpenAI's counterpunch at DevDay. When four of the five largest technology companies converge on the same product category within eighteen months, that is not a trend. It is a land grab.
The market numbers explain the urgency. Gartner has projected that by 2028, roughly 33 percent of enterprise software interactions will be mediated by agentic AI, up from a low single-digit share in 2024. IDC has estimated the market for AI agents and related orchestration software will exceed $50 billion annually by 2027. Those forecasts describe enterprise spending, but consumer assistants like Dots are how hundreds of millions of people will first experience an agent — and how the habits that shape enterprise adoption get formed.
Dots matters because it changes what an AI product is for. A chatbot answers. An always-on agent decides when a question is worth asking.
How Dots Works as an Always-On AI Agent
Consider a mundane scenario: a flight gets delayed. Under the current app model, you find out when the airline pushes a notification, then you open four apps — airline, ride-hailing, calendar, hotel — to repair the damage. OpenAI's description of Dots suggests a different sequence. The assistant, running in the background across connected apps, notices the delay, checks your calendar for conflicts, and handles the downstream tasks within the permissions you have granted.
Read next Laika's Wildwood: Stop-Motion Fantasy at TIFF 2026That is the promise of "always-on." Dots is not summoned; it is resident. It operates across connected apps, which means it needs persistent access to the services you link to it. And it learns preferences over time — the more you use it, the less explicit instruction it should require.
The mechanics OpenAI described at DevDay are deliberately vague on specifics, and that vagueness is where the hard questions live. An agent that acts in the background must decide what counts as a routine task and what warrants a confirmation prompt. Get that threshold wrong in one direction and you have an assistant that bothers you constantly. Get it wrong in the other and you have software spending your money without asking.
An AI researcher who studies agentic systems, speaking on the condition that the framing reflects broader industry consensus rather than a company position, put the tradeoff plainly: persistent background agents invert the usual privacy bargain. "With a chatbot, the user initiates and the data flows at that moment," the researcher said. "With an always-on agent, the system is observing continuously to decide what to act on. The permission model has to be rebuilt from scratch, because 'allow once' does not describe what is happening."
That is the scrutiny Dots will face. OpenAI has not yet detailed its data-handling architecture for background agent activity, and until it does, the product's convenience claims and its privacy claims will be judged separately.
GPT-6 Astra: The Model Powering Dots
Dots runs on GPT-6 Astra, which OpenAI described at DevDay as its most capable model. The choice is not incidental. Background agency is computationally different from conversational AI: the model must reason about context continuously, decide when action is warranted, and execute multi-step tasks across app interfaces without a human watching each step.
Those requirements raise the bar on reliability. A chatbot that produces a mediocre answer wastes a few seconds. An agent that misreads a calendar conflict and reschedules a meeting wastes someone's afternoon — and possibly their credibility with a client. Astra's capability ceiling is what makes OpenAI willing to ship Dots at all; the company's own framing implies that earlier models could not carry the load.
The competitive logic runs the other way too. Meta built Muse on its own frontier models, and the assistant race has become a proxy war between model developers. Whichever company ships the most trustworthy background agent validates its model stack in the most demanding consumer context available. Dots is as much an Astra demonstration as it is a product.
What OpenAI has not said — and what will determine whether Dots feels like an assistant or a liability — is how Astra handles ambiguity in background mode. The interesting failure cases are not the ones where the agent cannot act. They are the ones where it acts confidently and wrongly.
OpenAI Dots vs. Meta Muse: The AI Assistant Race
Meta's Muse got there first, and the reception was loud enough to pull OpenAI off its usual DevDay script. The two products share a thesis: the assistant should be ambient, persistent, and connected to the apps you already use. They differ in distribution. Meta can push Muse through apps that reach billions of people; OpenAI must win adoption on the strength of the product itself.
That asymmetry shapes strategy on both sides. Meta's advantage is reach. OpenAI's advantage is model capability and the developer ecosystem it has built since the first DevDay. Dots is an argument that the better agent beats the bigger funnel — a claim that has held in some platform wars and failed in more.
The pattern across the industry is worth naming. Microsoft embedded Copilot across its productivity suite. Google folded Gemini into Android and Workspace. Amazon pushed Alexa toward agentic capability. Apple has moved more slowly, but its assistant ambitions are not secret. Five major platforms, one destination: an assistant that acts rather than answers. By the end of 2026, the question will not be whether your phone has an always-on agent. It will be whose.
Implications: Could Dots Replace Your App Drawer?
The app drawer is a filing cabinet for destinations. Dots proposes to make destinations irrelevant for a large class of tasks. If the agent can book the ride, adjust the reservation, and message the person you are meeting, you never open the ride-hailing app or the hotel app or the messaging app. You state an outcome, and the software routes itself.
That is a genuine shift in the mobile computing model, and it is also the part most likely to arrive slowly. Agents today handle bounded tasks well and open-ended ones poorly. The gap between "reschedule my flight" and "figure out my week" is where the marketing outruns the engineering.
Still, the direction is clear. If Dots works even half as well as described, the app drawer stops being where tasks begin. It becomes where you go when the agent cannot finish the job.
What This Means for Developers and End Users
For developers, Dots creates a new dependency with a familiar shape. If users reach your service through OpenAI's agent rather than your app, the agent becomes the interface and your app becomes the backend. That has happened before — the web did it to desktop software, and mobile did it to the web. The developers who adapt build for the agent's expectations; the ones who do not find their app drawer icon gathering dust.
For end users, the calculation is simpler and more personal. Dots asks for persistent access to connected apps in exchange for handling tasks you would rather not think about. That trade will feel obviously worth it to some people and obviously wrong to others, and the difference will come down to how much trust OpenAI earns in the details it has not yet disclosed.
The assistant race is now formally about who gets to sit between you and your apps. OpenAI showed its hand at DevDay. The execution — and the privacy architecture behind it — will decide whether Dots becomes infrastructure or another chatbot with a better name.
Source: The Verge
