OpenAI says its new always-on assistant can "do nearly anything" across the apps you already use. It's called Dots, and it never asks for your attention.
Unveiled during the company's DevDay keynote on Tuesday, Dots represents OpenAI's most ambitious push yet beyond the chat window. Instead of waiting for a prompt, Dots works in the background, connecting to other apps, completing tasks, and quietly learning what you prefer along the way. The system runs on GPT-6 Astra, OpenAI's most capable model to date.
The pitch is simple, even if the engineering is not: an AI that acts more like a diligent assistant than a search box. The stakes are considerable. Whoever owns the layer that sits silently inside everyday software — calendars, email, messaging, shopping, travel — owns one of the most valuable positions in computing.
What Is OpenAI Dots and Why Does It Matter
McKinsey has estimated that generative AI could add trillions of dollars in annual value to the global economy, with a large share coming from automating knowledge work. Gartner projects that by 2028, a meaningful fraction of enterprise software interactions will be handled by autonomous agents rather than direct human input. Dots is OpenAI's bid to be the agent underneath those interactions.
The concept of an "agent" has dominated AI product roadmaps for two years, but most shipping products still behave reactively. You type, it answers. Dots flips that model. According to OpenAI's announcement, the assistant operates continuously, monitoring connected apps and taking action without a fresh prompt each time. It also adapts — the more it observes, the better it anticipates what you want.
That shift matters because it changes what an AI company sells. A chatbot is a destination. An always-on agent is infrastructure. OpenAI is signaling that it wants to be the latter, embedded in the tools people already open every morning rather than hoping they remember to open one more.
OpenAI DevDay 2026: The Announcement in Context
DevDay has become OpenAI's stage for platform-level moves. In 2023, the company used the event to launch plugins and the GPT Store, turning ChatGPT from a single product into an ecosystem. In 2024, it introduced GPT-4o, a model built for real-time, multimodal interaction across voice, vision, and text. Each DevDay has followed a pattern: a capability leap paired with a distribution play.
Read next Laika's Wildwood: Stop-Motion Fantasy at TIFF 2026Dots fits that mold. Announcing an agent framework at DevDay, alongside a flagship model in GPT-6 Astra, positions the product as both a technical milestone and a strategic land grab. OpenAI isn't shipping Dots as a standalone app you visit. It's shipping it as something that attaches to the apps you already live in.
The timing is instructive. Agentic AI has moved from demo reels to procurement conversations. Enterprises are piloting agents for customer support, scheduling, and internal workflows, and the vendors that win those pilots will shape defaults for years. DevDay gives OpenAI a captive audience of developers — the people who actually wire these systems into products.
How Dots Works Silently Across Your Apps
Picture a Tuesday morning. You haven't opened your calendar yet, but Dots has already noticed a scheduling conflict, drafted a rescheduling message, and flagged a flight that needs rebooking. Nothing flashes on screen. Nothing demands a tap. The work is simply done, pending your approval.
That's the operating model OpenAI describes: background execution across connected apps, with the assistant learning preferences over time. The GPT-6 Astra model supplies the reasoning horsepower, handling the multi-step planning that earlier systems struggled with — understanding that moving a meeting has downstream effects on travel, prep time, and who else needs to know.
The technical challenge isn't intelligence alone. It's coordination. An agent that touches email, calendars, and messaging needs permission structures, error handling, and a way to know when to stop. The hardest part of always-on AI is knowing when not to act. OpenAI's framing emphasizes learning your preferences, which suggests the system is designed to calibrate over time rather than act aggressively on day one.
For users, the practical difference is subtle but real. Reactive AI answers questions. Proactive AI removes tasks. That's a different contract, and it demands a different level of trust.
OpenAI vs Meta: The Race for Background AI Dominance
Meta's Muse gave the industry a template: an AI presence that follows you across surfaces rather than living in one app. OpenAI's Dots is a direct answer.
The competitive logic runs deeper than feature parity. Meta owns the social graph — messaging, feeds, and the daily habits of billions of users. OpenAI owns model capability and a developer ecosystem built through DevDay announcements like plugins and the GPT Store. Each company is trying to invade the other's territory. Meta wants better reasoning; OpenAI wants distribution.
Analysts have consistently ranked distribution as the harder problem. A superior model that nobody installs loses to a mediocre model that ships by default. That's why Dots matters beyond its technical specs. If OpenAI can persuade developers to build Dots into their apps, the assistant reaches users without OpenAI needing to win a single app-store download.
The race, then, isn't about who has the smartest agent. It's about who becomes the invisible default. Meta is betting on reach. OpenAI is betting on capability plus developer adoption — the same playbook that turned ChatGPT from a demo into a platform.
Privacy and Trust Concerns With Always-On AI
An assistant that watches your apps is an assistant that sees your life. That's not a side effect; it's the mechanism.
AI researchers have drawn a sharp distinction between reactive and proactive systems on exactly this point. A chatbot only knows what you tell it in the moment. An always-on agent accumulates context continuously — who you email, when you travel, what you buy, how you work. The value and the risk come from the same source.
Data sovereignty questions follow quickly. Where does that context live? Who can access it? What happens when an agent acts incorrectly — books the wrong flight, sends the wrong message, shares the wrong document? Consumer trust surveys have repeatedly shown that users are more comfortable with AI that asks permission than AI that assumes it. Dots' emphasis on learning preferences suggests OpenAI knows the approval loop is the product's weakest link.
Regulators are circling the same territory. The EU AI Act imposes transparency and oversight requirements on systems that act autonomously, and "always-on" agents sit squarely in that zone. OpenAI will need to demonstrate that background operation comes with foreground control.
What Dots Means for Developers and App Ecosystems
Every platform shift rewards the developers who move first. When Apple opened the App Store in 2008, the earliest builders captured categories that didn't exist a year earlier. The same dynamic is forming around agents.
For developers, Dots offers a way to add intelligence without building a model. Instead of training or fine-tuning their own systems, teams can connect their apps to OpenAI's agent and let it handle cross-app coordination. That's the plugin playbook from DevDay 2023, upgraded from "call this tool" to "act on my behalf."
The risk is dependency. Apps that hand their user relationships to an external agent may find themselves reduced to data sources. The companies that thrive in an agentic world will be the ones that integrate without becoming interchangeable — offering capabilities the agent can't replicate alone.
That tension will define the next few years. OpenAI Dots isn't just a product launch. It's a claim about where the intelligence layer should live. If OpenAI is right, the most important software of the next decade won't be the apps on your home screen. It will be the agent running quietly behind all of them.
Source: The Verge
