In its boldest structural upgrade to personal computing since the launch of ChatGPT, OpenAI unveiled Dots at its annual DevDay 2026 keynote. Described as always-on AI agents, Dots represent a departure from traditional prompt-and-response chatbots. Instead of sitting idle until a user enters a prompt, a Dot operates continuously in the background on its own isolated, cloud-based virtual machine.
Equipped with its own web browser, operating system, and deep integrations across thousands of workplace and personal applications, a Dot acts as an autonomous digital proxy. It can monitor workflows, perform proactive research, draft documents, and execute complex multi-step projects independently. The launch signals OpenAI most aggressive move yet into the agentic AI space, where the battleground has shifted from conversational assistants to persistent digital workers.
Technical architecture | How Dots work
Driven by OpenAI flagship GPT-6 Astra foundation model, each Dot is provisioned inside a secure, sandboxed cloud environment. The core capabilities include a dedicated cloud environment with native browser access and OS controls, an app integration network connecting natively to more than 4,000 application plugins via OAuth and API bridges, and a cross-platform interface letting users converse with their Dot in ChatGPT, invoke it on Slack or Teams, or grant local laptop access.
Dots also perform proactive background execution, conducting proactive research and identifying pending errands such as flagging unpaid invoices without being asked. This is the defining difference from a chatbot: a Dot anticipates work rather than waiting to be prompted. The same persistent-agent architecture Microsoft is pursuing with Autopilot is now a core product for OpenAI, but OpenAI approach is more tightly integrated with its consumer ChatGPT distribution.
Proactive research vs. human approval
To balance autonomous productivity with safety, OpenAI implemented a multi-tiered permission framework. When operating autonomously without direct human supervision, a Dot runs in Read-Only Mode, scanning connected calendars, inbox threads, and project boards to gather context without altering files or sending unauthorized messages.
When an action requires external impact, such as sending an invoice, booking a flight, or updating a database, the agent pauses and prompts the user for explicit approval. Sensitive tasks, such as modifying master account passwords or executing financial transfers, are strictly restricted from automated execution. This permission cascade mirrors the human-in-the-loop safeguards that power agentic commerce platforms like Grok and Stripe Link, where final approval always stays with the human.
Escalating rivalry | OpenAI Dots vs. Meta Muse
The introduction of Dots positions OpenAI in direct competition with Meta newly expanded Muse platform. While both tech giants are racing to capture the emerging agentic AI ecosystem, their go-to-market strategies cater to different user bases. Dots target enterprise knowledge workers and power users, running on the GPT-6 Astra cloud virtual machine with more than 4,000 app connectors, and ship bundled in ChatGPT Pro and Pro 500 tiers priced at $200 and $500 per month respectively.
Meta Muse, by contrast, targets broad consumers and small businesses, running on distributed open-source infrastructure and interfacing through WhatsApp, Instagram, Ray-Ban glasses, and a mobile app. The rapid consumer adoption of Muse has already driven a JPMorgan upgrade, but OpenAI is betting that enterprise-grade persistence and deep app integration will win the high-value productivity market.
By granting ChatGPT a persistent, virtualized computer environment, OpenAI is betting that the future of personal productivity lies not in chatting with an AI, but in managing a team of autonomous digital workers. That thesis aligns with the broader governance debate over agentic AI, where the central question is how much autonomy to grant systems that can now act, not just converse.