The OpenAI Dots agent is OpenAI’s answer to a simple question: what if your AI assistant didn’t stop working the moment you closed the chat window? Unveiled by Sam Altman at DevDay on September 29, Dots is a set of colorful, blob-shaped avatars that keep working on your tasks around the clock. It has turned AI agents into a genuine race, with Meta’s rival agent Muse drawing millions of downloads.
Unlike a chatbot that waits for your next prompt, a dot is given a job and keeps going after you close the laptop. Here’s what the OpenAI Dots agent actually is, what it costs, what it does well, and why privacy has become the battleground of this new agent war.
What the OpenAI Dots agent actually is
A dot is not another chat window. On stage at DevDay in San Francisco, Altman showed colored blobs, each a worker you hand a responsibility to, the way you’d brief a chief of staff. Each dot runs on OpenAI’s GPT-6 Astra model, gets its own Linux computer in the cloud, and has its own browser. You can open that computer and inspect exactly what it did.
OpenAI says a dot can plug into more than 4,000 apps, with Slack and Teams in the first set of connections. It can also operate your own laptop if you allow it, and texting a dot over iMessage or Android RCS was waitlisted for Pro users rather than switched on for everyone at launch.
What it costs and who can get it
Rollout started with people already paying. ChatGPT Pro is $100 a month, Business Premium is $100 per user a month, and enterprise customers, including education and healthcare workspaces, get a beta once an admin turns it on. The first dot is included in the plan. OpenAI also showed specialist dots for business jobs like accounts, email marketing, and legal review.
The funnel is enormous: OpenAI put ChatGPT’s weekly users at over 1.2 billion, and said Codex and ChatGPT Work together had passed 35 million weekly users, up from about 5 million Codex users in June. Altman promised a mass-market version “for billions of people someday.” For now, though, the always-on agent is a premium product.
What Dots does well in real use
Writing in Fast Company, global technology editor Harry McCracken gave Dots access to his calendar, email, and Dropbox, and found the agent genuinely proactive. It spotted a discrepancy he had missed: he’d entered a speaking engagement on one date while a confirmation email mentioned a different day. The organizers had rescheduled without telling him, and the agent caught it.
Without being asked, Dots started flagging messages it suspected were urgent, and its sense of which emails actually mattered was dead-on. On shopping research, it found a $100-off iPad Magic Keyboard deal he would never have located himself, and prepared a wireless-plan report by pulling his existing cost and data usage straight from emailed receipts and overage alerts.
OpenAI’s own guardrails: a dot learns what you care about over time, but sensitive actions, a password change is the example the company gave, still need your confirmation. Business data is not used for training by default, and personal users can opt out of having their conversations and dot work used that way.
OpenAI Dots vs Meta Muse: the privacy fight
Dots did not arrive in a vacuum. Meta had launched its own always-on agent, Muse, earlier in September, pitched for travel, mail, and everyday tasks, and it picked up millions of downloads, at one point topping Apple’s list of most-downloaded iPhone apps. We covered Meta’s latest frontier models, Muse Wonder, earlier today.
At DevDay, Altman turned privacy into a weapon. “We want to set a new standard for privacy in frontier AI,” he declared, taking shots at Muse’s data-safety record. The irony: Meta had launched Muse months earlier with nearly identical promises, calling it “built from the ground up for privacy and security” and positioning it as the safer successor to OpenClaw, its previous agent, which had faced criticism over data handling.
This is the structural story of the always-on AI agent era. To be useful, these agents need deep access to your emails, calendar, and accounts. The more they know, the better they work, and the bigger the liability if that data is compromised. Both have credibility problems: OpenAI fired three safety researchers for “breach of trust” just days earlier, while Meta’s data history means every Muse privacy promise gets scrutinized twice.
The rough edges
Dots is not polished yet. The Fast Company review notes that reaching Dots means rummaging around in the ChatGPT app’s sidebar each time, the anthropomorphization leans cloying, and voice runs through a simulated phone call rather than the voice mode ChatGPT already had. It also feels odd to choose between Dots and ChatGPT when the app already had Chat, Work, and the Codex coding agent.
OpenAI seems to agree the split is temporary. At the DevDay press Q&A, product chief Thibault Sottiaux said, “We’re going to be evolving ChatGPT in place to be this incredible personal assistant very soon.” The distinction between ChatGPT and Dots is expected to fade.
Competition is crowding in fast. A startup called Instinct has an agent inside texting apps like WhatsApp and Apple Messages, still on a waitlist but reportedly valued at $10 billion. Hark Pro, another contender, debuted on October 6 and is free, while Google’s own Gemini Spark, announced months ago, still doesn’t support paid Workspace accounts. The AI agent push into enterprise tools like Webex shows the same pattern: everyone wants their agent to be the one living inside your workflow.
Should you care yet?
For most people, the honest answer is: not yet, unless you’re already paying for ChatGPT Pro. The always-on AI agent is genuinely new as a product category, and hands-on testing suggests the proactive help is real, catching scheduling mistakes, triaging email, and researching purchases from your actual data. But at $100 a month with the mass-market version still a promise, Dots is a preview of where this is all heading.
What is worth watching now is the trust equation. Agents that book things, edit files, and read your email are a fundamentally different risk from chatbots that just talk. OpenAI’s pitch is permission rules plus a cloud computer you can inspect. Whether that holds once millions of dots are acting on their own is not something any launch video can settle.
