OpenAI on Sept. 29, 2026, introduced Dots — “remarkably capable, always-on agents built to handle everything,” in the company’s words — at its DevDay event. Per OpenAI’s announcement and ChatGPT Learn docs, a Dot is meant to keep work moving across tools and projects between conversations, ask for input when a decision needs the user, and keep going even when the user’s computer is off. Each Dot gets its own cloud computer and browser, OpenAI says, powered by GPT-6 Astra, and can connect through OpenAI’s plugin ecosystem to “over 4,000 apps.”
That is a different product door than Meta’s Muse, the consumer personal AI agent Meta introduced Sept. 8. Press coverage, including WIRED, has framed Dots as an “answer to Muse”; OpenAI’s own primaries reviewed for this piece do not name Muse. Treat the rivalry language as press framing, not OpenAI’s claim. This piece compares Dots to Muse the personal/agent product — not Meta’s Enterprise Platform packaging under CJ Desai.
OpenAI says users create a Dot in the ChatGPT desktop app or desktop browser, then can reach it in ChatGPT on desktop, web, and mobile after setup, plus Slack and Microsoft Teams, with voice calling in ChatGPT and texting “coming soon” or “coming next.” A user can inspect the Dot’s cloud computer and optionally connect one personal computer (online, with the ChatGPT app open) for local files and apps; the cloud machine stays separate from the user’s device sessions. OpenAI says Dots start with ChatGPT memory and context, learn from feedback, can run background or parallel work, and may do “proactive research” when not actively tasked — research tools that OpenAI says are read-only and cannot send messages, change plugin content, or control the browser or computer. Follow-ups that change anything need normal permissions. Custom Rules and Auto-review gate actions that affect accounts or share information; sensitive steps such as changing a password stay with the user or require take-over, OpenAI says. Dots can still make mistakes, OpenAI warns, and prompt-injection defenses are not claimed to be perfect.
Availability, per OpenAI Learn and the announcement: ChatGPT Pro users over 18 outside the EEA, United Kingdom, and Switzerland; Business Premium rolling out worldwide; Enterprise (including Education and Healthcare) as an admin-on beta, off by default. The first Dot is included with Pro or Business Premium at no extra cost; OpenAI describes a deeper-work allowance with extended limits in the first month and more Dots or scale “in the future,” without publishing prices for extra Dots. Conversations with a Dot do not count toward ChatGPT usage limits; Work and Codex tasks do. Rollout is gradual — an eligible plan does not mean the feature is immediately visible. Specialist Dots with org-provisioned identity, credentials, and tools are framed as focused enterprise pilots, with Microsoft Agent 365 called out for governance; teams of Dots remain a stated vision, not general availability. Those pilots sit outside this Muse comparison and outside Meta’s Enterprise Platform story.
Meta’s Muse, by Meta Newsroom, is a personal AI agent that “doesn’t just answer questions, it actually does the work” — goals to plans to progress, including after the app is closed, with approval asked for sensitive steps such as sending email or making a purchase. Muse runs on Muse Spark inside a Muse Secure VM with its own browser; a separate Sentinel agent must approve what reaches the internet, Meta says. Surfaces at launch were the Muse app, WhatsApp, and muse.ai in the U.S., with AI glasses listed as coming soon. Meta says Muse is free for most of what people need, with subscription plans for people who want to do more; exact dollar tiers are not in the Newsroom posts used here and are not invented below. On Sept. 29 — the same calendar day as Dots — Meta said Muse is available in the U.S. and Canada and expanded small-business skills and connectors (Asana, Shopify, Slack, Stripe, Meta business accounts, and others), with the claim that “nothing publishes, sends, or spends without your approval.” That SMB expansion is consumer/SMB agent framing, not the Enterprise Platform pillar.
Head to head, both companies pitch agents that do work rather than single-turn chat. OpenAI’s lane is paid ChatGPT tiers, workplace surfaces (ChatGPT, Slack, Teams), a named cloud computer per Dot, and plugin reach OpenAI pegs above 4,000 apps. Meta’s lane is a geo-rolled consumer agent on Muse app, WhatsApp, and web, a Secure VM plus Sentinel gate, and connectors the user chooses — with SMB tools expanded on DevDay Tuesday. Neither side’s safety or approval story is an independent audit: OpenAI’s Auto-review and read-only proactive research are OpenAI claims; Muse’s Sentinel and approval gates are Meta claims.
What Sept. 29 settled is product shipping and framing, not a scoreboard. OpenAI put always-on agents inside ChatGPT’s paid and workspace stack; Meta kept Muse on a consumer-and-SMB track the same day. Whether users trust an agent with a cloud computer, inbox connectors, and proactive research is the open question both launches leave on the table.