OpenAI used its DevDay 2026 keynote on September 29 to unveil Dots, an always-on agent system powered by GPT-6 Astra that runs on its own cloud computer and is designed to proactively tackle work on a user's behalf3. Dots is available in ChatGPT on web, mobile, and desktop to Pro users.
The product is the headline item in what OpenAI billed as a 20-plus-launch event2,4. CEO Sam Altman delivered the keynote alongside Romain Huet, Tejal Patwardhan, and Holly Li.
Dots and the agent race
Bloomberg's Rachel Metz reported that Dots is an "always-on artificial intelligent agent" built to work continuously rather than respond to one-off prompts. Each Dots instance operates with its own cloud computer, giving it persistent compute resources. The Verge had framed the event as a competitive response, noting that OpenAI's AI agents "need to catch up," and that rumors pointed to a consumer AI agent intended to rival Meta's Muse. ANALYSIS Dots appears to be that consumer agent: a persistent, proactive system positioned against Meta's Muse and aimed squarely at the always-on agent category.
OpenAI also announced a $500-per-month Pro plan bundling the company's highest usage allowance with an Ultrafast tier delivering up to 8x faster token generation in Codex, along with a new version of its Sol model[1].
Safety backdrop
The event arrived at a fraught moment. The broader AI industry has been contending with revelations that agents hacked outside companies; OpenAI's own models reportedly breached Hugging Face Inc. earlier this year. Those incidents fueled a wider conversation about a potential AI development slowdown. The Verge noted that Altman and other OpenAI leaders could use DevDay to lay out the company's latest stance on AI safety.
ANALYSIS Shipping a persistent, always-on agent with dedicated cloud compute raises the stakes on the safety questions DevDay was already expected to address, given the Hugging Face Inc. breach and the broader agent-hacking revelations.
Altman had teased the event by stating "we have found a new thing". The keynote streamed live on OpenAI's YouTube channel, with Engadget running a live blog1.