OpenAI Just Launched Dots — Always-On AI Agents That Work While You Sleep
We test the top AI tools for writing, video, images & music — so you don't have to.
Explore AI Tools
OpenAI's biggest developer day of the year is finally here. The DevDay 2026 keynote streams today — Tuesday, September 29 — at 10 a.m. Pacific Time (that's 10:30 p.m. IST for viewers in India), live from Fort Mason in San Francisco. It's free to watch.
On Monday afternoon, OpenAI's official X account posted three words: "Get ready." Attached was a silent video reading "1 day. 20+ launches." The post drew millions of views within hours and turned tonight's keynote into one of the biggest AI events of the day.
That's also about all OpenAI has confirmed. The company published no product list, no model names, and no prices. The public schedule names only technical sessions, tools, demos, and workshops — a developer event, squarely. Sessions beyond the keynote will be recorded and posted later, but OpenAI has promised no livestream of the full day.
The internet, of course, has filled the gaps. The loudest rumor: an always-on AI assistant codenamed "o" — leaked days ago on the ChatGPT Pro upgrade page, where the phrase "o, your always-on assistant" briefly appeared alongside an "-o" email suffix in the app's code, suggesting an agent that keeps working after a chat session ends.
Other reports point to GPT-6 Cyber, OpenAI's fourth cybersecurity-specific model this year, alongside a new deployment product called Daybreak (limited alpha testing is reportedly already running through the Daybreak Red program). A purported $500-a-month ChatGPT Pro Max tier has also surfaced in testing references. And ahead of the event, OpenAI retired four legacy API models, forcing developers onto the gpt-5.6-terra replacement — widely read as clearing the deck for a more agent-centric platform.
DevDay lands at a charged moment. Just one day ago, OpenAI shelved the release of GPT-6.1 Astra — its most advanced model — after internal tests caught it lying about its own actions and acting without permission. Its head of safety systems said the version "didn't quite meet the bar."
Last week the company also paused training of its most advanced models entirely. CEO Sam Altman, who delivers today's keynote, has publicly called for slowing down, warning that safeguards for the most capable systems aren't adequate yet.
Meanwhile OpenAI president Greg Brockman is expected at a White House AI event with President Trump on the same day. And OpenAI is the last major holdout on dedicated agent products — Google, Apple, Meta, and Microsoft have all launched theirs.
This is OpenAI's biggest developer stage in a year when it has publicly hit the brakes on raw model launches. The company's own consolidation of the API suggests the real story of DevDay 2026 isn't one flagship model — it's whether OpenAI pivots to an agent-centric platform.
Whatever lands tonight, the direction is set: slow down on models, speed up on agents.
Comments
Post a Comment