On September 26, OpenAI's official developer account posted a cryptic teaser containing just the letter "O." The same day, leak site TestingCatalog revealed that multiple references to "O" had appeared in ChatGPT's configuration, including the display name "O" and an email suffix "-o." Taken together, they point to one answer: OpenAI is preparing a persistent AI agent. And DevDay on September 29 in San Francisco — confirmed by OpenAI, with Sam Altman set to deliver the opening keynote — looks like the likely launch window.
Three clues dug out of the configuration
First: "O" shows up as a display name in ChatGPT's config. Second: the config includes an email suffix "-o," meaning "O" was designed from the start with its own email identity — a persistent agent needs to send and receive email and handle calendars on its own, not wait to be woken up every time. Third: some users briefly spotted "O" on the $100-per-month Pro plan upgrade page, suggesting it will likely ship to paying users first.
The three clues paint a clear product shape: an agent that doesn't depend on a chat window and keeps working in the background. TestingCatalog's description: "a persistent agent designed to keep working outside a normal chat session."
"Aeon": the other half of the codename puzzle
"O" didn't come out of nowhere. TestingCatalog notes it may be connected to OpenAI's internal "Aeon" project — Aeon has long been associated with OpenAI's persistent-agent R&D, and the name recently appeared in configuration for custom agents aimed at ChatGPT Workspace accounts. The reasonable guess: Aeon is the underlying infrastructure, "O" the consumer-facing product.
The o handle on X is currently in a suspended/reserved state, also seen as a sign that the official account is being held for launch. Until OpenAI makes a formal announcement, though, all of this remains at the "highly credible puzzle" stage — task scope, permissions, scheduling, and memory mechanics are all unknown.
What does a persistent agent actually do
Unlike today's "ask and answer" ChatGPT, the core of a persistent agent is "it keeps working whether you're there or not": regular industry research, ongoing information monitoring, email correspondence, deadline reminders — tasks that stretch across hours or even days, exactly what traditional chatbots do poorly. OpenAI employee Tibo posted on X the same day hinting at this shape: the agent moves on its own to advance work, and the user just sets the goal.
The competitive pressure is real, too. Meta's personal agent Muse, launched early this month, quickly hit No. 1 on the US App Store free chart — on the strength of "doing things for the user" rather than "chatting with the user." If OpenAI brings out "O" at DevDay, it would formally move the battlefield from "whose model is smarter" to "whose agent gets more done."
Two days remain until DevDay on September 29. The official teaser is out, and "O"'s config references, email suffix, and Pro-plan sighting are all in place — leaving just two questions hanging: how independently it can actually work, and whether OpenAI dares to let it truly stay on.
