OpenAI's leaked 'o' assistant may get its own email address

On Friday, someone at OpenAI apparently shipped a word a few days early. @JakeABoggs posted a screenshot of the ChatGPT upgrade screen and asked, "Did I miss a product launch or is this a dev day leak?" The screen mentioned something called "o."

The next morning, @testingcatalog, an account that tracks unreleased features by digging through app code, filled in more. It says the upcoming product is "an upcoming always-on assistant from OpenAI" named "o." It also reports that ChatGPT's internal config lists "o" as a display name and "-o" as an email suffix, and reads that to mean o will handle email.

Then OpenAI's developer account added its own hint. @OpenAIDevs posted "72 hours to OpenAI DevDay. We've been building. Time to show our work." The post passed a million views within a day. DevDay is Tuesday, September 29, in San Francisco.

What is actually confirmed

Not much. OpenAI has not announced a product called o. We have a screenshot, some config strings a leak-tracking account found, and an OpenAI countdown post that never names an assistant. The rest is inference.

There's a side joke going around too. Late Friday, after an outage, OpenAI's @thsottiaux opened a post about it with "o yes… we're back in action" and reset usage limits for paid users. Maybe that was a wink and maybe it was just a way to start a sentence. I wouldn't read anything into it.

TestingCatalog also clears up a separate rumor. It says "Aeon," another name that surfaced recently, is internal naming for the custom Agents that ChatGPT Workspace accounts already have, and it expects OpenAI to build the consumer version of o on top of that.

Why the email suffix matters

I care less about the name than about "-o" showing up as an email suffix.

If TestingCatalog is reading that right, o would have its own identity that can send and receive mail, on top of reading yours. That's a long way from a chatbot that waits for you to type, and it matches where the rest of the industry went this month. Microsoft just pitched Copilot Autopilot as an agent that keeps working after you log off. Meta's Muse already places calls for users. And as we wrote earlier this week, agents are starting to behave like scheduled jobs that check back later on their own.

TestingCatalog notes this too, saying Meta "set a super high standard" with Muse and OpenAI now has to clear that bar.

My honest read

I'd bet on some kind of always-on assistant showing up at DevDay, given the timing, the upgrade-screen copy and the countdown. I wouldn't bet on the details. Leaked config strings show what engineers were testing, which often isn't what ships, and the email piece could easily launch narrow, late, or only on the top paid tier.

For a business owner, what matters more is what happens once an assistant like this has an address and can act without being asked.

What this means for a business

An always-on agent with its own email is good at dull, recurring work, like chasing a quote that went quiet or nudging a late invoice. It also raises questions you haven't had to answer before. A customer might get a message nobody on your team wrote, and their reply might land in a mailbox nobody checks. Someone has to decide what the agent is allowed to promise.

You don't need o to prepare for that. The prep is the same whichever vendor wins:

• Write down which messages you'd let software send without a human reading them first
• Decide which inbox an agent replies from, and who watches it
• List the recurring follow-ups that eat your week, since those are where an agent earns its keep

If you'd like a second set of eyes on that list, our free process audit walks through your week and sorts the tasks an always-on agent could take today from the ones that should stay with a person. Tuesday's keynote will tell us what OpenAI actually built.