OpenAI Is Reportedly Readying ‘o,’ an AI Agent to Answer Meta’s Muse
Set Trending Topics as a preferred source on Google.
Meta made the first move, and now it is OpenAI’s turn: at its DevDay conference in San Francisco on Tuesday, the ChatGPT maker could unveil a personal AI agent called “o” that works for its users around the clock. Nothing has been officially confirmed yet, but the clues have piled up over the past few days.
What We Know About “o” So Far
The speculation started with the leak tracker TestingCatalog. According to its report, ChatGPT’s configuration contains several references to “o,” including “o” as a display name and an email suffix “-o.” That suggests the agent could come with its own email identity. For a brief moment, “o” also appeared on the upgrade page for the Pro plan, which costs $100 (roughly €85) a month, described as an “always-on assistant” that keeps working after the chat window is closed.
Internally, the project is said to be linked to “Aeon,” a name OpenAI already uses for its continuously running agents. Several reports, including one by Wccftech, expect “o” to run on a variant of the GPT-6 Astra model tuned for long-running tasks. Rumors say the agent will be available across all Pro tiers. OpenAI is also expected to introduce a new ChatGPT Pro Max plan at around $500 (about €430) a month that promises especially fast responses.
What exactly “o” will be able to do, which permissions it will get and when it might reach Europe are all still open questions. The name itself may not stick either: internal code names often fail to survive the final branding review.
DevDay Sets the Stage
A potential launch would take place at DevDay, OpenAI’s annual developer conference at Fort Mason in San Francisco. CEO Sam Altman opens the event with a keynote that will be livestreamed. Over the weekend, the company’s developer account stoked expectations with a countdown post saying the team had been building and it was time to show its work. The teaser image, featuring colorful round characters, was quickly read by the community as a nod to “o.”
An agent that works independently in the background would fit OpenAI’s strategy. The company recently introduced continuously running Workspace Agents for business customers, as Trending Topics reported. “o” could become their consumer counterpart.
Peter Steinberger’s Possible Role
How much Peter Steinberger contributed to “o” remains unclear. The Austrian developer built OpenClaw, one of the most closely watched open-source agents, and joined OpenAI earlier this year, as Trending Topics reported. At the time, Sam Altman wrote on X that Steinberger would “drive the next generation of personal agents” at OpenAI and said he expected this work to quickly become core to the company’s products.
In his blog post about the move, Steinberger set himself a clear goal: to build an agent that even his mother can use. That is precisely the audience a mass-market agent like “o” would target. So far, however, there is no official confirmation that Steinberger worked on “o” or that he will appear at DevDay. OpenClaw itself has since moved to a foundation and continues as an independent open-source project.
Meta Already Has Muse on the Market
Most of the pressure on OpenAI comes from Menlo Park. About three weeks ago, Meta launched Muse, a personal AI agent for a mass audience, though only in the United States for now, as Trending Topics reported. According to Meta, Muse runs in its own isolated virtual machine, connects to email, calendars and other services, writes emails, books travel, fills out forms and completes purchases. It keeps working on longer tasks after the app is closed and checks back in when it needs approval.
Basic use is free, with subscriptions of $20 (about €17) and $100 (about €85) a month kicking in at higher usage. The launch has gone well: Muse climbed to the top of the free iPhone apps in the U.S. and passed 2.5 million downloads within two weeks, according to Sensor Tower figures cited by CNN. At its Connect conference, Meta followed up with Muse Charm, a keychain device built specifically for the agent. Critics, however, question whether users will trust a company like Meta with access to so much personal data.
Other Competitors
Other players are pushing into the category as well. xAI has sent Grok Bot, a digital coworker, into an early beta, and Microsoft is building Autopilot, a continuously running agent, into Copilot. Anthropic also offers agentic features in Claude that can work on a computer independently, though they are aimed more at professional users.
Google Still Lacks a Comparable Product
Despite owning Gmail, Google Calendar, Android and the world’s largest search engine, Google has yet to launch a comparable agent for a mass audience. Its agent tool Gemini Spark is still in testing and only available to paying subscribers in the U.S. At the same time, Google is testing CC, a separate agent product aimed at families. Former Google Ventures partner M.G. Siegler criticized exactly this fragmentation in his newsletter Spyglass, calling Google’s approach timid. Investors are nervous too: Alphabet shares recently lost almost five percent over two trading days, a drop analysts linked to Muse’s success. Google DeepMind head Koray Kavukcuoglu then signaled that the next model, Gemini 4, could arrive much earlier than the end of the year.
