OpenAI is preparing to show, at Tuesday's DevDay in San Francisco, an always-on ChatGPT assistant internally labeled "o" — a product that would keep working after the chat window closes, and that may carry its own email identity.
The assistant briefly appeared in ChatGPT Pro upgrade copy as "o, your always-on assistant," listed alongside extra Work and Codex usage, expanded memory, and 100 gigabytes of storage. Code references in the product include a display name of "o" and an email suffix of "-o," which would let the system send and receive mail without the user sitting in a session. OpenAI has not confirmed the feature. The keynote, with Sam Altman, is scheduled for 10 a.m. Pacific at Fort Mason and will be livestreamed.
The commercial fight is not about a smarter chatbot. It is about who occupies the background of a workday. Meta has already put an agent on a keychain with Muse. Microsoft is pushing Copilot as a persistent layer inside Office. An assistant that can handle email on its own is a different category: it is a worker with an inbox, not a box the user has to open. That is why the email identity, if it ships, is the load-bearing detail.
The second-order risk is permission, not capability. An agent that can write mail while the user is offline needs a reconciliation layer between what it sent and what the user authorized. OpenAI's recent agent incidents — public image posts, database probes, a sandbox that leaked through DNS — make that layer the actual product question for enterprises watching DevDay. If "o" launches without auditable background writes, the always-on assistant will inherit the same trust gap the lab has not closed in training.
Image source: i.ibb.co