The company that started the chatbot revolution is playing catch-up in the always-on AI agent race, and it's betting big on a persistent assistant that works while you sleep
OpenAI ’s DevDay 2026 event, scheduled for September 29 in San Francisco at Fort Mason, is widely expected to showcase a persistent AI agent reportedly codenamed “Aeon” and potentially branded as “o.” The agent would represent OpenAI’s most significant product shift since ChatGPT’s original launch, moving from a tool you talk to into a tool that acts on your behalf, continuously, even when you’re not watching.
Where current chatbots wait for your input and respond, Aeon is designed to keep executing tasks after you close the chat interface. According to reports, the agent is expected to run on a specialized variant of GPT-6 Astra, OpenAI’s latest model architecture. It may incorporate individual user profiles, dedicated email identities, and proactive updates, meaning the agent could ping you with relevant information before you even think to ask for it.
There’s also speculation that OpenAI will use DevDay to announce a premium “ChatGPT Pro Max” subscription tier alongside potential hardware features, though neither has been officially confirmed. The company has stayed characteristically quiet ahead of the event, letting leaks and industry chatter do the marketing for free.
Meta launched its own proactive personal AI agent, Muse, on September 8, more than three weeks before DevDay. Muse already handles email management, bookings, and a range of everyday tasks across a three-tier pricing structure: a free option, a $20 per month plan, and a $100 per month premium tier.
JPMorgan has ranked Muse highly in recent benchmarks among consumer AI agents.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
OpenClaw, a community-built agent framework released in late 2025, has carved out a meaningful niche by letting users run agents locally on their own hardware. Its integration with popular messaging platforms gave it organic distribution that venture-backed competitors struggle to replicate.
The shift from conversational AI to persistent agents represents a fundamentally different product category. An agent with its own email identity raises immediate questions about accountability. If Aeon sends an email on your behalf that contains an error, or worse, commits you to something you didn’t intend, the liability picture gets complicated fast.
For OpenAI specifically, the challenge is bundling enough value into a paid tier to justify the compute costs of running persistent agents at scale. Always-on agents consume far more server resources than request-response chatbots. Meta can absorb those costs across its advertising empire. OpenClaw offloads them to the user’s own hardware.
