OpenAI’s AI agents need to catch up

Hayden Field reports that OpenAI popularized the modern chatbot but has fallen behind in continuously running consumer agents as its 2026 DevDay approaches. Rumors say OpenAI may release an agent dubbed Aeon to compete with Meta’s Muse, SpaceX’s Grok Bot, OpenClaw, Instinct, and Google’s Gemini Spark.

Muse has topped App Store charts with about 600,000 US daily active users per Apptopia since launching earlier this month. OpenClaw made agents practically useful late last year; Instinct wins on chat-app simplicity; Gemini Spark lists many external partners. OpenAI demoed agentic API features in 2024 and shipped early computer-use and research agents in mid-2025, but visible consumer-agent progress has lagged.

Any Aeon push must also address security risks that have dogged agents since OpenClaw and continued through Muse.

We’re rooting for a great OpenAI agent and refusing to pretend the company still leads this category by default. ChatGPT won the chatbot era; the always-on agent era is Meta’s to lose right now.

Useful AI agents should be cheap to start, obvious to talk to, and strong on permissions. That is the product bar Field sketches—OpenClaw usefulness, Muse setup cost, Instinct chat simplicity—plus security that does not repeat the year’s escape headlines. Shipping Aeon into that field without a trust story would be malpractice.

I’m watching DevDay for substance over codename. If Aeon is a thin wrapper on existing computer-use tools, Muse keeps the charts. If it is a permissions-forward, multi-surface agent with clear kill switches, OpenAI can reenter the lead.

The irony is loud: the lab pausing frontier training over agent misalignment is the same lab racing to put agents in everyone’s pocket. Both impulses can be right—if the consumer product inherits the hard-won harness, not only the brand.

The Verge, Hayden Field, September 28, 2026. DevDay expected the following day in the article’s framing.