OpenAI’s Dots Are Always-On AI Agents—and Its Answer to Meta’s Muse

WIRED reports Dots, announced at DevDay 2026, are always-on agents powered by GPT-6 Astra that complete multistep projects, crawl the web, pull context from connected apps, and learn preferences. Demos include proactive dinner planning from calendar context via GrubHub options and collaborative website launch.

Messaging spans ChatGPT, Slack, and Teams; Pro users can waitlist for iMessage/RCS. Rollout starts for ChatGPT Pro ($100/month). Altman previewed enterprise “specialist Dots” for accounting, email marketing, and legal analysis. Sensitive actions need explicit approval; Custom Rules set boundaries. Training-on-data settings carry over from the OpenAI account.

We’re comparing product surfaces, not mascots.

Muse proved consumer appetite for personal agents. Dots answer with Astra, cloud computers, and workplace chat hooks. Always-on plus calendar-aware food orders is the charming demo; always-on plus credentials is the threat model. WIRED is right to flag adversary prompting and accidental oversharing.

Useful agents that wait for approval on passwords and offer Custom Rules are the minimum viable trust layer. Pro-first rollout is how you burn in failure modes on people who pay for the blast radius. Enterprise specialists raise the stakes again — legal and accounting dots need audit trails, not vibes.

I’m watching whether proactive research stays read-only in practice, and how fast Muse-class download energy transfers to a $100 tier. The agent wars are now about permissions, not avatars.

WIRED, September 29, 2026. Developing story note on page. Meta Muse referenced as consumer benchmark.