Will it run?
Agents

OpenAI launches Dots, always-on agents aimed straight at Meta's Muse

By Desmond Okafor Clawpit staff
OpenAI launches Dots, always-on agents aimed straight at Meta's Muse

OpenAI pulled back the curtain on Dots at DevDay 2026 in San Francisco this week. A new family of AI agents built to run continuously in the background, executing multi-step tasks on your behalf. They're powered by GPT-6 Astra, the company's latest model, and show up as cute, customizable "blobs."

Unlike a regular chatbot that only reacts when prompted, Dots continuously scan the web, pull context from connected apps, and learn your preferences over time.

In one demo, a user's Dot spotted a calendar conflict — working through dinner — and pushed two GrubHub options with prices. The user just picked a dish and confirmed the delivery window. In another, a user collaborated with their agent to spin up a website from scratch.

Access lives in ChatGPT, Slack, and Microsoft Teams, with context preserved across channels. Pro subscribers can join a waitlist to talk to their Dot over iMessage or RCS on Android, too.

The rollout starts today for ChatGPT Pro subscribers at $100 a month. OpenAI's playbook: ship to the pricey tier first, then widen access gradually. For now, each user gets a single Dot. Multi-agent support is on the roadmap, but there's no target date.

Guardrails are baked in. Before sensitive actions — installing software, changing passwords, anything in that league — the agent asks for explicit approval. A Custom Rules tool lets you draw hard lines your Dot won't cross and flag tasks that need direct permission every time. It's a conservative stance for an "always-on" model, meant to keep an autonomous agent from making irreversible changes without you knowing.

The competitive backdrop: personal agents are having a moment. The year kicked off with early efforts like OpenClaw gaining traction in Silicon Valley. But Meta's Muse launch reached a mass audience and shot to the top of the smartphone download charts. Dots lands in that exact slot — OpenAI's direct answer to the product that turned personalized agents mainstream.

No benchmarks against Muse or other baselines yet. The story's still developing.