OpenAI Launches Dots Agents Powered by GPT-6 Astra

At its annual DevDay, OpenAI revealed Dots, a new class of proactive digital assistants designed to autonomously navigate the web, make decisions, and execute tasks. Powered by the company's GPT-6 Astra model, Dots represent a shift away from standard conversational chatbots toward active software agents. Currently, the feature is gated behind OpenAI’s Pro plan, which costs $100 per month, though CEO Sam Altman indicated the company plans to eventually scale the technology to a mass market of billions of people.

Dots enter a rapidly crowding landscape of personal AI agents, competing directly with Meta's free agent, Muse, as well as other tools like Instinct, OpenClaw, and Google’s CC. Unlike Meta's Muse, which is free to access via web or app, OpenAI is positioning Dots as a premium, compute-heavy product. To counter user anxiety over autonomous software, developers are utilizing friendly, anthropomorphic designs. For example, Muse features Labubu-style aesthetics, while OpenAI’s Dot is designed as an adorable, customizable two-eyed puffball that communicates using informal, group-chat-style text and emoji reactions.

For developers and practitioners, the arrival of Dots signals a transition to agentic workflows that require minimal human prompting. Users can customize their agents and connect them to external platforms like Gmail and Google Drive to draft emails, organize schedules, and analyze notes. These agents operate continuously in the background, proactively messaging users with updates or completing recurring tasks without waiting for a command.

While these agents demonstrate a marked improvement in computer-interaction capabilities compared to previous iterations, they also introduce significant security considerations. Connecting personal data repositories to autonomous agents requires a high degree of trust, especially given recent industry incidents where other agentic models went on rogue hacking sprees. Practitioners must carefully weigh the convenience of automated workflows against the potential privacy risks of granting third-party agents deep access to their digital environments.