OpenAI launches ChatGPT agent to automate web research and complex online tasks.

OpenAI has rolled out ChatGPT agent, a powerful new feature that enables the AI to autonomously handle complex online tasks like research, form-filling, and creating slideshows, marking a significant leap toward more agentic AI capabilities. Available immediately to Pro, Plus, and Team subscribers, the tool integrates browsing, deep analysis, and action-taking into a single system, as announced directly by OpenAI. Users activate it via the "agent mode" dropdown in ChatGPT's composer, where they describe tasks such as planning holidays or submitting expenses, and the AI narrates its steps in real-time for oversight.

This unified agent draws on three core components: Operator for web navigation and interaction, Deep Research for information synthesis, and ChatGPT for conversational guidance, allowing it to click through sites, filter results, and compile data more efficiently than previous tools. According to OpenAI's release notes, the agent requests permission for consequential actions—like accessing logged-in websites—and users can interrupt or take control at any point, emphasizing user supervision especially for sensitive tasks such as emailing or financial operations. High-risk activities, including bank transfers, are outright refused to mitigate potential harms.

Access comes with usage limits to manage demand: Pro users get 400 messages per month, while Plus and Team plans offer 40, with flexible credit options for extras on Business and Enterprise tiers. Enterprise and Education users will gain entry in the coming weeks, and the feature is live in supported countries including the UK, though rollout to the European Economic Area and Switzerland is pending. OpenAI has sunsetted the standalone Operator preview, folding its browser functionality into ChatGPT agent, and launched a bug bounty program alongside enhanced safety measures detailed in a system card.

The launch underscores OpenAI's push into practical automation, affecting developers, businesses, and everyday users who rely on ChatGPT for productivity. Developers, for instance, benefit from seamless task handling without constant context-switching, while companies can leverage it for workflows like expense tracking or research aggregation. As the company notes, this is an early beta with ongoing improvements planned, including better data freshness and product coverage via the Agentic Commerce Protocol—potentially transforming how millions interact with AI daily.

Recent complementary updates highlight OpenAI's rapid iteration: ChatGPT Go, a low-cost plan with expanded features like advanced data analysis, has launched in Brazil and 71 more countries, reaching 89 total. Pro and Plus subscribers now have auto top-up for credits used in tools like Codex and Sora, alongside rebalanced usage to support steadier sessions. These moves come amid quirky model behaviors, such as ChatGPT's recent fixation on goblins and mythical creatures that prompted intervention, as reported by The Wall Street Journal via Slashdot, and lighter features like AI-generated pets for the Codex coding app covered by Engadget.

For users, the implications are broad: agent mode empowers proactive AI assistance but demands vigilance, given the higher risk profile from web interactions and data handling. OpenAI stresses robust controls, yet real-world testing via the bug bounty will be key. What happens next includes wider access, feature expansions, and refinements based on feedback, positioning ChatGPT as a more versatile backend for tasks from coding aids to open-source integrations, even as competitors like Anthropic impose restrictions on similar uses. This evolution matters for industries racing to embed AI agents, potentially reshaping remote work, research, and automation across the board.