Artificial Intelligence Giant, OpenAI, Nears Initial Release of AI Agent
In a groundbreaking development, software developer Tibor Blaho has uncovered evidence of an OpenAI agent codenamed "Operator." This AI tool is designed to perform tasks by interacting with remote visual browser environments, executing code in a controlled terminal, and accessing external data sources via connectors.
The OpenAI agent "Operator" forms a key component within OpenAI’s newer ChatGPT agent system, which integrates Operator's strengths with those of Deep Research and ChatGPT's conversational capabilities.
Operator's core ability is to execute tasks through a remote visual browser environment, enabling it to navigate websites, click around, fill forms, and perform complex web-based interactions automatically. This capability, combined with Deep Research’s multi-step research and synthesis skills and the ability to run code, analyze data, and generate documents such as reports or slide decks, makes the ChatGPT agent system a powerful tool for diverse tasks.
The system operates through a "virtual computer," switching fluidly between reasoning and action to complete workflows end-to-end. Examples of tasks that can be performed include navigating user calendars, analyzing competitors, or buying ingredients online—all while requiring explicit user permission before executing consequential actions for safety reasons.
In relation to specific tasks, Operator's visual browser interactions and the ChatGPT agent's multi-step web handling suggest strong suitability for web navigation benchmarks such as OSWorld and WebVoyager. While direct test results of Operator on these benchmarks are not explicitly detailed, the ChatGPT agent system's design for general purpose task execution across the web and research tasks implies it can perform well on such tests.
Operator’s ability to interact visually with websites and run code also implies it can handle multi-step form submissions and software interactions required to create Bitcoin wallets online, under user supervision and with safety measures in place. The success rate for creating a Bitcoin wallet using Operator is reported to be 10%. Similarly, Operator could automate the process of registering on cloud provider websites by completing form entries, navigating UI steps, and managing authentication flows. Registering with a cloud provider using Operator has a success rate of 60%.
Blaho found references to the AI agent "Operator" on OpenAI's website, and the evidence was first shared on X (formerly Twitter) by Tibor Blaho on January 19, 2025. A user on X (formerly Twitter) nicknamed M1 independently confirmed similar details about the macOS version of ChatGPT. The ChatGPT macOS desktop app has hidden options to define shortcuts for the desktop launcher to "Toggle Operator" and "Force Quit Operator."
The evidence of "Operator" can be found at this link: https://t.co/rSFobi4iPN and a picture related to the evidence of "Operator" can be found at this link: pic.twitter.com/j19YSlexAS.
In sum, Operator enables proactive, autonomous web and code task execution within OpenAI’s new agentic ChatGPT system, making it a powerful tool for diverse tasks like web navigation benchmarks (e.g., OSWorld, WebVoyager), Bitcoin wallet creation, and cloud provider registration. The system emphasizes safety by allowing users to interrupt or control the agent and by requiring approval before consequential actions are executed.
The OpenAI agent "Operator" is not only a key component of the ChatGPT agent system but also demonstrates the ability to handle web navigation tasks, including creating Bitcoin wallets online and registering with cloud providers. This suggests a potential integration with decentralized finance (Defi) and blockchain technology, as Operator can run code and interact with remote visual browser environments. Furthermore, with advancements in artificial intelligence (AI) and technology, it is possible that Operator could be developed to execute more complex Defi tasks in the future.
