OpenAI unveiled ChatGPT Agent this week, a new upgrade that turns the popular AI chatbot into a proactive task-master. This new feature allows ChatGPT to operate on its own virtual computer, handling complex jobs from research to execution without constant user input.
This is designed to bridge the gap between thinking and doing, making it a game-changer for productivity in everyday workflows. With the ability to technically execute complex tasks on your behalf, the reality is you’ll need to give it credentials to leverage your accounts, many of which will have credit cards or payment methods connected.
Built on the strengths of previous tools like Operator for browser actions and deep research for information synthesis, ChatGPT Agent combines conversational AI with real-world capabilities. Users can now instruct it to perform tasks such as planning meals, analysing competitors, or creating presentations, all while maintaining context across steps.
Currently the browser session will stop at the checkout, so it can find the right items at an online store, add them to the cart, but will stop short and request permission before the final step of placing the order and spending the money.
Initially this will give users comfort, but I predict, in a fairly short period of time, we’ll grow comfortable with the Agent spending money on our behalf and simply set spending limits for the Agent, but allow them to complete the process and have the good arrive in your inbox (digital) or at your door (physical).
You could set a limit to the amount the agent is permitted to spend at Amazon for example, then as you request it to order, the experience is seamless, as long as what you order is less than say $500 for the week, month, or year.
Key Features
Visual Browser
This tool lets the agent interact with websites through a graphical interface, clicking links, filling forms, and even handling logins with user confirmation for security.
Text-Based Browser
For quicker, reasoning-focused queries, the agent uses a streamlined text browser to fetch and filter web data efficiently.
Terminal Access
ChatGPT Agent can run commands in a virtual terminal, enabling code execution, file management, and other technical tasks directly.
Direct API Integration
It connects to APIs and ChatGPT connectors like Gmail or GitHub to pull in personal data and automate actions across platforms.
Dynamic Tool Selection
The agent intelligently chooses the best tools for the job, switching seamlessly while preserving full task history for accurate results.
Collaborative Workflow
Users can interrupt, provide feedback, or steer the agent mid-task, with options for notifications on completion via the mobile app.
Editable Outputs
Generates spreadsheets, slides, and other files that users can edit, with beta support for presentation creation.
Safety remains a top priority, with built-in mitigations against risks like prompt injection and real-world actions requiring explicit user approval. OpenAI has classified this as high capability in biological and chemical domains under their Preparedness Framework, implementing safeguards like refusal training and monitoring.
For Australian users, ChatGPT Agent is rolling out now to paid subscribers, with full access expected shortly outside restricted regions like the EEA. Pro users get 400 messages per month, while Plus and Team tiers offer 40, with options to buy extra credits. Pricing starts at A$31 per month for Plus (approximately, based on current exchange rates), A$308 for Pro, and A$38 per user for Team on annual plans.
This launch highlights OpenAI’s push towards more agentic AI, where tools don’t just respond but actively solve problems. While it’s exciting for professionals and creators, the emphasis on safety ensures responsible use. As AI continues to advance, features like this could redefine efficiency in tech-savvy Australia.
For more information, head to https://openai.com/index/introducing-chatgpt-agent/
