OpenAI has launched ChatGPT Agent, a major upgrade to its flagship AI product that enables users to delegate multi-step digital tasks to an AI system. The feature goes beyond text generation to perform actions across websites, terminals, APIs, and apps—marking a shift towards more autonomous and hands-on AI capabilities in the workplace.
Designed for business and professional use, ChatGPT Agent can browse the web, log into secure accounts (with user permission), fill in forms, run code, analyse spreadsheets, and carry out research—all while adapting to the context of the task. It offers clear potential for streamlining workflows, automating routine processes, and enhancing productivity by reducing the manual effort required for digital tasks.
The system operates transparently, asking for approval before carrying out sensitive actions and allowing users to monitor, pause, or take over tasks at any point. It’s part of a broader move by OpenAI to develop AI tools that don’t just assist with information, but act on it.
Here’s what you need to know.
What is the ChatGPT Agent?
ChatGPT Agent enables the AI to act rather than just respond. It can navigate websites, log into applications, run code, interact with terminals, use APIs, and manipulate files like spreadsheets and presentations—adapting to multi-step workflows along the way.
How does it work?
Users remain in control at all times, with the option to pause or view the agent’s activity as it happens.
The agent combines browser control, terminal access, and app integrations (like Gmail and GitHub) to complete tasks. It doesn’t act autonomously without oversight—it requires user approval before accessing accounts or performing high-impact actions.
Is it reliable?
In industry benchmarks, the agent has delivered strong results. It scored over 40 percent on the ‘Humanity’s Last Exam’, designed to test complex reasoning and execution, and posted high marks in tests involving coding, web navigation, and spreadsheet manipulation.
What about safety?
- OpenAI has built in safeguards to mitigate risk:
- Explicit user consent before executing key actions
- Isolated sessions for secure logins
- Resistance to prompt injection attacks
- Refusal to engage with sensitive tasks in areas like biology or chemistry
Who can use it and when?
The agent is rolling out now for ChatGPT Pro users, with Plus and Team tier users gaining access over the coming days. Enterprise and education rollouts will follow, with EEA and Swiss users seeing a delay due to regulatory requirements.
Pro subscribers get 400 message credits per month for using the agent. Other paid tiers start with 40 credits, with the option to purchase more.
Why does it matter for business?
The launch of ChatGPT Agent signals a shift from AI as a passive assistant to an active digital operator. By executing tasks across systems—securely and under human supervision—it introduces new possibilities for automation, especially in areas like research, admin, data processing, and customer operations.
As AI takes on more of the digital workload, the focus for professionals increasingly becomes about supervision, strategy, and decision-making—rather than execution.
Discussion about this post