TLDR: OpenAI has unveiled ChatGPT Agent, an advanced AI tool designed to autonomously perform complex, multi-step tasks like online shopping, planning, and document creation. While showcasing impressive capabilities in navigating digital environments, early user experiments and demos have highlighted a significant limitation in its processing speed, with tasks sometimes taking 15 to 60 minutes to complete, underscoring the current challenges in deploying AI agents for real-time, complex operations.
OpenAI has taken a significant leap into the realm of ‘agentic’ artificial intelligence with the official announcement of ChatGPT Agent on July 18, 2025. This new tool transcends the traditional chatbot model, empowering AI to autonomously execute end-to-end assignments within a virtual computer environment, without requiring continuous user prompts. The ambition is to allow users to delegate complex tasks, from scanning calendars and summarizing meetings to planning events, online shopping, and generating presentations or spreadsheets.
The ChatGPT Agent is powered by a new, purpose-built model that integrates the browsing capabilities of OpenAI’s earlier ‘Operator’ tool and the analytical prowess of ‘Deep Research’. This fusion enables the agent to interact with websites, manage files, and operate applications, effectively bridging the gap between conversational AI and actionable execution. Product lead Yash Kumar and research lead Isa Fulford demonstrated the agent’s ability to plan a night out by checking Google Calendar and searching OpenTable for restaurant options, and even creating research reports comparing toy trends.
However, early feedback and demonstrations have quickly brought to light a key limitation: speed. While the agent exhibits impressive autonomy, its performance speed leaves something to be desired. Reports indicate that tasks can take anywhere from 15 to 30 minutes to complete. In one notable test highlighted by Wired, the agent took over an hour to order cupcakes for Isa Fulford. Despite this sluggishness, Fulford noted that the time saved was still valuable, as the user was spared the manual effort. Kumar emphasized that the development team is prioritizing accuracy and the ability to handle ‘harder jobs’ over real-time speed, suggesting that users are not expected to monitor the agent’s progress continuously.
OpenAI has also implemented robust safety measures for ChatGPT Agent. The system is designed to pause and seek explicit user confirmation before executing any irreversible actions, such as making purchases or sending emails. Financial transactions are currently off-limits, and a ‘Watch Mode’ is in place to ensure user oversight during sensitive tasks involving financial websites; if a user navigates away from the agent’s tab, the operation automatically halts. These safeguards aim to mitigate risks associated with an AI system capable of controlling a user’s computer.
Also Read:
- OpenAI’s New ChatGPT Agent Poses Significant Bioweapon and Financial Misuse Risks
- China’s AI Agent Sector Trails US in Commercialization, Accelerates Efforts Amid OpenAI’s Advancements
The introduction of ChatGPT Agent arrives amidst growing industry interest in autonomous AI systems, with companies like Klarna having previously deployed AI agents for customer service. While the technology is still experimental, OpenAI positions it as a significant step towards enhancing productivity and streamlining workflows, challenging users to consider the extent of control they are willing to delegate to artificial intelligence.
