OpenAI Enters AI Agent Race with ChatGPT Agent
OpenAI has officially launched ChatGPT Agent, placing it squarely in the growing competition for AI agents. This launch allows their popular chatbot to perform complex, multi-step tasks on behalf of users, such as scheduling meetings, creating presentations, and utilizing tools like code terminals or web browsers.
Initial Access Limitations
Currently, this functionality is exclusively available to paid subscribers. Pro users can leverage up to 400 Agent prompts per month, while Plus and Team users are capped at 40. Broader access for enterprise and education users is expected in the coming weeks.
Growing AI Agent Competition
This move follows significant investments from major tech companies like Google, Salesforce, and HubSpot in the AI agent space over the past 12 months. It also coincides with OpenAI's recent announcement of its own web browser, following a similar move by Perplexity AI.
How ChatGPT Agent Works
According to OpenAI, ChatGPT Agent "combines the strengths of three early breakthroughs: the Operator's ability to interact with websites, in-depth research information integration, and ChatGPT's intelligence and conversational fluency."
Users can instruct ChatGPT to perform tasks like "Check my schedule and provide a brief summary of upcoming client meetings based on the latest news" or "Analyze three competitors and create a slide presentation".
The Agent operates on its own virtual computer and can switch between various tools, including a text browser, visual browser, code terminal, and third-party app connectors such as Gmail and GitHub.
OpenAI explains that this combination allows the model to "seamlessly transition from reasoning to action, handling complex workflows in their entirety".
Designed for Collaboration
ChatGPT Agent is designed for collaboration. Users can interrupt, take over the browser, or reassign tasks as they progress, and ChatGPT will adjust accordingly. "It will pick up where it left off, and update with new information, but not lose previous progress," the company wrote in a blog post.
Current Limitations
However, the tool does not yet support real-time responses. In a demonstration, ChatGPT Agent took approximately 25 minutes to generate a slide presentation based on company financial reports.
Product lead Yash Kumar stated that many tasks could be "started in the background and returned to later". Research lead Isa Fulford added:
"Even if it takes 15 minutes or half an hour, that's already a huge efficiency gain compared to the user doing it manually."
For safety reasons, some features, including memory, were disabled in the initial release. Kumar told The Verge, "It's not that we think it's unsafe, it's that we're taking extra precautions."
ChatGPT Agent Sets New Benchmarks, Raises Safety Concerns
OpenAI says that the model behind ChatGPT Agent has significantly outperformed previous versions of the technology in multiple benchmark tests.
But the release comes with heightened risks. OpenAI classified the model in its internal "preparedness framework" as having "high bio and chem capabilities," a designation applied to instances where models could "exacerbate existing pathways toward serious harm." OpenAI said, "While we have no conclusive evidence that the model could substantially help non-experts manufacture serious biohazards -- which is our bar for designating 'high capability' -- we are acting with caution and immediately implementing necessary safeguards."
The company stated it has taken new measures to reduce real-world misuse risks, including requiring users to explicitly confirm before executing irreversible actions, and a "monitor mode," which requires users to oversee high-stakes tasks (such as sending emails or interacting with financial websites).
OpenAI has also introduced protections to prevent prompt injection. Prompt injection is when malicious instructions are hidden in content that the agent might encounter online. The company wrote:
"Because ChatGPT Agent can take direct action, successful attacks could have a bigger impact and pose higher risks."
However, OpenAI emphasizes that today's release is "just the beginning." The company plans to continuously enhance and optimize the Agent's capabilities over time.
"We expect ChatGPT Agent's efficiency, depth, and versatility to continue to improve, including realizing smoother interactions," OpenAI stated.
TH
