OpenAI has launched a suite of new tools aimed at helping businesses and developers build advanced AI agents. These agents—designed to independently perform tasks—are powered by OpenAI’s latest AI models and frameworks, promising enhanced automation capabilities across industries.
A Major Leap in AI Agent Technology
The newly introduced Responses API enables enterprises to create AI-driven systems that can conduct web searches, analyze company databases, and navigate online platforms. The API replaces OpenAI’s previous Assistants API, which is set to be phased out by mid-2026.
“It’s pretty easy to demo your agent,” said Olivier Godement, OpenAI’s API product head, in an interview with TechCrunch. “To scale an agent is pretty hard, and to get people to use it often is very hard.” With this in mind, OpenAI’s latest advancements aim to bridge the gap between promising AI agent demos and real-world applications.
Expanding AI Capabilities
The Responses API leverages two new AI-powered search models—GPT-4o search and GPT-4o mini search—which can fetch real-time data and provide source citations. OpenAI claims that these models outperform previous versions in accuracy, with GPT-4o search scoring 90% on the company’s SimpleQA benchmark, compared to 63% for its predecessor, GPT-4.5.
Additionally, the API features a file search utility capable of scanning enterprise databases for quick information retrieval. OpenAI assures users that it will not use these files for training its models, addressing privacy concerns often associated with AI systems.
Another key component is OpenAI’s Computer-Using Agent (CUA) model, which enables AI agents to simulate human interactions with digital interfaces by generating mouse and keyboard actions. This model—currently in research preview—has the potential to streamline tasks such as data entry, workflow automation, and software navigation. Enterprises can also deploy the CUA model locally, ensuring greater control over AI-driven operations.
Addressing AI Agent Challenges
Despite these advancements, OpenAI acknowledges that AI agents still face limitations. While AI-powered search tools reduce errors compared to traditional models, they are not entirely free from inaccuracies. GPT-4o search still gets 10% of factual queries wrong, and issues such as unreliable citations and struggles with short, navigational queries persist.
Furthermore, OpenAI concedes that its CUA model is “not yet highly reliable for automating tasks on operating systems,” highlighting the need for continued refinement.
Empowering Developers with Open-Source Tools
To support AI agent development, OpenAI has also introduced the Agents SDK—an open-source toolkit designed to help developers integrate models into their systems, implement safeguards, and monitor AI behavior. The Agents SDK builds upon OpenAI’s Swarm framework, which was launched last year for multi-agent coordination.
Godement emphasized that AI agents represent one of the most transformative applications of artificial intelligence. OpenAI CEO Sam Altman echoed this sentiment earlier in 2025, stating that “this is the year AI agents enter the workforce.”
