OpenAI Introduces Innovative Tools for Businesses to Create AI Agents
OpenAI Unveils New Tools for Developing AI Agents
On Tuesday, OpenAI introduced a set of innovative tools aimed at helping developers and businesses create automated systems known as AI agents. These agents are designed to perform tasks independently using OpenAI’s advanced models and frameworks.
What is the Responses API?
The new tools are part of OpenAI’s Responses API, which allows organizations to build customized AI agents capable of tasks like conducting web searches, sifting through company documents, and navigating websites. This API is set to take over from the previous Assistants API, which OpenAI plans to retire by the first half of 2026.
Growing Interest in AI Agents
Interest in AI agents has surged in recent years, yet the tech community often struggles with how to define them or demonstrate their capabilities effectively. This disconnect was evident when the Chinese startup Butterfly Effect launched Manus, an AI agent platform that quickly came under scrutiny for falling short of its promises.
Olivier Godement, the head of OpenAI’s API product, remarked on this challenge, stating that while showcasing an agent might be straightforward, scaling usage and ensuring frequent engagement is significantly more difficult.
Previous Offerings from OpenAI
Earlier this year, OpenAI launched two AI agents within ChatGPT: Operator, which helps users navigate the web, and Deep Research, which assists with compiling research reports. These tools provided an early look at the potential of agent technology, although they had limitations when it came to autonomy.
Features of the Responses API
The Responses API is designed to enable developers to create AI agents similar to Operator and Deep Research but with enhanced autonomy. Through this API, developers gain access to powerful AI models, including GPT-4o search and GPT-4o mini search, that can perform web searches and cite sources accurately when answering queries.
OpenAI highlights that these models demonstrate high factual accuracy, achieving scores of 90% and 88% on its SimpleQA benchmark, respectively. In comparison, OpenAI’s larger model, GPT-4.5, scores significantly lower at 63%.
Additional Functionalities for Enterprises
The Responses API also features a file search tool that quickly scans a company’s database files for information retrieval. OpenAI assures users that it will not use these files to train its models. Additionally, the API allows access to the Computer-Using Agent (CUA) model, which powers the Operator tool, automating tasks like data entry and app workflows.
Companies can opt to run the CUA model locally, which is currently available in a research preview. The consumer version of this model is limited to action on the internet.
Limitations of Current AI Agents
While the Responses API presents advancements, it does not address all existing issues related to AI agents. Despite enhanced accuracy, AI-powered search tools still face challenges, particularly with web search queries and the persistence of "hallucinations," where AI generates incorrect or misleading information. For instance, GPT-4o search still misidentifies 10% of factual questions.
OpenAI has acknowledged that the CUA model is not fully reliable for automating tasks within operating systems and may make mistakes.
Future Improvements and Open-Source Tools
OpenAI describes its agent tools as early iterations, with ongoing enhancements in the works. Alongside the Responses API, the company is releasing an open-source toolkit called the Agents SDK. This toolkit will provide developers with free resources to integrate AI models with internal systems, implement safeguards, and track AI agent performance.
The Agents SDK is seen as an evolution of OpenAI’s Swarm framework, aimed at multi-agent orchestration introduced last year. Godement expresses hope that OpenAI can transition from showcasing flashy agent demos to meaningful products in the near future.
Sam Altman, OpenAI’s CEO, has previously indicated that 2025 could be significant for AI agents in the workforce, emphasizing the goal of making AI agents an integral part of various industries.
With these releases, OpenAI aims to shift from merely demonstrating capabilities to providing impactful tools that enhance productivity through automation.
