OpenAI has unveiled new tools to help developers and businesses build AI agents, which are automated systems that can independently perform tasks. These tools are part of OpenAI's new Responses API, allowing enterprises to create custom AI agents that can search the web, navigate websites, and scan company files, similar to OpenAI's existing Operator product. The company plans to phase out its older Assistants API by 2026, replacing it with the new capabilities.

The Responses API provides developers with access to powerful AI models, such as GPT-4o search and GPT-4o mini search, which are designed for high factual accuracy. OpenAI claims these models can offer more reliable answers than previous versions, with GPT-4o search achieving a 90% accuracy rate. Additionally, the platform includes a file search feature to help companies quickly retrieve information from their databases. The CUA model, which automates tasks like data entry, is also available, allowing developers to automate workflows with more precision.

Despite its promise, OpenAI acknowledges that there are still challenges to address, such as AI hallucinations and occasional errors in task automation. However, the company continues to improve its models, and the introduction of the Agents SDK gives developers the tools they need to build, debug, and optimise AI agents. OpenAI's goal is to move beyond demos and create impactful tools that will shape the future of AI in enterprise applications.