OpenAI releases new business tools for building AI agents

OpenAI released a new set of tools that will help businesses create specialized AI Agents. The tools will make it easy for businesses to create automated systems that can complete specific tasks using OpenAI existing LLMs and frameworks.

The new set of AI tools is similar to OpenAI’s Operator product. They are part of the company’s new Responses API service. The AI Agents will perform several tasks like internet search, website navigation and scanning through company files. The new AI Agents and tools are essentially a replacement for OpenAI’s Assistants API, which will discontinued in late 2026.

Responses API helps developers build agentic applications

Responses API has components that power AI agents and help developers build applications that utilize OpenAI’s Operator and Deep Research. Creating such agentic applications could lead to solutions that are more autonomous than today’s AI apps.

The Responses API will let developers access same AI models that operate within ChatGPT Search which include GPT-4o search and GPT-4o mini search. These models can browse the internet to answer questions while citing sources as replies being written.

OpenAI’s SimpleQA benchmark results showed that GPT-4o search and GPT-4o mini have a high degree of factual accuracy. When it comes to answering short and factual questions, GPT-4o Search scored 90 percent, followed by GPT-4o Mini Search at 88 percent. On the other hand, GPT-4.5 scored 63% only, even though it’s a larger model.

The Responses API comes with a file search functionality that can be used to quickly search through the files in an organization’s databases to retrieve information. OpenAI claimed that it won’t train its LLM models using such files or databases. Developers can use the model to automate tasks such as data entry and application workflows by generating mouse and keyboard actions.

Moreover, OpenAI’s Computer-Using Agent (CUA) model, which powers Operator, is available to developers using the Responses API. OpenAI stated that CUA is available in research preview mode only, and businesses have the option to run CAU models on the local servers.

According to a report by The Wall Street Journal, Stripe, the largest private fintech company, used OpenAI’s generative technology to build a prototype of an AI agent. This agent is capable of reading a sales spreadsheet, creating invoices, and sending them to customers using Stripe’s AI platform.

Aaron Levie, CEO of Box, stated that the cloud-storage company is simplifying the process of developing custom agents using OpenAI’s platform for its business clients. These agents seamlessly integrate with any data stored on Box.

OpenAI’s new API isn’t a leap for agentic AI

The ChatGPT creator has stated that the CUA model is not that reliable for task automation on operating systems and the model could make unintentional mistakes. The company said that these tools are early versions of their agentic tools and they’re continuously working to improve them.

Consistent with the release of the Responses API, OpenAI is also releasing an open-source toolkit known as the Agents SDK. This toolkit helps developers integrate LLM models into internal company systems, oversee agentic AI activities for optimization and debugging purposes, and implement safety measures at no cost. The Agents SDK is kind of the successor to Swarm, which is a framework that simplifies the orchestration of multiple agents through the provision of a transparent and minimalist interface.

Olivier Godemont, OpenAI’s API product head said that “agents are the most impactful application of AI that will happen.” He stated that its easy to demo an AI agent but its difficult to scale it and get people to actually use it. Godemont remains hopeful about seeing OpenAI bringing this vision into reality.

Cryptopolitan Academy: Tired of market swings? Learn how DeFi can help you build steady passive income. Register Now