OpenAI launches new tools to help businesses build AI agents

On Tuesday, OpenAI released new tools that were designed to help developers and companies build AI agents – automated systems that can perform independent tasks – using their own AI models and frameworks of the company.

The tools are part of OpenAI’s new answers API, with which companies can develop custom AI agents who can carry out web search assignments, can scan business files and navigate websites, just like OpenAI’s product product. The API answers effectively replaces the API of OpenAi, who is planning the company to sun sons in the first half of 2026.

The hype surrounding AI agents has grown dramatically in recent years, despite the fact that the technical industry has had difficulty showing people, or even define what “AI agents” are real. In the most recent example of agenthype that ran before the utility earlier this week, the Chinese startup Butterfly effect went viral for a new AI agent platform called Manus that users quickly discovered that many of the promises of the company were not true.

In other words, the deployment is high for OpenAi to get agents well.

“It’s pretty easy to demonstrate your agent,” Olivier Godemont, OpenAi’s API product head, told WAN in an interview. “It is quite difficult to scale up an agent, and to let people use it often is very difficult.”

Earlier this year, OpenAi introduced two AI agents in Chatgpt: Operator, who navigates on your behalf, and deep research, who prepare research reports for you. Both tools offered a glimpse into some agent technology, but left quite a bit desired in the “autonomy” department.

Now with the answers API, OpenAi wants to sell access to the components that AI agents authorize, so that developers can build their own agent applications in the operator and deep research style. OpenAI hopes that developers can make some applications with his agent technology that feel autonomous than what is available today.

With the help of the answers API, developers can tap the same AI models (in preview) under the hood of OpenAI’s Chatgpt Search Web Search Tool: GPT-4O Search and GPT-4O Mini search. The models can browse the internet for answers to questions, with reference to sources when generating answers.

OpenAi claims that GPT-4O search and GPT-4O mini-search assignment are very actually accurate. On the SimpleQa benchmark of the company, which measures the ability of models to answer short, fact-seeking questions, GPT-4O search scores 90% while GPT-4O mini-seeking scores 88% (higher is better). For comparison, GPT-4.5 it scores much larger, recently released model of OpenAi-Slechts 63%.

The fact that AI-driven search aids are more accurate than traditional AI models is not necessarily surprising-in theory, GPT-4O can simply look for the correct answer. Web search, however, does not make hallucinations solved. In addition to their factual accuracy, AI search tools also tend to struggle with short, navigation queries (such as “Lakers score today”), and recent reports suggest that that Chatgpt’s quotes are not always reliable.

The answers API also contains a tool for searching for files that can quickly scan files in the databases of a company to collect information. (OpenAI claims that it will not train models on these files.) Moreover, developers who use the answers can tap API on OpenAI’s Computer -Use Agent Agent (CUA) model, which feeds operator. The model generates mouse and keyboard actions, allowing developers to automate the use of computer use, such as data entry and app workflows.

Enterprises can optionally perform the CUA model that spends locally on their own systems, said OpenAi. The consumer version of the CUA that is available in operator can only take actions on the internet.

To be clear, the answers API will not solve all the technical problems that AI agents today solve too much.

Although AI-driven search aids are more accurate than traditional AI-models-a fact that it is not surprising as they can simply look up the correct answer, web search no AI-hallucinations for a solved problem. GPT-4O search still gets 10% of the actual questions wrong. In addition to their accuracy, AI search tools also tend to struggle with short, navigation queries (such as “Lakers score today”), and recent reports suggest that Chatgpt’s quotes are not always reliable.

In a blog post to WAN, OpenAI said that the CUA model “is not yet reliable for automating tasks on operating systems”, and that it is sensitive to making “unintended” mistakes.

However, OpenAi said that these are early iterations of their agent tools, and it is constantly working to improve them.

In addition to the answers API, OpenAi releases an open-source toolkit called the Agents SDK, which offers developers free tools to integrate models with their internal systems, guarantee and check AI agent activities for error detection and optimization purposes. The Agents SDK is a kind of continuation of OpenAi’s Swarm, a framework for multi-agent orchestration that the company released at the end of last year.

Godemont said that he hopes that OpenAi can bridge the gap between AI agent demos and products this year, and that in his opinion “agents are the most impactful application of AI that will happen.” That reflects a Proclamation OpenAi CEO Sam Altman made in January: That 2025 is the year that AI agents enter the workforce.

Or 2025 really becomes the ‘Year of the AI ​​agent’, the latest releases from OpenAi show that the company wants to shift from flashy agent demos to impactful tools.