OpenAI unveiled last week a suite of new tools designed to simplify the development of AI agents—systems that can independently (but not yet autonomously) accomplish tasks on behalf of users. OpenAI said in the official announcement that this release provides developers and enterprises with "the first set of building blocks" to create valuable and reliable AI agents based on the startup's technologies.

That this set is referred to as the first subtly suggests OpenAI may have more tools in preparation, in line with the current excitement surrounding AI agents. The blog post also touched on the reportedly positive adoption that AI agents are having within OpenAI's customer base, as the startup claims that the current tool release was motivated by feedback from customers trying to build agents that leverage some of OpenAI's latest model capabilities like advanced reasoning or multimodal interactions.

Key Components of the Release

The new offerings include access to:

  • Responses API: A new API primitive designed to become a flexible foundation for building agents that can leverage OpenAI's built-in tools described below. According to the startup, the Responses API is a superset of the Chat Completions API, combining its simplicity with the Assistants API's tool-use capabilities.
  • Built-in tools:
    • Web search is available as a tool whenever GPT-4o and GPT-4o-mini are in use. This is unsurprising, as the ChatGPT search consumer product is based on GPT-40. As its name indicates, web search enables agents to use the internet to deliver grounded answers and relevant citations. This tool is available as a preview to everyone using the Responses API starting at $30 and $25 per thousand queries for GPT-40 and 40-mini.
    • File search enables agents to extract information from large collections of documents (such as knowledge bases) and is ideal for setting up a retrieval-augmented generation (RAG) pipeline. File search costs $2.50 per thousand queries, while file storage is $0.10/GB/day, with the first GB free.
    • The computer use preview is powered by the same model OpenAI's Operator is built on, letting customers build agents capable of performing tasks directly on computers.
  • Agents SDK: An open-source toolkit for orchestrating single-agent and multi-agent workflows. It includes configurable LLMs, handoffs between agents, safety checks, and tracing and observability tools. This SDK works with OpenAI's API services and third-party APIs providing a Chat Completions-style endpoint.

Early adopters are already seeing success

OpenAI reports on several success stories from existing customers who have already put these tools to work. Some use cases include:

  • Hebbia uses web search to help asset managers and law practices extract insights from extensive datasets
  • Navan employs file search in its AI travel agent to provide users with precise answers from knowledge-base articles
  • Luminai integrated computer use to automate complex operational workflows for enterprises with legacy systems
  • Box has created agents that leverage web search and the Agents SDK to help enterprises search and extract insights from unstructured data

Looking Ahead

OpenAI also detailed its plans for the Chat Completions and Assistants API services. The startup says it will continue to support and enhance the Chat Completions API with new models. The company recommends Chat Completions for any application that does not require built-in tools. In contrast, the Assistants API will be deprecated once it and the Responses API achieve full feature parity. Once this is officially announced, OpenAI has vowed to provide migration paths from the Assistants to the Responses API.

OpenAI views these releases as just the beginning of its agent platform. The company has indicated plans to continue investing in deeper integrations across their APIs and develop new tools to help deploy, evaluate, and optimize agents in production. As AI models become increasingly capable of handling complex, multi-step tasks, these new tools aim to streamline the development process, making it significantly easier for developers to build practical AI agents that can deliver real-world impact across industries.