OpenAI has launched a set of new tools aimed at simplifying the process of building agents that can autonomously complete tasks on behalf of users, rivalling Anthropic’s launch of its new AI agent last month.
“Building agents is harder than it needs to be”, said an OpenAI spokesperson.
To streamline this process, the ChatGPT owner released five new tools to help businesses and customers build and scale agentic AI.
These aim to help developers quickly integrate OpenAI’s powerful models with real-world tools like web search, file search, and computer use.
OpenAI has pledged that these tools will allow agents to perform complex, multi-step tasks.
New AI tools
The web search tool will allow agents to search through vast data sets, or even interact with computer systems directly.
During early testing, developers have already leveraged this function, applying it to research assistants and booking agents.
It has also been used in testing to accurately and quickly extract insights from both public and private datasets.
The tool is able to provide real time, relevant information directly from the web, with in-line citations to sources.
OpenAI has also launched a new file search tool, which allows AI agents to search across large document databases.
Travel software firm Navan has used it to enhance its AI-powered travel assistant, by providing employees with fast, accurate access to internal knowledge bases.
This both saves time and improves the quality of customer support.
The third tool that the tech giant has launched is a ‘computer use tool’, which allows agents to complete tasks traditionally requiring human interaction with a computer, like browsing on data entry.
Meanwhile, the ChatGPT maker has released an ‘Agents software development kit (SDK)’: an open source framework designed to streamline the communication of multiple agents with one another.
“These new tools provide developers with an easier path to building and deploying AI agents, enabling applications across various industries”, said the company on a blog post on Tuesday.
“By offering capabilities that were once difficult to integrate into agent systems, OpenAI is reducing barriers for businesses looking to harness the power of autonomous AI in real-world applications.”
These new tools follow the launch of two AI agents that the company launched in January.
The first, Operator, was released to independently browse the web to do things like booking flights or purchasing things online.
The next, Deep Research, has the capability of performing more complicated internet tasks.
The firm pledged that the latter accomplished in minutes tasks that take humans hours.
A ‘watershed year’ for agentic AI
OpenAI’s newly released agent-building tools are entering an increasingly competitive space, with several companies already offering AI development platforms.
Notably, Anthropic and Google’s DeepMind have both released large language models (LLMs) that can be used to power intelligent agents.
Anthropic’s new intelligent model was the first hybrid reasoning model on the market.
The firm announced that its new model will make it easier for users and developers to tackle problems that require step by step cogitation, as well as instinctive output.
It also featured a new ‘scratchpad’ used to reveal the model’s reasoning and working out.
Yet, OpenAI’s new tools claim to distinguish themselves by entering a new space by having an open-ended approach, while giving some guardrails to developers.
Unlike some competitors that offer LLMs with limited integration, OpenAI’s approach is centred around providing developers with the necessary tools to seamlessly combine AI with real world actions.
This will allow businesses to build agents that generate text whilst also interacting with external systems and searching through documents.