Funtap - stock.adobe.com
OpenAI rides agentic wave, intros new agent-building tools
The vendor's new Responses API includes various tools such as web search, file search and computer use that developers can use to build agentic applications.
OpenAI on Tuesday launched new agent-building tools for enterprises, including a new Responses API and Agents SDK.
The new tools focus on agentic AI, one of the fastest-growing segments of AI technology. They are the latest in a steady stream of product releases by the independent generative AI vendor known for the first mass market large language model, ChatGPT.
The Responses API, aimed at app developers, combines the capabilities of OpenAI's Chat Completions API with the tool use capabilities of the Assistants API for building agents that work autonomously or semi-autonomously to carry out business tasks.
The API supports new tools such as web search, file search and computer use, which lets generative AI models use computers.
In Responses API, web search enables developers to get fast, precise answers with relevant citations from the web, according to the generative AI vendor. It is available when using OpenAI's GPT-4o or GPT-4o mini.
File search lets developers extract large amounts of information from documents. It supports multiple file types, metadata files, filtering and custom reranking. For example, it can help a customer support agent access FAQs or a legal assistant reference past cases.
The computer use tool in the Responses API is powered by the same model that underlies Operator, an agent released in January that can perform tasks on the web. The model is the Computer Using Agent Model, which combines GPT-4o's vision capabilities with advanced reasoning through reinforcement learning.
The computer use tool captures mouse and keyboard actions generated by the model. Developers can use the tool to automate browser-based workflows, OpenAI said.
More agentic AI
Agentic AI is the latest generative AI trend, and the market is filled with new agentic AI tools.
In December, Google introduced Agentspace, a platform for AI agents. In October, Anthropic launched computer use, an agent that can use a computer like a human would.
Agentic AI is also flooding the open source market. Most recently, Chinese startup Manus last week introduced an AI agent it claims is the first fully autonomous agent.
The agentic trend highlights another trend, said Gartner analyst Arun Chandrasekaran.
"The value is increasingly moving to the layer above the model," Chandrasekaran said. He added that LLMs are now an underlying foundation, and vendors are trying to provide enterprises with more ways they can use models by building capabilities on top of the models.
"Can you embed that into an application? Can you build like an agent?" he said. "The value is moving up the stack."
A logical step
Specifically, for OpenAI, the new agentic tools and building blocks are part of the vendor's strategy for making money from its products geared toward enterprises, which hold the key to generating revenue, said Mark Beccue, an analyst at Enterprise Strategy Group, now part of Omdia.
"These are logical directions for them when I think about enterprise," he said.
Paul Baier, CEO of GAI Insights, said the tools are also helpful for enterprises and developers who don't want to spend time building the agentic tools themselves and want more support than open source providers are able to provide.
"These building blocks for software developers are easy and more powerful, and that will be embraced by software developers and will save them time," Baier said. "Software developers need to focus on building vertical or function-specific features and applications, not these building block components."
Some challenges
However, one issue is cost, Beccue said.
Some enterprises might not be able to afford to use OpenAI tools on their systems, he said. "The smaller the firm, the less likely they have to run that. The idea of expense is probably the biggest challenge for them," he said.
The ability to customize the SDKs from OpenAI can also be challenging for users.
"It's not just as easy to customize these because you might have to do more training, which means more computing resources," Beccue continued.
There is also the issue of accuracy, said Rowan Curran, an analyst at Forrester Research.
OpenAI's computer use model tested below certain benchmarks in real-world use cases, "which is definitely well below the accuracy level that you would want for an enterprise application," Curran said.
"Even with these tools that are provided, you still need to do quite a bit of work to make these things enterprise-ready," he added. "There are a lot of challenges around how you hook these things together and make them run with the level of efficacy that we expect from an enterprise application."
OpenAI's agentic tools also comes a day after news that the vendor will pay data center provider CoreWeave nearly $12 billion over the next five years. As part of the deal, CoreWeave will provide AI infrastructure to OpenAI and OpenAI will have a stake in the company.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
