OpenAI has introduced new capabilities to its Agents SDK to help enterprises create more secure AI agents.
The ChatGPT maker on April 15 launched a model-native harness that lets agents work across files and tools on a computer, as well as a new native sandbox for execution. The Agents SDK harness provides configurable memory and sandbox-aware orchestration (a process for managing AI agents within secure environments).
OpenAI’s update to the Agents SDK comes more than a year after the vendor positioned the agentic open source framework as a way for enterprise developers to build AI agents. The SDK was touted as the evolution of OpenAI’s “Swarm” project, which also enabled enterprise developers to build multi-agent systems. The update to Agents SDK shows how OpenAI wants to continue pointing enterprises toward its environment and ecosystem.
A Tool for OpenAI Users
“This update just makes it easier to use OpenAI native tooling in their agents,” said William McKeon-White, an analyst at Forrester. He added that with the update, enterprise developers no longer need to manage their agentic configurations and tooling themselves; they can instead build an OpenAI agent and use what OpenAI provides in ChatGPT.
“This is perfect for people who want to build agents in the OpenAI ecosystem,” McKeon-White said. “For larger organizations that want to remain a bit more provider agnostic, this is a less relevant update.”
However, OpenAI’s desire to funnel more enterprise developers to its ecosystem is a viable strategy for the vendor, said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget.
“OpenAI wants to become profitable,” Su said. “It wants to keep everything within its ecosystem.”
He added that the vendor also has a large customer base and wants to provide customers with effective security access and protection, which the new update aims to do.
“It’s only fair that it’s being deployed this way,” Su said.
Moreover, Agents SDK update makes agent deployment easier and less technical for enterprise developers, and it also benefits OpenAI.
“If agent deployments become a lot easier, it increases the token consumption, it increases the demand for AI applications,” Su said. “It helps OpenAI identify and invest in new services that they can enable in the future.”
One downside, though, is that enterprise developers who do not want to surrender the whole process of building AI agents to OpenAI won’t have access to the security and protection this update provides.
“Some enterprises are building their own AI agents as well,” Su said.
