OpenAI Expands Agents SDK to Boost Enterprise AI Capabilities

New sandbox and harness features aim to help businesses build safer and more powerful AI agents.

Apr. 15, 2026 at 7:28pm by

Got story updates? Submit your updates here. ›

A highly detailed, glowing 3D illustration of an AI agent represented as a series of illuminated geometric shapes and lines, surrounded by a network of pulsing neon-colored cables and circuits, conveying a sense of advanced, cybernetic technology powering the agent's capabilities.An AI agent operating within a secure, high-tech digital infrastructure, symbolizing the advanced capabilities and safety features powering the next generation of enterprise automation.San Francisco Today

OpenAI has updated its Agents software development toolkit (SDK) with new capabilities designed to help enterprises create more capable and secure AI agents. The key additions include a sandboxing feature that isolates agents in controlled environments, as well as an in-distribution harness that allows agents to work with approved files and tools within a workspace. These features are intended to make it easier for businesses to deploy and test advanced AI agents built on OpenAI's frontier models.

Why it matters

As agentic AI continues to grow in popularity, companies are racing to provide the tools enterprises need to build their own automated assistants and AI-powered applications. OpenAI's updates to its Agents SDK aim to address key challenges around the safety and reliability of these AI agents, which can sometimes behave in unpredictable ways when deployed in real-world environments.

The details

The new sandbox integration allows agents to operate in isolated computer environments, restricting their access to files and code to specific approved operations. This helps protect the overall system integrity when running these potentially unpredictable AI models. Additionally, the updated SDK includes an in-distribution harness that gives developers more control over how agents interact with files and tools within a workspace. Together, these features are intended to enable enterprises to build 'long-horizon' AI agents capable of more complex, multi-step tasks.

  • The updated Agents SDK was launched on April 15, 2026.

The players

OpenAI

An artificial intelligence research company that develops advanced language models and other AI technologies.

Karan Sharma

A member of OpenAI's product team who discussed the Agents SDK updates.

Got photos? Submit your photos here. ›

What they’re saying

“This launch, at its core, is about taking our existing agents SDK and making it so it's compatible with all of these sandbox providers.”

— Karan Sharma, Product team member

What’s next

OpenAI plans to continue expanding the Agents SDK over time, with future updates including support for TypeScript and additional agent capabilities like code mode and subagents.

The takeaway

OpenAI's updates to its Agents SDK highlight the growing importance of providing enterprises with the tools to build safe and reliable AI agents. By introducing sandboxing and harness features, the company aims to help businesses deploy more capable and secure automated assistants that can handle complex, multi-step tasks.