OpenAI has upgraded its Agents SDK with new capabilities aimed at accelerating enterprise adoption of agentic AI systems. The update introduces sandboxing functionality, allowing AI agents to operate within controlled environments, reducing risks associated with unsupervised execution while maintaining access to specific files and tools.
The release also includes an in-distribution harness, enabling developers to deploy and test agents more effectively alongside frontier models. Karan Sharma indicated the enhancements are designed to support more complex, long-horizon tasks and improve compatibility with different infrastructure environments.
Initially launching in Python with TypeScript support planned, the updated SDK is available through OpenAI’s API under standard pricing. The company signaled continued investment in expanding agent capabilities as demand for automated AI workflows grows.
