OpenAI Updates Agents SDK With New Sandbox and Harness Features for Enterprise AI Development
Agentic AI has cemented itself as the fastest-growing new trend in the modern tech industry, with leading AI developers like OpenAI and Anthropic rushing to equip enterprise clients with the tools required to build custom automated AI assistant agents. To advance this mission, OpenAI has released a major update to its Agents Software Development Kit (SDK), adding a suite of new features designed to help businesses build their own custom agents that run on top of OpenAI’s foundational models.
One of the flagship new capabilities added to the toolkit is sandboxing functionality, which allows agents to operate within fully controlled, isolated computing environments. This addition addresses a key industry risk: running AI agents in completely unsupervised settings carries significant safety concerns, due to the occasionally unpredictable behavior of autonomous agent tools.
With sandbox integration, agents work within a segregated, dedicated workspace. They can only access specific files and code for their assigned tasks, while the broader integrity of an organization’s core system remains fully protected.
Alongside sandboxing, the updated SDK also provides developers with an in-distribution harness built specifically for frontier AI models, which lets agents work with files and pre-approved tools within a secured workspace, OpenAI confirmed. For context, in AI agent development, the term “harness” refers to all supporting components of an agent outside of the core base model it runs on. An in-distribution harness typically enables organizations to both deploy and test agents built on frontier models, which are classified as the most advanced, general-purpose AI models available today.
“At its core, this launch is about taking our existing Agents SDK and making it compatible with every major sandbox provider teams already use,” Karan Sharma, a member of OpenAI’s product team, told TechCrunch.
Sharma explained that this compatibility, paired with the new harness capabilities, will allow users “to go build these long-horizon agents using our harness and with whatever infrastructure they already have in place.” Long-horizon tasks are broadly defined as complex, multi-step work that requires agents to operate over extended timelines.
Meet Your Next Investor or Portfolio Startup at Disrupt
Your next funding round. Your next key hire. Your next breakout opportunity. Find it all at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, high-impact introductions, and market-defining innovation. Register now to save up to $410.
Meet Your Next Investor or Portfolio Startup at Disrupt
Your next funding round. Your next key hire. Your next breakout opportunity. Find it all at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, high-impact introductions, and market-defining innovation. Register now to save up to $410.
OpenAI plans to continue expanding the Agents SDK gradually over time. For this initial release, the new harness and sandbox capabilities are launching first for Python, with TypeScript support scheduled for a later rollout. The company is also actively working to add more popular agent capabilities, including code mode and subagent tools, to both the Python and TypeScript versions of the SDK.
All OpenAI customers can access the new Agents SDK features via the company’s API, and all usage is covered under OpenAI’s standard existing pricing.
OpenAI Updates Agents SDK With New Sandbox and Harness Features for Enterprise AI Development
