New sandboxing and advanced testing capabilities aim to help businesses safely build and deploy AI agents, as competition intensifies among tech firms to deliver scalable, enterprise-ready automation solutions.
As interest in agentic AI continues to grow, OpenAI has introduced significant upgrades to its Agents Software Development Kit (SDK), aimed at enabling enterprises to build more capable and secure AI-driven automation tools.
The updated SDK focuses on improving how businesses develop AI agents—software systems designed to autonomously perform tasks—by adding new features that enhance both functionality and safety. These enhancements come at a time when companies are increasingly exploring agent-based systems to handle complex, multi-step workflows.
A key addition is the introduction of sandboxing capabilities, which allow AI agents to operate within controlled environments. This approach limits access to specific files, code, and tools required for a given task, helping organisations reduce risks associated with unsupervised or unpredictable AI behaviour.
Focus on safety and scalable deployment
The sandbox feature ensures that agents function within isolated workspaces, preventing unintended interactions with broader systems. This is particularly important as enterprises begin deploying AI for more critical operations, where errors or misuse could have significant consequences.
In addition to sandboxing, OpenAI has introduced a new “in-distribution harness” designed for use with its most advanced models. This framework enables developers to test and deploy agents more effectively by providing structured access to approved tools and datasets within a defined environment.
Together, these capabilities are expected to support the development of “long-horizon” agents—systems capable of executing complex, multi-step tasks over extended periods. Such agents are increasingly seen as the next step in enterprise automation, moving beyond simple, single-action AI tools.
Broader rollout and future expansion
The enhanced SDK is initially being released with support for Python, with TypeScript compatibility planned in future updates. OpenAI also indicated that additional features, including advanced coding modes and multi-agent systems, are in development to further expand the platform’s capabilities.
The new tools are being made available through OpenAI’s API under its standard pricing model, allowing businesses of varying sizes to experiment with and deploy agent-based solutions.
With competitors like Anthropic also advancing in the agentic AI space, the latest update reflects a broader industry push to deliver enterprise-grade AI systems that balance innovation with operational safety.
See What’s Next in Tech With the Fast Forward Newsletter
Tweets From @varindiamag
Nothing to see here - yet
When they Tweet, their Tweets will show up here.
