OpenAI Enhances Governance with New Agents SDK

OpenAI has introduced a new Agents SDK that allows enterprise governance teams to deploy automated workflows with controlled risks. Teams transitioning systems from prototype to production face architectural compromises regarding where their operations occur. Using model-agnostic frameworks provided initial flexibility but failed to fully leverage the capabilities of frontier models. Model-provider SDKs remained closer to the underlying models but often lacked sufficient visibility into the control harness.

Managed agent APIs simplified the deployment process but severely constrained where systems could run and how they accessed sensitive corporate data. To address this, OpenAI is introducing new capabilities in the Agents SDK, offering developers standardized infrastructure featuring a model-native harness and native sandbox execution.

The updated infrastructure aligns execution with the natural operating pattern of the underlying models, improving reliability when tasks require coordination across diverse systems. Oscar Health serves as an example of this efficiency, having tested the new infrastructure to automate a clinical records workflow that older approaches could not handle reliably.

The engineering team required the automated system to extract correct metadata while accurately understanding the boundaries of patient encounters within complex medical files. By automating this process, the provider could parse patient histories faster, expediting care coordination and enhancing the overall member experience.

Rachael Burns, Staff Engineer & AI Tech Lead at Oscar Health, stated, “The updated Agents SDK made it production-viable for us to automate a critical clinical records workflow that previous approaches couldn’t handle reliably enough.” For us, the difference was not just extracting the right metadata, but correctly understanding the boundaries of each encounter in long, complex records.

OpenAI optimizes AI workflows with a model-native harness. To deploy these systems, engineers must manage vector database synchronization, control hallucination risks, and optimize costly compute cycles. Without standard frameworks, internal teams often resort to building brittle custom connectors to manage these workflows.

The new model-native harness helps alleviate this friction by introducing configurable memory, sandbox-aware orchestration, and Codex-like filesystem tools. Developers can integrate standardized primitives such as tool use via MCP, custom instructions via AGENTS.md, and file edits using the apply patch tool.