OpenAI just released a major update to the Agents SDK that gives developers better tools to build production ready agents. The focus is on making agents more capable at handling real work with files and systems while keeping everything secure.
What the new Agents SDK brings to developers 🤖
Agents now get native sandbox execution so they can work in isolated workspaces with the files tools and dependencies they need. This update addresses common pain points when moving from prototypes to production. As reported in TechCrunch coverage of the enterprise impact the enhancements help enterprises build safer agents.
The SDK separates the harness from the compute layer which improves security durability and scale. Developers can use built in support for providers like Blaxel Cloudflare Daytona E2B Modal Runloop and Vercel or bring their own.
Sandbox execution and long horizon tasks 💻
Developers can define a Manifest to set up the agent workspace with local files directories or cloud storage from AWS S3 Google Cloud and others. This setup makes it easier for agents to inspect evidence run commands and edit code across extended tasks. More details appear in The New Stack analysis of sandbox features.
The harness now includes configurable memory Codex like filesystem tools and integrations with standards like MCP. OpenAI designed it to align closely with how frontier models work best. Check out related best AI tools for programming that complement these capabilities.
One early tester from Oscar Health highlighted how the update made a clinical records workflow reliable enough for production. You can explore agent monitoring tools such as OpenHands coding agents or similar platforms listed in best AI tools for developers on BestAITools.com.
Further technical breakdown is available in this Junia AI blog post on the release.
The features use standard API pricing and are generally available now.
