OpenAI updates its Agents SDK with native sandboxing and an in-distribution harness for frontier models to help deploy and test agents on long-horizon tasks
Agentic AI is the tech industry's newest success story, and companies like OpenAI and Anthropic are racing to give enterprises the tools they need to create these automated little helpers.
TechCrunchLucas Ropek
Context & Ripple Effects
OpenAI had already moved from models toward agent-building infrastructure with AgentKit, while Anthropic’s Managed Agents launch put a competing managed harness into public beta days earlier. The overlap makes deployment tooling—not just model capability—a central competitive surface.
Related coverage also points to a parallel push for open agent standards and to coding agents becoming recurring professional tools. That raises the value of reliable execution, testing, and containment for agents operating across longer workflows.
First-order effects
Developers using OpenAI’s Agents SDK gain native isolation and an integrated way to evaluate agents on long-horizon tasks, reducing the need to assemble those deployment and testing layers separately.
OpenAI broadens its agent platform from construction tooling toward operational controls needed to move agents into production environments.
Second-order effects
Anthropic and other agent-platform providers face pressure to match or differentiate on runtime safety, evaluation quality, and managed deployment—not solely on underlying model performance.
Enterprise buyers can more directly compare agent platforms on their ability to test and constrain multi-step workflows, increasing scrutiny of tool reliability before broader rollout.
Third-order effects
If these capabilities become table stakes, agent platforms will compete as integrated application-runtime stacks, with harnesses, sandboxes, and evaluation systems becoming durable sources of platform stickiness.
The simultaneous interest in open standards could limit lock-in at the interface layer, while leaving providers to differentiate on the operational quality of their managed agent environments.
The trend:Agentic AI is shifting from model-led demonstrations toward production infrastructure for safely evaluating, governing, and operating agents over extended workflows.
Long-running agents are the future - we're excited to partner with OpenAI as a sandboxing partner for their new Agents SDK launch! Get started: https://developers.cloudflare.com/ ...
Customize the Agents SDK to fit your stack. Use it out of the box, or adapt the models, tools, instructions, and orchestration logic your agent needs. The Agents SDK capabilities are available to all API customers. https://openai.com/...
Bring your own environment. The Agents SDK now supports sandbox execution with providers including @Cloudflare, @Vercel, @modal, @e2b, and more. Keep files, credentials, and execution state in your environment while passing approved context to the model. [image]
you can build a Python agent that accepts a coding task, executes it inside a Cloudflare Sandbox, and copies the output files to your local machine @OpenAIDevs x @CloudflareDev Check out our guide here: https://developers.cloudflare.com/ ...
Improve agent performance with a harness that keeps long-running agents on track. It manages the agent loop across tools, context, and traces. The sandbox preserves working state across pauses, retries, and resumptions. [image]
The new Agents SDK is what I wish I had in September 2025. It leans into the truth that all agents are coding agents and the thoughtfulness the team has put into “bring your own sandbox” is just fantastic. Let the SDK do the plumbing; focus your energy on your users. Let's build!
Use Vercel Sandbox with the OpenAI agents SDK as an official extension. Build agents that can run code, read files, and analyze data safely inside isolated microVMs. Control the compute and data flow from your secure cloud environment.
we just launched a big update to the Agents SDK. you can now bring your own compute/storage and connect it up to our agent loop to handle things like computer-use, skills, memory, and compaction
Build long-running agents with more control over agent execution. New capabilities in the Agents SDK: • Run agents in controlled sandboxes • Inspect and customize the open-source harness • Control when memories are created and where they're stored [image]