The Agentic AI Foundation went from three maintained projects to ten in one round of donations. OpenAI, Anthropic, Hugging Face and Alibaba each supplied a different chunk of what a functioning agent requires.
OpenAI handed over Open Responses, a shared template for the interfaces that expose models. Vendors shape those interfaces differently, so carrying code from one to another usually costs real effort. Because it mirrors the layout OpenAI already serves its own models through, workloads can migrate to compatible endpoints and back with no rewrite, and familiarity should speed uptake.
Anthropic contributed Agent Skills, the open packaging format it released last December for bundling instructions, sample walkthroughs and automation scripts that lift an agent’s output quality. A common format lets teams exchange such packages without reorganizing files, and steadier data organization also reduces hallucination. Anthropic says 70-plus platforms now use it, among them ChatGPT, Cursor, GitHub Copilot and VS Code.
The remaining pair address sandboxes. Hugging Face donated OpenEnv, which trims the effort of changing training systems, and Alibaba donated OpenSandbox, which offers application programming interfaces for running and managing sandbox environments and linking the agents inside them to outside software.
David Soria Parra, an Anthropic engineer, credited the foundation with giving the Model Context Protocol a neutral home after his employer donated it less than a year earlier, and pledged continued support for open agent tooling.
The Linux Foundation set the body up last December. Ten projects remains a modest base for a sector this size, but agreeing on common ground matters when agents already cross vendor lines constantly.
