The Agentic AI Foundation (AAIF) announced four new hosted projects on October 22, 2026, at AGNTCon + MCPCon North America in San Jose: Agent Skills, contributed by Anthropic; Open Responses, contributed by OpenAI; OpenEnv, contributed by Hugging Face; and OpenSandbox, contributed by Alibaba. The additions bring the foundation to 10 hosted projects in its first year, up from the three it launched with in December 2025.
The four projects cover four different layers of how agents are built and run: how agents gain new skills, how applications talk to models across providers, how agents are trained and evaluated, and where agents run safely. All four will operate under the foundation’s open, vendor-neutral governance, alongside the projects AAIF already hosts.
Four Projects, Four Layers Of The Agent Stack
Agent Skills is an open format for giving agents new capabilities and domain expertise. Skills package instructions, scripts and resources in a reusable form that agents can load on demand, so specialised know-how becomes portable across agent products and platforms. Anthropic’s David Soria Parra, co-creator and lead maintainer of the Model Context Protocol (MCP), said Anthropic launched Agent Skills a year ago as an open standard and is now donating it to the AAIF so anyone building with skills can help shape what comes next.
Open Responses is an open specification for building multi-provider, interoperable interfaces for large language models, based on the OpenAI Responses API. It defines a shared schema, consistent streaming events and extensible tooling, so developers can describe requests and outputs once and run them across model providers with minimal translation. OpenAI’s Steve Coffey said the project joined the AAIF to help developers build on today’s language models through simple, unified interfaces.
OpenEnv is an open framework and community hub for creating reinforcement learning environments. It makes those environments interoperable, plug-and-play, easy to build and easy to share through a containerised service that exposes a familiar agent interface, MCP, and the Gym training protocol used by training libraries. Hugging Face’s Ben Burtenshaw said the committee behind OpenEnv — which includes contributors from Hugging Face, Microsoft, Reflection, Meta-PyTorch, Fleet, Mercor, RadixArk, Modal and Prime Intellect — wants to democratise AI with a fully open stack.
OpenSandbox is a general-purpose sandbox platform for AI applications. It offers multi-language SDKs, unified sandbox APIs and Docker and Kubernetes runtimes for use cases including coding agents, GUI agents, agent evaluation, AI code execution and reinforcement learning training, plus a sandbox protocol that defines lifecycle and execution APIs so teams can extend it with custom runtimes. Alibaba’s Xiaobin Xu said the company brought the project to the AAIF so its future is not defined by a single company.
Joining MCP, A2A And Goose Under Neutral Governance
The new projects join six existing AAIF projects: the Agent2Agent (A2A) protocol, Agent Router, agentgateway, AGENTS.md, goose and the Model Context Protocol (MCP). AAIF describes itself as the neutral home for open standards, protocols and infrastructure in agentic AI, with founding projects including MCP, goose and AGENTS.md.
“A year ago, AAIF launched with three projects and a bet that agentic AI needed a neutral home to thrive. Today we host 10,” said Manik Surtani, CTO and co-founder of the foundation. He said the four new projects each cover a different part of the stack, from how agents gain capabilities to where they run, and that four different companies chose to contribute them to neutral governance.
The announcement coincided with AGNTCon + MCPCon North America, held October 22–23 at the San Jose McEnery Convention Center, which brings the open agentic AI community together to share implementation experience and advance approaches to building, securing and operating AI agents. The pattern matches how agent infrastructure is consolidating elsewhere, from Walmart’s Agent Gateway for shopping agents to 1Password’s sign-in system for agents.
Why It Matters
The Agentic AI Foundation’s expansion matters because the contest over agent standards is shifting from competing proprietary formats to shared infrastructure governed in the open. When Anthropic donates Agent Skills and OpenAI donates a specification based on its own Responses API to the same foundation that already governs MCP, the practical effect is that skills, tool calls and model responses become portable: a developer can build once and carry that work across Claude, OpenAI models and other providers instead of rewriting for each platform. For enterprises, neutral governance lowers the risk of building on a format a single vendor could later change or withdraw. For the wider agent ecosystem — the world surveyed in our explainer on what AI agents are — ten hosted projects in a first year signals that interoperability, training environments and safe execution are now treated as common infrastructure, not differentiators.
Background
The Agentic AI Foundation launched in December 2025 with three founding projects — MCP, goose and AGENTS.md — as a neutral home for the protocols and frameworks that let agents operate across platforms. MCP, originally created at Anthropic, had already become the most widely adopted way for agents to connect to tools and data, and its move into foundation governance set the template the four new projects now follow: a company builds a piece of agent infrastructure, it gains adoption, and it is then handed to vendor-neutral governance so the whole industry can build on it.
Source: Agentic AI Foundation, PR Newswire
0 Comments