The world of AI agents just took a big step toward openness. The Agentic AI Foundation — the neutral, vendor-independent home for agent standards — has announced four new hosted projects, growing its portfolio from three projects at launch to ten in its first year.
The announcement was made at AGNTCon + MCPCon North America in San Jose, California, on October 22, 2026, the community conference that brings together the open agentic AI ecosystem to share implementation experience and advance how AI agents are built, secured and operated.
What is striking about this announcement is who contributed the projects: Anthropic, OpenAI, Hugging Face and Alibaba each donated a project to the foundation, placing their work under open, vendor-neutral governance rather than keeping it under corporate control.
AGENT SKILLS, CONTRIBUTED BY ANTHROPIC
Agent Skills is about how AI agents gain new capabilities. Instead of every developer inventing their own way to extend an agent, the project aims to create a shared standard for packaging, sharing and applying skills to agents.
OPEN RESPONSES, CONTRIBUTED BY OPENAI
Open Responses tackles how applications talk to AI models across different providers. As agents increasingly need to work with models from more than one company, a common response format becomes essential. This project sits at the layer where applications and models meet, making it easier to switch providers without rewriting everything.
OPENENV, CONTRIBUTED BY HUGGING FACE
OpenEnv, from Hugging Face, covers the environments where agents are trained and evaluated. Good evaluation is one of the hardest problems in the agent world — how do you reliably test whether an agent actually did the job? OpenEnv aims to give the community shared, reproducible environments for training and benchmarking agents.
OPENSANDBOX, CONTRIBUTED BY ALIBABA
OpenSandbox, contributed by Alibaba, addresses where agents run safely. Agents that can take real actions — browse the web, run code, call tools — need secure, isolated places to operate. A shared open standard for sandboxed agent execution would raise the safety bar for the whole industry.
TEN PROJECTS, ONE NEUTRAL ROOF
These four join the foundation’s existing projects: the Agent2Agent (A2A) protocol, Agent Router, agentgateway, AGENTS.md, goose and the Model Context Protocol (MCP). Each covers a different layer of the agentic stack — how agents gain skills, how applications talk to models, how agents are trained and evaluated, and where agents run safely.
“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 Agentic AI Foundation. “Agent Skills, Open Responses, OpenEnv, and OpenSandbox each cover a different part of that stack, from how agents gain capabilities to where they run, and four different companies chose to contribute them to neutral governance.”
That last point is the real story. Four major players — including direct competitors — chose to put core pieces of the agent stack into a neutral foundation rather than keep them proprietary. It suggests the industry is converging on a shared understanding: agentic AI will only scale if the infrastructure beneath it is open and nobody single-handedly controls it.
WHY THIS MATTERS FOR BUILDERS
For developers, neutral standards mean less lock-in. An agent built against open protocols can move between providers, plug into shared tooling, and benefit from community improvements rather than waiting on one vendor’s roadmap. The four new projects fill in some of the most practical gaps — skills, model interoperability, training environments and safe execution — which are exactly the places where teams today spend the most time reinventing the wheel.
WHAT TO WATCH NEXT
The conference runs through October 23, and the foundation’s project count may keep climbing. The next milestone to watch is adoption: standards only matter if developers actually build on them. With Anthropic, OpenAI, Hugging Face and Alibaba all backing the same neutral home, the agent ecosystem’s foundation is looking stronger than ever.
Which of these four projects — Agent Skills, Open Responses, OpenEnv or OpenSandbox — do you think developers will adopt first? Tell us in the comments.
Comments
Post a Comment