Agentic AI Foundation Reaches 10 Open Projects in Year One

SAN JOSE — The Agentic AI Foundation announced four new hosted projects on Wednesday, October 22, 2026, growing from three projects to ten in its first year. The contributions come from four of the largest players in AI — Anthropic, OpenAI, Hugging Face and Alibaba — and were unveiled at AGNTCon + MCPCon North America at the San Jose McEnery Convention Center.

According to the foundation, the four projects address four different layers of the agentic stack: how agents gain new skills, how applications talk to models across providers, how agents are trained and evaluated, and where agents run safely. They join the Agent2Agent (A2A) protocol, Agent Router, agentgateway, AGENTS.md, goose, and the Model Context Protocol (MCP) under the foundation's open, vendor-neutral governance.

The Four New Projects

Agent Skills, contributed by Anthropic, is an open format for giving agents new capabilities and domain expertise. Skills package instructions, scripts and resources in a reusable manner that agents can load on demand, making specialized know-how portable across agent products and platforms. The scale is already significant: more than 70 agent platforms support the format, over one million skills have been published with nearly 280 million installations, and the skills.sh registry listed a million skills in its first seven months. The specification is open source under Apache 2.0 and CC BY 4.0.

Anthropic launched Agent Skills a year ago, and the company said donating it to the foundation ensures that anyone building with skills can shape what comes next, with governance through a Technical Steering Committee of maintainers drawn from multiple organizations serving as individuals.

Open Responses, contributed by OpenAI, is an open specification for building multi-provider, interoperable LLM interfaces, 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. Extensions are identified by provider-specific names so features from different implementations do not conflict, and because the specification applies at the API layer, adopting it requires no model changes or retraining. Two implementations already exist: Databricks Model Serving exposes an Open Responses endpoint for Claude, Gemini and Databricks-hosted open models, and Vercel AI Gateway exposes an OpenResponses API for accessing models through its gateway.

OpenEnv, contributed by Hugging Face, is an open framework and community hub for creating reinforcement learning environments. A containerized service exposes a familiar agent interface, MCP, and the Gym training protocol used by training libraries. The committee behind it draws on work built and used at Hugging Face, Microsoft, Reflection, Meta-PyTorch, Fleet, Mercor, RadixArk, Modal, Prime Intellect and others. One reported example: Turing built Calendar Gym on OpenEnv and saw roughly 90% task success when tasks named explicit calendar identifiers, compared with roughly 40% when the same tasks used natural-language descriptions. Roadmap priorities include connecting agent harnesses to training systems, validating environments, and supporting task sampling, curriculum and evaluation during training.

OpenSandbox, contributed by Alibaba, 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. Its sandbox protocol defines lifecycle APIs — creating, inspecting, pausing, resuming, renewing, snapshotting and deleting sandboxes — and execution APIs for running commands and code, working with files, keeping sessions open and collecting resource measurements. An execution daemon called execd runs inside each sandbox. The project is licensed under Apache 2.0.

A Neutral Home for Agent Standards

The announcement marks a notable moment in the young agent ecosystem: rival labs handing core pieces of their infrastructure to neutral governance in the same breath. The foundation's CTO and co-founder Manik Surtani said the Agentic AI Foundation launched a year ago with a bet that agentic AI needed a neutral home to thrive, and that four different companies choosing to contribute projects to neutral governance validates that bet.

The Linux Foundation announced the Agentic AI Foundation's formation on December 9, 2025, with founding contributions of Anthropic's Model Context Protocol, Block's goose, and OpenAI's AGENTS.md. Platinum members at formation included Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. At the time, MCP had more than 10,000 published servers and AGENTS.md had been adopted by more than 60,000 open source projects.

The trajectory suggests the industry is converging on a shared, open stack for agents faster than many expected — with the interoperability, training and execution layers now standardized alongside the communication protocols.

What Comes Next

AGNTCon + MCPCon North America continues through October 23 in San Jose, where the open agentic AI community is expected to dig into implementation experience and shared challenges around building, securing and operating agents. For the new projects, the real test begins now: whether developers adopt the shared formats quickly enough to make them the default rather than one option among many.

Sources: Linux Foundation press release, Unite.AI analysis.