Agentic AI Foundation adds four open agent projects
On 22 October 2026, the Linux Foundation announced that the Agentic AI Foundation (AAIF) is adding four open projects from four different contributors: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba. The announcement was made at AGNTCon + MCPCon North America in San Jose. AAIF says its portfolio has grown from three projects at launch to ten in its first year.
The additions cover different layers of the agent stack. Agent Skills is a format for packaging instructions, scripts and reference material that an agent can load when it needs specialised knowledge. Open Responses defines a shared schema, streaming events and tooling so an application can communicate with different model providers without rebuilding every integration. OpenEnv provides a framework for creating and sharing reinforcement-learning environments, while OpenSandbox supplies APIs, SDKs and Docker or Kubernetes runtimes for running and evaluating AI applications in an isolated environment.
The projects are not new commercial products that automatically work together. They are contributions to a foundation that presents itself as a vendor-neutral home for open standards, protocols and infrastructure. The existing AAIF portfolio includes the Model Context Protocol (MCP), the Agent2Agent protocol, AGENTS.md and other projects. Engineers from the contributing organisations will continue to maintain their projects, but the foundation’s governance is intended to give the wider community a role in their direction.
That matters because AI agents increasingly depend on more than a model: they need reusable procedures, model interfaces, training environments and safe execution. Shared building blocks can reduce duplicate integration work and make it easier for a team to compare providers or replace one component. The practical benefits will depend on implementation quality, security guidance and adoption outside the founding organisations. A standard can be open while the tools, permissions and data around it remain difficult to combine.
For developers and companies, the announcement is therefore a signal about where the agent ecosystem may be heading: towards a stack of interoperable components instead of one supplier’s complete platform. It is not proof that vendor lock-in has disappeared. Teams still need to inspect licenses, maintenance, compatibility and security before building production workflows on any of the four projects.