Agentic AI Foundation Adopts Agent Skills and Open Responses

The Agentic AI Foundation has taken a major step toward becoming the neutral home of the agent era. At AGNTCon + MCPCon North America in San Jose on October 22, 2026, the foundation announced four new hosted projects, bringing its total to 10 in its first year. The additions cover four different layers of the agentic stack, and each one arrives from a different major contributor.

According to the foundation's announcement, Agent Skills comes from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba. The Agentic AI Foundation launched in December 2025 with three projects, and the new arrivals join the Agent2Agent protocol, Agent Router, agentgateway, AGENTS.md, goose, and the Model Context Protocol under open, vendor-neutral governance.

The timing matters. Enterprises are racing to deploy agents that can plan multi-step work, call tools, and act across systems. Until now, much of the plumbing underneath those agents belonged to individual vendors. The Agentic AI Foundation is betting that shared infrastructure will move faster when no single company controls it.

Four projects, four layers of the agent stack

Agent Skills, contributed by Anthropic, is an open format for packaging instructions, scripts, and resources that agents can load on demand. The idea is that specialized know-how, such as how to qualify a lead or how to write a renewal note, can be written once and reused across agent products instead of being rebuilt for every platform.

Open Responses, contributed by OpenAI, is an open specification for multi-provider model interfaces based on the Responses API. It defines a shared schema, consistent streaming events, and extensible tooling, so a request written once can run across model providers with minimal translation. For teams worried about being locked into one provider, that portability is the point.

OpenEnv, contributed by Hugging Face, is a framework and community hub for reinforcement-learning environments. It gives researchers and builders a containerized service exposing an agent interface, the Model Context Protocol, and the Gym training protocol, making it easier to train and evaluate agents in reproducible settings.

OpenSandbox, contributed by Alibaba, is a general-purpose sandbox platform with multi-language software development kits and unified application programming interfaces. It supports Docker and Kubernetes runtimes for coding agents, graphical interface agents, evaluation workloads, and reinforcement-learning training, giving agents a safe place to run code and act on systems.

Why a neutral home matters for agent builders

The Agentic AI Foundation's chief technology officer and co-founder, Manik Surtani, framed the announcement as proof that the neutral-governance model is working. In reported remarks, Surtani noted that the foundation launched with a bet that agentic AI needed a neutral home to thrive, and that four different companies choosing to contribute projects to that home shows the model taking hold.

That framing points to a practical concern for builders. When skills, interfaces, training environments, and execution sandboxes all live under separate vendor umbrellas, switching providers means rewriting the glue between them. A shared, community-governed layer lowers that cost. Standards do not make an agent good, but they do make it cheaper to leave a bad one, as one independent analysis of the announcement put it.

The pattern is already visible elsewhere in the agent economy. Recent enterprise launches show agents being embedded into the tools companies already use, from customer intelligence suites joining the Gemini Enterprise Agent Marketplace to outside agents being asked to register before touching enterprise platforms. A neutral stack underneath those integrations makes the whole market more interoperable, not just any single product.

What comes next for the agent ecosystem

The foundation's growth from three projects to 10 in under a year signals how quickly the agent stack is standardizing. A year ago, the interoperability story was mostly the Model Context Protocol, the plug agents use to reach tools. Now the neutral layer also covers skills, model interfaces, training environments, and execution sandboxes, and the contributors are the very labs competing hardest with each other.

For agent builders, the practical takeaway is straightforward. Packaging reusable skills in an open format, keeping model calls behind one request shape, and running agents in standard sandboxes are all becoming baseline practices rather than experiments. Teams that adopt those patterns early will find it easier to swap models, share know-how, and prove that their agents behave safely.

The Agentic AI Foundation's announcement also sets up the next question for the ecosystem: which parts of the stack standardize next. Identity, payments, and discovery for agents are all still fragmented, and each one could follow the same path from vendor feature to shared infrastructure. If the foundation's first year is any guide, that transition may happen faster than most teams expect.

Sources: the Agentic AI Foundation's October 22 announcement, coverage by the Linux Foundation, and analysis from Unite.AI and Civic. Related coverage on this site includes the arrival of customer-intelligence agents in Gemini Enterprise and the registration rules facing outside agents on enterprise platforms.